Compare commits

..

143 Commits

Author SHA1 Message Date
János Benjamin Antal
cb5ecfcad0 Make expand one work with MultiFrame 2022-11-15 16:40:16 +01:00
János Benjamin Antal
03a7c439db Make ScanAll work with MultiFrame 2022-11-11 18:11:43 +01:00
János Benjamin Antal
7ff5b902ba Extend comment about ScanAll 2022-11-11 08:46:21 +01:00
János Benjamin Antal
378cc6443a Make Once, Produce and DistributedScanAllAndFilter work with Multiframe 2022-11-11 08:43:42 +01:00
János Benjamin Antal
cca9e373e9 Handle non-existing vertices on shard 2022-11-10 20:56:16 +01:00
János Benjamin Antal
78b1830615 Make ScanAllByLabelProperty work using direct lookup 2022-11-10 17:34:07 +01:00
János Benjamin Antal
4b368452d4 Make planner use singlular primary keys as label property indices 2022-11-10 15:41:50 +01:00
János Benjamin Antal
3c07a5dc04 Merge pull request #639 from memgraph/T1145-MG-handle-insertions-of-vertices-that-already-exist-in-a-better-way
Return an error for insertions of vertices that already exist
2022-11-08 17:11:22 +01:00
János Benjamin Antal
32fe4d94d9 Merge branch 'project-pineapples' into T1145-MG-handle-insertions-of-vertices-that-already-exist-in-a-better-way 2022-11-08 16:35:38 +01:00
János Benjamin Antal
94b66a4e81 Merge pull request #638 from memgraph/T1148-MG-fix-expand-one-source-vertex
Fix setting the primary key of source vertex in ExpandOne
2022-11-08 16:35:30 +01:00
János Benjamin Antal
159b30ba5f Merge branch 'project-pineapples' into T1148-MG-fix-expand-one-source-vertex 2022-11-08 08:20:25 +01:00
gvolfing
b67e5b9a6c Merge branch 'project-pineapples' into T1145-MG-handle-insertions-of-vertices-that-already-exist-in-a-better-way 2022-11-08 07:53:45 +01:00
gvolfing
79756ae6fb Modify unit test
Instead of Creating an exception that is would be only used in this
file, just assert the type of the error the double vertex insertion
operation should yield.
2022-11-08 07:31:01 +01:00
gvolfing
91550128a5 Conform unit test with the new error-handling 2022-11-07 11:46:24 +01:00
jbajic
88487e2513 Extract vertex creation from FillUpSourceVertexSecondaryLabels 2022-11-07 10:54:28 +01:00
gvolfing
39b40ecf00 Extend the Error enum instead of a separate type
The error representing that a vertex is already inserted into the
skip-list was represented by the struct AlreadyInseertedElement. Instead
of using that struct, extend the memgraph::storage::v3::Error scoped
enum and use that to represent the double-insertion error.
2022-11-07 10:00:34 +01:00
Tyler Neely
d85fb94bc7 Merge pull request #634 from memgraph/T1122-MG-ShardManager-ThreadPool
Add shard manager thread pool
2022-11-04 16:41:53 +01:00
jbajic
4d2036249e Fix e2e test 2022-11-04 16:37:35 +01:00
Jure Bajic
d23643d4ff Merge branch 'project-pineapples' into T1148-MG-fix-expand-one-source-vertex 2022-11-04 16:19:55 +01:00
jbajic
e52ce1e363 Add expandone test 2022-11-04 16:18:18 +01:00
Tyler Neely
1abfe28806 Correctly use a moved transport interface while constructing CoordinatorWorker 2022-11-04 15:11:32 +00:00
Tyler Neely
528e30a9be Avoid warning for not using captured this. Use std::move instead of forward where appropriate 2022-11-04 14:13:15 +00:00
Tyler Neely
bb7c7f7627 Make popped messages non-const to allow for RVO 2022-11-04 14:07:49 +00:00
jbajic
bab5e1386a Address review comments 2022-11-04 15:05:55 +01:00
gvolfing
3d954e7abc Restructure SchemaResult type and uts usage
Rename SchemaResult to ShardOperationResult move it into a separate
header and add a new type to the underlying variant that indicates that
the vertex, the user would like to insert into the skip-list already
exist.
2022-11-04 15:04:25 +01:00
Tyler Neely
fa5c9a2568 Make items popped from *worker::Queue const. Use std::move instead of std::forward in one place 2022-11-04 11:33:02 +00:00
Tyler Neely
c745f8c877 Fix build after breaking code suggestion 2022-11-04 11:32:17 +00:00
Tyler Neely
486231b1b9 Update src/machine_manager/machine_manager.hpp
Co-authored-by: Kostas Kyrimis  <kostaskyrim@gmail.com>
2022-11-04 12:28:01 +01:00
Tyler Neely
43ad5855c4 Update src/machine_manager/machine_manager.hpp
Co-authored-by: Kostas Kyrimis  <kostaskyrim@gmail.com>
2022-11-04 12:27:40 +01:00
Tyler Neely
24864ff7d2 Update src/coordinator/coordinator_worker.hpp
Co-authored-by: Kostas Kyrimis  <kostaskyrim@gmail.com>
2022-11-04 12:17:57 +01:00
Tyler Neely
8598f6edf4 Fix a race condition that happens when logging from a detached thread in the cluster property test. Improve the ShardManager dtor and log statements 2022-11-04 11:14:39 +00:00
gvolfing
1cee7ecb8a Make ShardRsm aware of trying to write the same vertex into the skip-list 2022-11-04 08:12:37 +01:00
Tyler Neely
9203616283 Merge branch 'project-pineapples' into T1122-MG-ShardManager-ThreadPool 2022-11-03 17:01:07 +01:00
Tyler Neely
d198819bf1 Merge pull request #637 from memgraph/tyler_remove_storage_v2_durability_test
Remove unit tests for storage v2
2022-11-03 17:00:46 +01:00
jbajic
3a8f01af79 Fix expand one source vertex pk setting 2022-11-03 16:29:28 +01:00
Tyler Neely
7e6ec8bb26 Capture this instead of all references in scope for call to std::visit to route CoordinatorWorker messages 2022-11-03 14:38:16 +00:00
Tyler Neely
6138d76690 Remove unit tests for storage v2 2022-11-03 14:21:17 +00:00
Tyler Neely
25fdb1a1f0 Make the high_density_shard_create_scan test run much faster 2022-11-03 13:48:11 +00:00
Tyler Neely
2de1d6c359 Fix UB due to integer overflow 2022-11-03 13:27:45 +00:00
Tyler Neely
0364311dd0 Log latency histograms in the high_density_shard_create_scan test 2022-11-03 12:44:47 +00:00
Tyler Neely
9235515dab Reduce high_density_shard_create_scan shard+thread combinations. Log Raft write request demangled names 2022-11-03 12:38:12 +00:00
Tyler Neely
b83fb287ad Apply feedback from clang-tidy 2022-11-03 11:00:07 +00:00
Tyler Neely
b685a21171 Merge branch 'project-pineapples' of github.com:memgraph/memgraph into T1122-MG-ShardManager-ThreadPool 2022-11-03 09:42:28 +00:00
Tyler Neely
4db83b8159 Add a few safety checks to the new concurrent Queue structures and the promise maps in the transport layer 2022-11-03 09:41:28 +00:00
Tyler Neely
dd8dd4f6c4 Restructure responsibilities for assigning request ids to the transport handles. Simplify promise tracking to avoid replier addresses, enabling eventual direct server return (DSR) 2022-11-03 09:31:06 +00:00
Tyler Neely
fa1ddfea12 Fix a bug where the MachineManager's destructor became incorrect after being moved 2022-11-02 17:45:27 +00:00
Tyler Neely
a815ec9617 Handle Coordinator work on a separate thread, unblocking the MachineManager to route additional messages to Shards 2022-11-02 17:15:52 +00:00
Tyler Neely
78528bd609 Avoid the ShutDown explicit acknowledgement due to jthread already blocking on this 2022-11-02 17:12:21 +00:00
Tyler Neely
6239f4fc3e Simplify usage of PromiseKey in LocalTransportHandle to avoid replier address 2022-11-02 17:11:26 +00:00
János Benjamin Antal
9f71ce0f78 Merge pull request #635 from memgraph/MG-expand-fixes 2022-11-02 15:55:49 +01:00
János Benjamin Antal
589dd36bf2 Make function parameter constant 2022-11-02 14:44:54 +01:00
János Benjamin Antal
57e7169203 Eliminate dangling reference from lambda 2022-11-02 14:22:01 +01:00
János Benjamin Antal
bb3b053375 Create edge in the correct direction 2022-11-02 14:22:01 +01:00
János Benjamin Antal
a96f489756 Count the number of requests per operator 2022-11-02 14:22:01 +01:00
János Benjamin Antal
c92c795b1a Merge pull request #632 from memgraph/MG-improve-parameters 2022-11-02 14:06:30 +01:00
János Benjamin Antal
0fee412f92 Merge branch 'project-pineapples' into MG-improve-parameters 2022-11-02 13:37:35 +01:00
Tyler Neely
7596e85358 When message conversion fails to happen, demangle the concrete type name and error log a useful message 2022-11-02 12:37:34 +00:00
Tyler Neely
68654b5a19 Merge branch 'project-pineapples' of github.com:memgraph/memgraph into T1122-MG-ShardManager-ThreadPool 2022-11-02 10:55:18 +00:00
János Benjamin Antal
4d96bf5006 Merge pull request #628 from memgraph/T1138-MG-fix-edge-id-allocator
Fix passing of edge IdAllocator not by value but as pointer
2022-11-02 10:40:35 +01:00
Marko Budiselić
9bb70eb0b9 Merge branch 'project-pineapples' into T1138-MG-fix-edge-id-allocator 2022-11-01 19:38:10 +01:00
Tyler Neely
ee4be9aa5b Merge pull request #631 from memgraph/tyler_remove_shard_map_from_machine_manager
Remove redundant ShardMap copy from MachineManager to avoid race conditions
2022-11-01 18:33:37 +01:00
Tyler Neely
84509fa477 Fix merge conflict with MachineManager constructor 2022-11-01 17:07:14 +00:00
Tyler Neely
19b5fe3caf Merge branch 'project-pineapples' into tyler_remove_shard_map_from_machine_manager 2022-11-01 17:57:03 +01:00
Tyler Neely
74db669e23 Merge pull request #622 from memgraph/tyler_1k_shard_unit_test
Add test for 1k shards, 1k create vertices, and then a scan all
2022-11-01 17:56:42 +01:00
Tyler Neely
599802033f Reformat TODO comment 2022-11-01 17:21:15 +01:00
Tyler Neely
6dd57426f8 Revert broken code suggestion 2022-11-01 15:51:37 +00:00
Tyler Neely
3c9f0c48e9 Merge branch 'project-pineapples' into tyler_1k_shard_unit_test 2022-11-01 16:31:29 +01:00
Tyler Neely
9a1258a708 Update tests/unit/1k_shards_1k_create_scanall.cpp
Co-authored-by: Kostas Kyrimis  <kostaskyrim@gmail.com>
2022-11-01 16:31:14 +01:00
Tyler Neely
c55ca836a4 Update tests/unit/1k_shards_1k_create_scanall.cpp
Co-authored-by: Kostas Kyrimis  <kostaskyrim@gmail.com>
2022-11-01 16:31:06 +01:00
Tyler Neely
5674ef4016 Update tests/unit/1k_shards_1k_create_scanall.cpp
Co-authored-by: Kostas Kyrimis  <kostaskyrim@gmail.com>
2022-11-01 16:30:37 +01:00
Tyler Neely
77c2afc9e7 Add TODO about ID mapping improvements 2022-11-01 15:27:35 +00:00
Tyler Neely
27292dd921 Merge branch 'project-pineapples' into tyler_remove_shard_map_from_machine_manager 2022-11-01 16:20:54 +01:00
János Benjamin Antal
4608af9d00 Merge branch 'project-pineapples' into MG-improve-parameters 2022-11-01 15:56:16 +01:00
János Benjamin Antal
50e72a7c28 Merge branch 'project-pineapples' into T1138-MG-fix-edge-id-allocator 2022-11-01 15:56:12 +01:00
János Benjamin Antal
17a104677d Merge pull request #617 from memgraph/T1105-MG-profile-query-in-distributed 2022-11-01 15:55:52 +01:00
Tyler Neely
a6add80fc9 Use static RSM partitioning function for achieving a smooth Shard->ShardWorker distribution 2022-11-01 14:52:38 +00:00
Tyler Neely
5d3eaf6a55 Rename 1k_shards_1k_create_scanall test to high_density_shard_create_scan 2022-11-01 14:02:07 +00:00
Tyler Neely
a13f260236 Parameterize shard worker threads in the MachineConfig and simplify test output 2022-11-01 13:57:57 +00:00
János Benjamin Antal
7b4b1ba8ed Speed up Parameters::AtTokenPosition 2022-11-01 11:51:35 +01:00
Tyler Neely
bb1e8aa164 Improve histogram output by adding a pretty table printing function 2022-11-01 10:46:25 +00:00
Tyler Neely
27a1311966 Merge branch 'tyler_1k_shard_unit_test' of github.com:memgraph/memgraph into T1122-MG-ShardManager-ThreadPool 2022-11-01 09:49:15 +00:00
János Benjamin Antal
30ff6487f2 Merge remote-tracking branch 'origin/project-pineapples' into T1105-MG-profile-query-in-distributed 2022-11-01 10:32:44 +01:00
Tyler Neely
119da2d7a7 Update CMakeLists.txt 2022-11-01 09:51:43 +01:00
Tyler Neely
fa972813d2 Update src/coordinator/shard_map.cpp
Co-authored-by: Jure Bajic <jure.bajic@memgraph.com>
2022-10-31 19:05:19 +01:00
Tyler Neely
951b058116 Complete migration from single-threaded ShardManager to multi-threaded ShardWorker processing 2022-10-31 18:04:30 +00:00
Tyler Neely
d0cad6e6ba Temporarily duplicate shard management logic from ShardManger in ShardWorker 2022-10-31 17:25:08 +00:00
Tyler Neely
6138277972 Merge branch 'tyler_remove_shard_map_from_machine_manager' of github.com:memgraph/memgraph into T1122-MG-ShardManager-ThreadPool 2022-10-31 17:14:01 +00:00
Tyler Neely
2f77eb96ff Remove redundant ShardMap copy from MachineManager to avoid race conditions 2022-10-31 16:49:46 +00:00
Tyler Neely
d7bc93c55f Fill Cron next time promise from each worker thread 2022-10-31 16:17:34 +00:00
Tyler Neely
039d5f51d3 Merge branch 'project-pineapples' into T1122-MG-ShardManager-ThreadPool 2022-10-31 17:06:36 +01:00
Tyler Neely
cebe6f62fa Implement skeleton worker threadpool for the ShardManager 2022-10-31 16:03:47 +00:00
jbajic
6834ce01fe Fix edge id allocator 2022-10-31 16:26:56 +01:00
János Benjamin Antal
f9e2a66961 Add const qualifier to function parameters 2022-10-31 15:53:04 +01:00
János Benjamin Antal
d5700ab5ff Use [[unlikely]] attribute 2022-10-31 15:53:04 +01:00
Tyler Neely
b0aaba6765 Merge branch 'project-pineapples' into tyler_1k_shard_unit_test 2022-10-31 15:47:38 +01:00
Kostas Kyrimis
59c7d81ae8 [🍍 < T1086-MG] Test distributed operators e2e (#607)
* Fix Explain queries
* Add Vertex/Edge accessor support for properties
* Fix projections
* Fix expansions to fetch destination vertex properties
* Fix improper use of ShardMap on bolt and replaced it with the ShardRequestManager 
* Add NameToId mappers on ShardRequestManager
* Add e2e tests for operators
* Fix OPTIONAL MATCH
2022-10-31 11:52:20 +02:00
János Benjamin Antal
d920d7c293 Merge branch 'project-pineapples' into T1105-MG-profile-query-in-distributed 2022-10-31 09:27:05 +01:00
Tyler Neely
6b0168cb3d Add LocalTransport::ResponseLatencies 2022-10-28 08:51:12 +00:00
Tyler Neely
03cf264b76 Add test for 1k shards, 1k create vertices, and then a scan all 2022-10-28 08:29:27 +00:00
Tyler Neely
9448e23dc9 Check-in basic shard scheduler skeleton 2022-10-27 13:36:53 +00:00
Tyler Neely
eafccaea84 Add a logarithmically bucketing histogram and record response latencies per message type (#616) 2022-10-26 15:57:11 +02:00
János Benjamin Antal
0c4f591b3c Merge branch 'project-pineapples' into T1105-MG-profile-query-in-distributed 2022-10-26 14:48:27 +02:00
János Benjamin Antal
534e365271 Suppress warning about exception escape for destructor 2022-10-26 11:22:27 +02:00
János Benjamin Antal
281ae158ec Make ReadTSC noexcept 2022-10-26 11:21:27 +02:00
János Benjamin Antal
39c9c215b1 Suppress clang-tidy warnings for 2022-10-26 11:14:24 +02:00
János Benjamin Antal
f6e78ce6da Fix TODOs in storage engine (#614)
Fixed various TODOs that were easy to fix to improve the code quality of the
newly implemented storage.
2022-10-26 10:40:35 +02:00
János Benjamin Antal
6a31c49432 Flatten nested conditional statements 2022-10-25 20:30:29 +02:00
János Benjamin Antal
fea92f4829 Merge remote-tracking branch 'origin/project-pineapples' into T1105-MG-profile-query-in-distributed 2022-10-25 20:28:54 +02:00
János Benjamin Antal
1703cd039d Populate custom data of profile query with request wait times 2022-10-25 20:15:06 +02:00
gvolfing
ca2351124b Make primary labels act as label indices (#605)
Because of the lexicographical sharding, the primary labels themselves are
acting as indexes. If a primary label is specified in a MATCH query we can
safely narrow the range of shards we have to scan through based on that label.
This PR introduces the necessary changes in order to achieve that.
2022-10-25 19:48:17 +02:00
János Benjamin Antal
5784c0d473 Return the custom data for profile queries 2022-10-25 15:26:42 +02:00
János Benjamin Antal
8ebc704819 Fix profile queries with ScanAll 2022-10-25 14:37:18 +02:00
János Benjamin Antal
5939fb2b0c Start transaction properly 2022-10-25 14:35:26 +02:00
János Benjamin Antal
ee64684b0b Add ScopedCustomProfile 2022-10-25 13:37:57 +02:00
János Benjamin Antal
1e4c02f8a5 Make profile query work 2022-10-25 13:37:27 +02:00
János Benjamin Antal
8353228ba7 Upgrade json lib to 3.11.2 2022-10-25 13:10:03 +02:00
János Benjamin Antal
332afadf21 Split file parsing (#600)
Add temporary support for split files. This is only temporary solution until we
get the shard splitting implemented.
2022-10-25 10:27:13 +02:00
Tyler Neely
acc655f4fd Model-based testing of simulated full cluster (#584)
This PR adds support for generating randomized workloads that will be executed
against a simulated cluster, as well as against a correctness model. Initially
this just generates ScanAll and CreateVertex requests, and anything that it
creates, it also inserts into a `std::set`, and when we do a ScanAll, it asserts
that we get the same number of requests back. This will become much more
sophisticated over time, but it's already hitting pay-dirt.
2022-10-24 19:54:09 +02:00
Tyler Neely
12e7a261aa operator<< implementations for a few foundational types (#604) 2022-10-21 15:25:40 +02:00
Marko Budiselić
1c3bb969e9 Decouple interactive planning manual test (#585) 2022-10-21 13:34:13 +02:00
Jure Bajic
e5437080c5 Handle OrderBy in ScanVertices request (#594)
- Refactor shard_rsm and move function into expr.hpp/expr.cpp and request_helper.hpp/request_helper.cpp
2022-10-21 12:33:40 +02:00
Tyler Neely
17090dd8ac Load SPDLOG_LEVEL environment variable in unit tests binaries (#606)
Co-authored-by: János Benjamin Antal <antaljanosbenjamin@users.noreply.github.com>
2022-10-20 18:27:52 +02:00
gvolfing
281cebd386 Add filtering capabilities to ExpandOne request (#581) 2022-10-20 17:51:44 +02:00
János Benjamin Antal
f89a2bbf42 Make ExpandOne work in the query engine (#589) 2022-10-20 11:35:00 +02:00
gvolfing
85b8ce9101 Implement filtering capabilities for ScanAll (#578)
ScanVerticesrequest was not able to utilize filtering capabilities
before. With these modification it is now able to filter the scanned
vertices based on the filter_expressions field in the
ScanVerticesRequest message type.
2022-10-19 16:09:00 +02:00
Jure Bajic
6bb40a7f49 Create ExpandOne request (#580)
Refactor CreateEdge into CreateExpand
2022-10-19 13:55:46 +02:00
gvolfing
07f34838bd Add e2e test to check ScanAll on new architecture (#603)
Add testing capabilities similar to what existed before on the older
architecture. With this test specifically, test if we can create
vertices and return them with ScanAll. There is a hard-coded 3s wait in
the test which is needed because the MachineManager need this time to
initialize itself.
2022-10-18 21:38:56 +02:00
gvolfing
d06132cb33 Make ShardRequestManager work with futures (#588)
The communication between the ShardRequestManager and the RsmClient
used to be direct. In this PR this changes into a future-based
communication type. The RsmClient stores state about the currently
processed future (either read or write request) and exposes blocking
and non-blocking functionality to obtain the filled future. The
ShardRequestManager -for now- will send of the set of requests present
in the ExecutionState and block on each of them until the requests are
completed or the set of paginated responses(caused by, for example the
batch-limit in ScanAll) are ready for the next round.
2022-10-18 09:53:37 +02:00
Marko Budiselić
5347c06d76 Add shutdown of LocalSystem during the shutdown process (#592) 2022-10-14 11:59:50 +02:00
Tyler Neely
4634ac484a Optimize simulator RNG usage (#590)
This causes instructions for `basic_request.cpp` to drop from 281 million to 28 million
2022-10-12 19:19:06 +03:00
Jeremy B
4cb3b064c4 Add filter to scan all (#575)
Add several versions of ScanAll with filters.
Add helper function to transform an expression into string that can be parsed again once on the storage.
2022-10-12 11:46:59 +02:00
Jure Bajic
23171e76b6 Integrate bolt server (#572)
* Use query-v2 in the main executable
* Set up machine manager in memgraph
* Add `ShardRequestManager` to `Interpreter`
* Make vertex creation work
* Make scan all work
* Add edge type map in shard request manager
* Send schema over request
* Empty out DbAccessor
* Store shard mapping at creation
* Remove failing CI steps

Cooltura is the best place in Zagreb!

Co-authored-by: János Benjamin Antal <benjamin.antal@memgraph.io>
2022-10-11 16:31:46 +02:00
Jeremy B
6fd64d31f2 Update usage of Shard (#574)
Updating tests to use new constructor of Shard
Commenting test shard_request_manager
2022-10-05 11:56:36 +02:00
gvolfing
87111b2f89 Implement missing message handlers (#563)
Implement the missing message handlers with basic functionality. The
implementation does not include any capabilities to filter based on
expressions.
2022-10-03 15:31:06 +02:00
Tyler Neely
b5c7078c7d Stitch request manager with shard (#570)
Fix various bugs
2022-09-23 20:07:41 +02:00
Kostas Kyrimis
925835b080 Implement query engine client (#531)
- Add shard request manager
2022-09-22 15:05:43 +02:00
Tyler Neely
ce788f5f65 Machine manager and shard stitch (#569) 2022-09-22 13:55:16 +02:00
János Benjamin Antal
b8186bea2e Lexicographically ordered storage (#454) 2022-09-21 18:58:31 +02:00
János Benjamin Antal
3c4856dcb7 Merge remote-tracking branch 'origin/project-pineapples' into E118-MG-lexicographically-ordered-storage 2022-09-21 18:50:00 +02:00
Marko Budiselić
b4d6dc0930 Add proper transaction handling (#550) 2022-09-21 18:25:51 +02:00
Jure Bajic
817161a915 Implement expression evaluator in storage v3 (#534)
- Enable `mg-expr` in mg-storage-v3
-  Adapt id_mapper
- Change conversion function from PropertyValue to TypedValue
- Remove memgraph functions
- Enable expression tests on for storage
2022-09-21 17:26:55 +02:00
Tyler Neely
58eb2caf0f Add machine manager prototype (#533) 2022-09-21 16:57:25 +02:00
gvolfing
ecda71168c Implement message based actions (#538)
Create shard-side handlers for basic messages

Implement the handlers for CreateVertices, CreateEdges and ScanAll. Use
or modify the defined messages to interact with individual Shards and
test their behavior. Shard is currently being owned by ShardRsm
instances. The two top level dispatching functions Read() and Apply()
are responsible for read- and write operations respectively. Currently
there are a handful of messages that are defined but not utilized, these
will be used in the near future, as well as a couple of handler
functions with empty implementations.
2022-09-20 11:15:19 +02:00
240 changed files with 18100 additions and 41682 deletions

View File

@@ -164,52 +164,14 @@ jobs:
cmake ..
make -j$THREADS
- name: Run leftover CTest tests
- name: Run simulation tests
run: |
# Activate toolchain.
source /opt/toolchain-v4/activate
# Run leftover CTest tests (all except unit and benchmark tests).
# Run simulation tests.
cd build
ctest -E "(memgraph__unit|memgraph__benchmark|memgraph__simulation)" --output-on-failure
- name: Run drivers tests
run: |
./tests/drivers/run.sh
- name: Run integration tests
run: |
cd tests/integration
for name in *; do
if [ ! -d $name ]; then continue; fi
pushd $name >/dev/null
echo "Running: $name"
if [ -x prepare.sh ]; then
./prepare.sh
fi
if [ -x runner.py ]; then
./runner.py
elif [ -x runner.sh ]; then
./runner.sh
fi
echo
popd >/dev/null
done
- name: Run cppcheck and clang-format
run: |
# Activate toolchain.
source /opt/toolchain-v4/activate
# Run cppcheck and clang-format.
cd tools/github
./cppcheck_and_clang_format diff
- name: Save cppcheck and clang-format errors
uses: actions/upload-artifact@v2
with:
name: "Code coverage"
path: tools/github/cppcheck_and_clang_format.txt
ctest -R memgraph__simulation --output-on-failure -j$THREADS
release_build:
name: "Release build"
@@ -240,19 +202,6 @@ jobs:
cmake -DCMAKE_BUILD_TYPE=release ..
make -j$THREADS
- name: Run GQL Behave tests
run: |
cd tests/gql_behave
./continuous_integration
- name: Save quality assurance status
uses: actions/upload-artifact@v2
with:
name: "GQL Behave Status"
path: |
tests/gql_behave/gql_behave_status.csv
tests/gql_behave/gql_behave_status.html
- name: Run unit tests
run: |
# Activate toolchain.
@@ -267,7 +216,7 @@ jobs:
# Activate toolchain.
source /opt/toolchain-v4/activate
# Run unit tests.
# Run simulation tests.
cd build
ctest -R memgraph__simulation --output-on-failure -j$THREADS
@@ -278,167 +227,4 @@ jobs:
./setup.sh
source ve3/bin/activate
cd e2e
LD_LIBRARY_PATH=$LD_LIBRARY_PATH:../../libs/mgclient/lib python runner.py --workloads-root-directory .
- name: Run stress test (plain)
run: |
cd tests/stress
./continuous_integration
- name: Run stress test (SSL)
run: |
cd tests/stress
./continuous_integration --use-ssl
- name: Run durability test
run: |
cd tests/stress
source ve3/bin/activate
python3 durability --num-steps 5
- name: Create enterprise DEB package
run: |
# Activate toolchain.
source /opt/toolchain-v4/activate
cd build
# create mgconsole
# we use the -B to force the build
make -j$THREADS -B mgconsole
# Create enterprise DEB package.
mkdir output && cd output
cpack -G DEB --config ../CPackConfig.cmake
- name: Save enterprise DEB package
uses: actions/upload-artifact@v2
with:
name: "Enterprise DEB package"
path: build/output/memgraph*.deb
- name: Save test data
uses: actions/upload-artifact@v2
if: always()
with:
name: "Test data"
path: |
# multiple paths could be defined
build/logs
release_jepsen_test:
name: "Release Jepsen Test"
runs-on: [self-hosted, Linux, X64, Debian10, JepsenControl]
#continue-on-error: true
env:
THREADS: 24
MEMGRAPH_ENTERPRISE_LICENSE: ${{ secrets.MEMGRAPH_ENTERPRISE_LICENSE }}
MEMGRAPH_ORGANIZATION_NAME: ${{ secrets.MEMGRAPH_ORGANIZATION_NAME }}
steps:
- name: Set up repository
uses: actions/checkout@v2
with:
# Number of commits to fetch. `0` indicates all history for all
# branches and tags. (default: 1)
fetch-depth: 0
- name: Build release binaries
run: |
# Activate toolchain.
source /opt/toolchain-v4/activate
# Initialize dependencies.
./init
# Build only memgraph release binarie.
cd build
cmake -DCMAKE_BUILD_TYPE=release ..
make -j$THREADS memgraph
- name: Run Jepsen tests
run: |
cd tests/jepsen
./run.sh test --binary ../../build/memgraph --run-args "test-all --node-configs resources/node-config.edn" --ignore-run-stdout-logs --ignore-run-stderr-logs
- name: Save Jepsen report
uses: actions/upload-artifact@v2
if: ${{ always() }}
with:
name: "Jepsen Report"
path: tests/jepsen/Jepsen.tar.gz
release_benchmarks:
name: "Release benchmarks"
runs-on: [self-hosted, Linux, X64, Diff, Gen7]
env:
THREADS: 24
MEMGRAPH_ENTERPRISE_LICENSE: ${{ secrets.MEMGRAPH_ENTERPRISE_LICENSE }}
MEMGRAPH_ORGANIZATION_NAME: ${{ secrets.MEMGRAPH_ORGANIZATION_NAME }}
steps:
- name: Set up repository
uses: actions/checkout@v2
with:
# Number of commits to fetch. `0` indicates all history for all
# branches and tags. (default: 1)
fetch-depth: 0
- name: Build release binaries
run: |
# Activate toolchain.
source /opt/toolchain-v4/activate
# Initialize dependencies.
./init
# Build only memgraph release binaries.
cd build
cmake -DCMAKE_BUILD_TYPE=release ..
make -j$THREADS
- name: Run macro benchmarks
run: |
cd tests/macro_benchmark
./harness QuerySuite MemgraphRunner \
--groups aggregation 1000_create unwind_create dense_expand match \
--no-strict
- name: Get branch name (merge)
if: github.event_name != 'pull_request'
shell: bash
run: echo "BRANCH_NAME=$(echo ${GITHUB_REF#refs/heads/} | tr / -)" >> $GITHUB_ENV
- name: Get branch name (pull request)
if: github.event_name == 'pull_request'
shell: bash
run: echo "BRANCH_NAME=$(echo ${GITHUB_HEAD_REF} | tr / -)" >> $GITHUB_ENV
- name: Upload macro benchmark results
run: |
cd tools/bench-graph-client
virtualenv -p python3 ve3
source ve3/bin/activate
pip install -r requirements.txt
./main.py --benchmark-name "macro_benchmark" \
--benchmark-results-path "../../tests/macro_benchmark/.harness_summary" \
--github-run-id "${{ github.run_id }}" \
--github-run-number "${{ github.run_number }}" \
--head-branch-name "${{ env.BRANCH_NAME }}"
- name: Run mgbench
run: |
cd tests/mgbench
./benchmark.py --num-workers-for-benchmark 12 --export-results benchmark_result.json pokec/medium/*/*
- name: Upload mgbench results
run: |
cd tools/bench-graph-client
virtualenv -p python3 ve3
source ve3/bin/activate
pip install -r requirements.txt
./main.py --benchmark-name "mgbench" \
--benchmark-results-path "../../tests/mgbench/benchmark_result.json" \
--github-run-id "${{ github.run_id }}" \
--github-run-number "${{ github.run_number }}" \
--head-branch-name "${{ env.BRANCH_NAME }}"
LD_LIBRARY_PATH=$LD_LIBRARY_PATH:../../libs/mgclient/lib python runner.py --workloads-root-directory ./distributed_queries

4
.gitignore vendored
View File

@@ -54,6 +54,7 @@ src/distributed/token_sharing_rpc_messages.hpp
src/distributed/updates_rpc_messages.hpp
src/query/v2/frontend/ast/ast.hpp
src/query/frontend/ast/ast.hpp
src/storage/v3/bindings/ast/ast.hpp
src/query/distributed/frontend/ast/ast_serialization.hpp
src/query/v2/distributed/frontend/ast/ast_serialization.hpp
src/durability/distributed/state_delta.hpp
@@ -61,12 +62,15 @@ src/durability/single_node/state_delta.hpp
src/durability/single_node_ha/state_delta.hpp
src/query/frontend/semantic/symbol.hpp
src/query/v2/frontend/semantic/symbol.hpp
src/expr/semantic/symbol.hpp
src/query/distributed/frontend/semantic/symbol_serialization.hpp
src/query/v2/distributed/frontend/semantic/symbol_serialization.hpp
src/query/distributed/plan/ops.hpp
src/query/v2/distributed/plan/ops.hpp
src/query/plan/operator.hpp
src/query/v2/plan/operator.hpp
src/parser/opencypher/generated
src/expr/semantic/symbol.hpp
src/raft/log_entry.hpp
src/raft/raft_rpc_messages.hpp
src/raft/snapshot_metadata.hpp

View File

@@ -83,13 +83,9 @@ modifications:
value: "true"
override: true
- name: "query_modules_directory"
value: "/usr/lib/memgraph/query_modules"
override: true
- name: "auth_module_executable"
value: "/usr/lib/memgraph/auth_module/example.py"
override: false
# - name: "query_modules_directory"
# value: "/usr/lib/memgraph/query_modules"
# override: true
- name: "memory_limit"
value: "0"

View File

@@ -116,7 +116,7 @@ declare -A primary_urls=(
["pymgclient"]="http://$local_cache_host/git/pymgclient.git"
["mgconsole"]="http://$local_cache_host/git/mgconsole.git"
["spdlog"]="http://$local_cache_host/git/spdlog"
["nlohmann"]="http://$local_cache_host/file/nlohmann/json/4f8fba14066156b73f1189a2b8bd568bde5284c5/single_include/nlohmann/json.hpp"
["nlohmann"]="http://$local_cache_host/file/nlohmann/json/9d69186291aca4f0137b69c1dee313b391ff564c/single_include/nlohmann/json.hpp"
["neo4j"]="http://$local_cache_host/file/neo4j-community-3.2.3-unix.tar.gz"
["librdkafka"]="http://$local_cache_host/git/librdkafka.git"
["protobuf"]="http://$local_cache_host/git/protobuf.git"
@@ -141,7 +141,7 @@ declare -A secondary_urls=(
["pymgclient"]="https://github.com/memgraph/pymgclient.git"
["mgconsole"]="http://github.com/memgraph/mgconsole.git"
["spdlog"]="https://github.com/gabime/spdlog"
["nlohmann"]="https://raw.githubusercontent.com/nlohmann/json/4f8fba14066156b73f1189a2b8bd568bde5284c5/single_include/nlohmann/json.hpp"
["nlohmann"]="https://raw.githubusercontent.com/nlohmann/json/9d69186291aca4f0137b69c1dee313b391ff564c/single_include/nlohmann/json.hpp"
["neo4j"]="https://s3-eu-west-1.amazonaws.com/deps.memgraph.io/neo4j-community-3.2.3-unix.tar.gz"
["librdkafka"]="https://github.com/edenhill/librdkafka.git"
["protobuf"]="https://github.com/protocolbuffers/protobuf.git"

View File

@@ -37,13 +37,13 @@ include_directories(${CMAKE_CURRENT_BINARY_DIR})
# Memgraph Single Node v2 Executable
# ----------------------------------------------------------------------------
set(mg_single_node_v2_sources
glue/communication.cpp
memgraph.cpp
glue/auth.cpp
glue/v2/communication.cpp
memgraph.cpp
glue/v2/auth.cpp
)
set(mg_single_node_v2_libs stdc++fs Threads::Threads
telemetry_lib mg-query mg-communication mg-memory mg-utils mg-auth mg-license mg-settings)
telemetry_lib mg-query-v2 mg-communication mg-memory mg-utils mg-auth mg-license mg-settings mg-io mg-coordinator)
if (MG_ENTERPRISE)
# These are enterprise subsystems
set(mg_single_node_v2_libs ${mg_single_node_v2_libs} mg-audit)
@@ -126,19 +126,3 @@ install(CODE "file(MAKE_DIRECTORY \$ENV{DESTDIR}/var/log/memgraph
# ----------------------------------------------------------------------------
# Memgraph CSV Import Tool Executable
# ----------------------------------------------------------------------------
add_executable(mg_import_csv mg_import_csv.cpp)
target_link_libraries(mg_import_csv mg-storage-v2)
# Strip the executable in release build.
if (lower_build_type STREQUAL "release")
add_custom_command(TARGET mg_import_csv POST_BUILD
COMMAND strip -s mg_import_csv
COMMENT "Stripping symbols and sections from mg_import_csv")
endif()
install(TARGETS mg_import_csv RUNTIME DESTINATION bin)
# ----------------------------------------------------------------------------
# Memgraph CSV Import Tool Executable
# ----------------------------------------------------------------------------

View File

@@ -30,6 +30,7 @@
#include "communication/context.hpp"
#include "communication/v2/pool.hpp"
#include "communication/v2/session.hpp"
#include "utils/message.hpp"
#include "utils/spin_lock.hpp"
#include "utils/synchronized.hpp"
@@ -58,10 +59,10 @@ class Listener final : public std::enable_shared_from_this<Listener<TSession, TS
bool IsRunning() const noexcept { return alive_.load(std::memory_order_relaxed); }
private:
Listener(boost::asio::io_context &io_context, TSessionData *data, ServerContext *server_context,
Listener(boost::asio::io_context &io_context, TSessionData &data, ServerContext *server_context,
tcp::endpoint &endpoint, const std::string_view service_name, const uint64_t inactivity_timeout_sec)
: io_context_(io_context),
data_(data),
data_(&data),
server_context_(server_context),
acceptor_(io_context_),
endpoint_{endpoint},
@@ -110,7 +111,7 @@ class Listener final : public std::enable_shared_from_this<Listener<TSession, TS
return OnError(ec, "accept");
}
auto session = SessionHandler::Create(std::move(socket), data_, *server_context_, endpoint_, inactivity_timeout_,
auto session = SessionHandler::Create(std::move(socket), *data_, *server_context_, endpoint_, inactivity_timeout_,
service_name_);
session->Start();
DoAccept();

View File

@@ -72,7 +72,7 @@ class Server final {
* Constructs and binds server to endpoint, operates on session data and
* invokes workers_count workers
*/
Server(ServerEndpoint &endpoint, TSessionData *session_data, ServerContext *server_context,
Server(ServerEndpoint &endpoint, TSessionData &session_data, ServerContext *server_context,
const int inactivity_timeout_sec, const std::string_view service_name,
size_t workers_count = std::thread::hardware_concurrency())
: endpoint_{endpoint},

View File

@@ -41,6 +41,7 @@
#include <boost/beast/websocket/rfc6455.hpp>
#include <boost/system/detail/error_code.hpp>
#include "communication/buffer.hpp"
#include "communication/context.hpp"
#include "communication/exceptions.hpp"
#include "utils/logging.hpp"
@@ -139,7 +140,7 @@ class WebsocketSession : public std::enable_shared_from_this<WebsocketSession<TS
private:
// Take ownership of the socket
explicit WebsocketSession(tcp::socket &&socket, TSessionData *data, tcp::endpoint endpoint,
explicit WebsocketSession(tcp::socket &&socket, TSessionData &data, tcp::endpoint endpoint,
std::string_view service_name)
: ws_(std::move(socket)),
strand_{boost::asio::make_strand(ws_.get_executor())},
@@ -311,13 +312,13 @@ class Session final : public std::enable_shared_from_this<Session<TSession, TSes
}
private:
explicit Session(tcp::socket &&socket, TSessionData *data, ServerContext &server_context, tcp::endpoint endpoint,
explicit Session(tcp::socket &&socket, TSessionData &data, ServerContext &server_context, tcp::endpoint endpoint,
const std::chrono::seconds inactivity_timeout_sec, std::string_view service_name)
: socket_(CreateSocket(std::move(socket), server_context)),
strand_{boost::asio::make_strand(GetExecutor())},
output_stream_([this](const uint8_t *data, size_t len, bool have_more) { return Write(data, len, have_more); }),
session_(data, endpoint, input_buffer_.read_end(), &output_stream_),
data_{data},
data_{&data},
endpoint_{endpoint},
remote_endpoint_{GetRemoteEndpoint()},
service_name_{service_name},
@@ -373,7 +374,7 @@ class Session final : public std::enable_shared_from_this<Session<TSession, TSes
spdlog::info("Switching {} to websocket connection", remote_endpoint_);
if (std::holds_alternative<TCPSocket>(socket_)) {
auto sock = std::get<TCPSocket>(std::move(socket_));
WebsocketSession<TSession, TSessionData>::Create(std::move(sock), data_, endpoint_, service_name_)
WebsocketSession<TSession, TSessionData>::Create(std::move(sock), *data_, endpoint_, service_name_)
->DoAccept(parser.release());
execution_active_ = false;
return;
@@ -465,7 +466,7 @@ class Session final : public std::enable_shared_from_this<Session<TSession, TSes
if (timeout_timer_.expiry() <= boost::asio::steady_timer::clock_type::now()) {
// The deadline has passed. Stop the session. The other actors will
// terminate as soon as possible.
spdlog::info("Shutting down session after {} of inactivity", timeout_seconds_);
spdlog::info("Shutting down session after {} of inactivity", timeout_seconds_.count());
DoShutdown();
} else {
// Put the actor back to sleep.

View File

@@ -1,10 +1,9 @@
set(coordinator_src_files
coordinator.hpp
shard_map.hpp
hybrid_logical_clock.hpp)
coordinator.cpp
shard_map.cpp)
find_package(fmt REQUIRED)
find_package(Threads REQUIRED)
add_library(mg-coordinator STATIC ${coordinator_src_files})
target_link_libraries(mg-coordinator stdc++fs Threads::Threads fmt::fmt mg-utils)
target_link_libraries(mg-coordinator stdc++fs Threads::Threads fmt::fmt mg-utils mg-storage-v3)

View File

@@ -0,0 +1,129 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#include <coordinator/coordinator.hpp>
namespace memgraph::coordinator {
CoordinatorWriteResponses Coordinator::ApplyWrite(HeartbeatRequest &&heartbeat_request) {
spdlog::info("Coordinator handling HeartbeatRequest");
// add this storage engine to any under-replicated shards that it is not already a part of
auto initializing_rsms_for_shard_manager =
shard_map_.AssignShards(heartbeat_request.from_storage_manager, heartbeat_request.initialized_rsms);
return HeartbeatResponse{
.shards_to_initialize = initializing_rsms_for_shard_manager,
};
}
CoordinatorWriteResponses Coordinator::ApplyWrite(HlcRequest &&hlc_request) {
HlcResponse res{};
auto hlc_shard_map = shard_map_.GetHlc();
MG_ASSERT(!(hlc_request.last_shard_map_version.logical_id > hlc_shard_map.logical_id));
res.new_hlc = Hlc{
.logical_id = ++highest_allocated_timestamp_,
// TODO(tyler) probably pass some more context to the Coordinator here
// so that we can use our wall clock and enforce monotonicity.
// .coordinator_wall_clock = io_.Now(),
};
// Allways return fresher shard_map for now.
res.fresher_shard_map = std::make_optional(shard_map_);
return res;
}
CoordinatorWriteResponses Coordinator::ApplyWrite(AllocateEdgeIdBatchRequest &&ahr) {
AllocateEdgeIdBatchResponse res{};
uint64_t low = highest_allocated_edge_id_;
highest_allocated_edge_id_ += ahr.batch_size;
uint64_t high = highest_allocated_edge_id_;
res.low = low;
res.high = high;
return res;
}
/// This splits the shard immediately beneath the provided
/// split key, keeping the assigned peers identical for now,
/// but letting them be gradually migrated over time.
CoordinatorWriteResponses Coordinator::ApplyWrite(SplitShardRequest &&split_shard_request) {
SplitShardResponse res{};
if (split_shard_request.previous_shard_map_version != shard_map_.shard_map_version) {
res.success = false;
} else {
res.success = shard_map_.SplitShard(split_shard_request.previous_shard_map_version, split_shard_request.label_id,
split_shard_request.split_key);
}
return res;
}
/// This adds the provided storage engine to the standby storage engine pool,
/// which can be used to rebalance storage over time.
CoordinatorWriteResponses Coordinator::ApplyWrite(
RegisterStorageEngineRequest && /* register_storage_engine_request */) {
RegisterStorageEngineResponse res{};
// TODO
return res;
}
/// This begins the process of draining the provided storage engine from all raft
/// clusters that it might be participating in.
CoordinatorWriteResponses Coordinator::ApplyWrite(
DeregisterStorageEngineRequest && /* register_storage_engine_request */) {
DeregisterStorageEngineResponse res{};
// TODO
return res;
}
CoordinatorWriteResponses Coordinator::ApplyWrite(InitializeLabelRequest &&initialize_label_request) {
InitializeLabelResponse res{};
std::optional<LabelId> new_label_id = shard_map_.InitializeNewLabel(
initialize_label_request.label_name, initialize_label_request.schema, initialize_label_request.replication_factor,
initialize_label_request.last_shard_map_version);
if (new_label_id) {
res.new_label_id = new_label_id.value();
res.fresher_shard_map = std::nullopt;
res.success = true;
} else {
res.fresher_shard_map = shard_map_;
res.success = false;
}
return res;
}
CoordinatorWriteResponses Coordinator::ApplyWrite(AllocatePropertyIdsRequest &&allocate_property_ids_request) {
AllocatePropertyIdsResponse res{};
auto property_ids = shard_map_.AllocatePropertyIds(allocate_property_ids_request.property_names);
res.property_ids = property_ids;
return res;
}
} // namespace memgraph::coordinator

View File

@@ -12,26 +12,30 @@
#pragma once
#include <optional>
#include <set>
#include <string>
#include <unordered_set>
#include <variant>
#include <vector>
#include "coordinator/hybrid_logical_clock.hpp"
#include "coordinator/shard_map.hpp"
#include "io/simulator/simulator.hpp"
#include "io/time.hpp"
#include "io/transport.hpp"
#include "storage/v3/id_types.hpp"
#include "storage/v3/schemas.hpp"
#include <boost/uuid/uuid.hpp>
#include <coordinator/hybrid_logical_clock.hpp>
#include <coordinator/shard_map.hpp>
#include <io/simulator/simulator.hpp>
#include <io/time.hpp>
#include <io/transport.hpp>
#include <storage/v3/id_types.hpp>
#include <storage/v3/schemas.hpp>
namespace memgraph::coordinator {
using memgraph::io::Address;
using memgraph::storage::v3::LabelId;
using memgraph::storage::v3::PropertyId;
using Address = memgraph::io::Address;
using SimT = memgraph::io::simulator::SimulatorTransport;
using memgraph::storage::v3::SchemaProperty;
using SimT = memgraph::io::simulator::SimulatorTransport;
using PrimaryKey = std::vector<PropertyValue>;
struct HlcRequest {
Hlc last_shard_map_version;
@@ -50,6 +54,11 @@ struct GetShardMapResponse {
ShardMap shard_map;
};
struct AllocateHlcBatchRequest {
Hlc low;
Hlc high;
};
struct AllocateHlcBatchResponse {
bool success;
Hlc low;
@@ -76,7 +85,7 @@ struct AllocatePropertyIdsResponse {
struct SplitShardRequest {
Hlc previous_shard_map_version;
LabelId label_id;
CompoundKey split_key;
PrimaryKey split_key;
};
struct SplitShardResponse {
@@ -102,26 +111,34 @@ struct DeregisterStorageEngineResponse {
struct InitializeLabelRequest {
std::string label_name;
std::vector<SchemaProperty> schema;
size_t replication_factor;
Hlc last_shard_map_version;
};
struct InitializeLabelResponse {
bool success;
LabelId new_label_id;
std::optional<ShardMap> fresher_shard_map;
};
struct HeartbeatRequest {};
struct HeartbeatResponse {};
struct HeartbeatRequest {
Address from_storage_manager;
std::set<boost::uuids::uuid> initialized_rsms;
};
struct HeartbeatResponse {
std::vector<ShardToInitialize> shards_to_initialize;
};
using CoordinatorWriteRequests =
std::variant<HlcRequest, AllocateEdgeIdBatchRequest, SplitShardRequest, RegisterStorageEngineRequest,
DeregisterStorageEngineRequest, InitializeLabelRequest, AllocatePropertyIdsRequest>;
using CoordinatorWriteResponses =
std::variant<HlcResponse, AllocateEdgeIdBatchResponse, SplitShardResponse, RegisterStorageEngineResponse,
DeregisterStorageEngineResponse, InitializeLabelResponse, AllocatePropertyIdsResponse>;
DeregisterStorageEngineRequest, InitializeLabelRequest, AllocatePropertyIdsRequest, HeartbeatRequest>;
using CoordinatorWriteResponses = std::variant<HlcResponse, AllocateEdgeIdBatchResponse, SplitShardResponse,
RegisterStorageEngineResponse, DeregisterStorageEngineResponse,
InitializeLabelResponse, AllocatePropertyIdsResponse, HeartbeatResponse>;
using CoordinatorReadRequests = std::variant<GetShardMapRequest, HeartbeatRequest>;
using CoordinatorReadResponses = std::variant<GetShardMapResponse, HeartbeatResponse>;
using CoordinatorReadRequests = std::variant<GetShardMapRequest>;
using CoordinatorReadResponses = std::variant<GetShardMapResponse>;
class Coordinator {
public:
@@ -141,14 +158,10 @@ class Coordinator {
private:
ShardMap shard_map_;
uint64_t highest_allocated_timestamp_;
uint64_t highest_allocated_timestamp_{0};
/// Query engines need to periodically request batches of unique edge IDs.
uint64_t highest_allocated_edge_id_;
static CoordinatorReadResponses HandleRead(HeartbeatRequest && /* heartbeat_request */) {
return HeartbeatResponse{};
}
uint64_t highest_allocated_edge_id_{0};
CoordinatorReadResponses HandleRead(GetShardMapRequest && /* get_shard_map_request */) {
GetShardMapResponse res;
@@ -156,104 +169,28 @@ class Coordinator {
return res;
}
CoordinatorWriteResponses ApplyWrite(HlcRequest &&hlc_request) {
HlcResponse res{};
CoordinatorWriteResponses ApplyWrite(HeartbeatRequest &&heartbeat_request);
auto hlc_shard_map = shard_map_.GetHlc();
CoordinatorWriteResponses ApplyWrite(HlcRequest &&hlc_request);
MG_ASSERT(!(hlc_request.last_shard_map_version.logical_id > hlc_shard_map.logical_id));
res.new_hlc = Hlc{
.logical_id = ++highest_allocated_timestamp_,
// TODO(tyler) probably pass some more context to the Coordinator here
// so that we can use our wall clock and enforce monotonicity.
// .coordinator_wall_clock = io_.Now(),
};
// Allways return fresher shard_map for now.
res.fresher_shard_map = std::make_optional(shard_map_);
return res;
}
CoordinatorWriteResponses ApplyWrite(AllocateEdgeIdBatchRequest &&ahr) {
AllocateEdgeIdBatchResponse res{};
uint64_t low = highest_allocated_edge_id_;
highest_allocated_edge_id_ += ahr.batch_size;
uint64_t high = highest_allocated_edge_id_;
res.low = low;
res.high = high;
return res;
}
CoordinatorWriteResponses ApplyWrite(AllocateEdgeIdBatchRequest &&ahr);
/// This splits the shard immediately beneath the provided
/// split key, keeping the assigned peers identical for now,
/// but letting them be gradually migrated over time.
CoordinatorWriteResponses ApplyWrite(SplitShardRequest &&split_shard_request) {
SplitShardResponse res{};
if (split_shard_request.previous_shard_map_version != shard_map_.shard_map_version) {
res.success = false;
} else {
res.success = shard_map_.SplitShard(split_shard_request.previous_shard_map_version, split_shard_request.label_id,
split_shard_request.split_key);
}
return res;
}
CoordinatorWriteResponses ApplyWrite(SplitShardRequest &&split_shard_request);
/// This adds the provided storage engine to the standby storage engine pool,
/// which can be used to rebalance storage over time.
static CoordinatorWriteResponses ApplyWrite(RegisterStorageEngineRequest && /* register_storage_engine_request */) {
RegisterStorageEngineResponse res{};
// TODO
return res;
}
static CoordinatorWriteResponses ApplyWrite(RegisterStorageEngineRequest && /* register_storage_engine_request */);
/// This begins the process of draining the provided storage engine from all raft
/// clusters that it might be participating in.
static CoordinatorWriteResponses ApplyWrite(DeregisterStorageEngineRequest && /* register_storage_engine_request */) {
DeregisterStorageEngineResponse res{};
// TODO
// const Address &address = register_storage_engine_request.address;
// storage_engine_pool_.erase(address);
// res.success = true;
static CoordinatorWriteResponses ApplyWrite(DeregisterStorageEngineRequest && /* register_storage_engine_request */);
return res;
}
CoordinatorWriteResponses ApplyWrite(InitializeLabelRequest &&initialize_label_request);
CoordinatorWriteResponses ApplyWrite(InitializeLabelRequest &&initialize_label_request) {
InitializeLabelResponse res{};
bool success = shard_map_.InitializeNewLabel(initialize_label_request.label_name, initialize_label_request.schema,
initialize_label_request.last_shard_map_version);
if (success) {
res.fresher_shard_map = shard_map_;
res.success = false;
} else {
res.fresher_shard_map = std::nullopt;
res.success = true;
}
return res;
}
CoordinatorWriteResponses ApplyWrite(AllocatePropertyIdsRequest &&allocate_property_ids_request) {
AllocatePropertyIdsResponse res{};
auto property_ids = shard_map_.AllocatePropertyIds(allocate_property_ids_request.property_names);
res.property_ids = property_ids;
return res;
}
CoordinatorWriteResponses ApplyWrite(AllocatePropertyIdsRequest &&allocate_property_ids_request);
};
} // namespace memgraph::coordinator

View File

@@ -21,5 +21,4 @@ using memgraph::io::rsm::RsmClient;
template <typename IoImpl>
using CoordinatorClient = RsmClient<IoImpl, CoordinatorWriteRequests, CoordinatorWriteResponses,
CoordinatorReadRequests, CoordinatorReadResponses>;
} // namespace memgraph::coordinator

View File

@@ -0,0 +1,156 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#pragma once
#include <chrono>
#include <deque>
#include <memory>
#include <queue>
#include <variant>
#include "coordinator/coordinator.hpp"
#include "coordinator/coordinator_rsm.hpp"
#include "coordinator/shard_map.hpp"
#include "io/address.hpp"
#include "io/future.hpp"
#include "io/messages.hpp"
#include "io/rsm/raft.hpp"
#include "io/time.hpp"
#include "io/transport.hpp"
#include "query/v2/requests.hpp"
namespace memgraph::coordinator::coordinator_worker {
/// Obligations:
/// * ShutDown
/// * Cron
/// * RouteMessage
using coordinator::Coordinator;
using coordinator::CoordinatorRsm;
using io::Address;
using io::RequestId;
using io::Time;
using io::messages::CoordinatorMessages;
using msgs::ReadRequests;
using msgs::ReadResponses;
using msgs::WriteRequests;
using msgs::WriteResponses;
struct ShutDown {};
struct Cron {};
struct RouteMessage {
CoordinatorMessages message;
RequestId request_id;
Address to;
Address from;
};
using Message = std::variant<RouteMessage, Cron, ShutDown>;
struct QueueInner {
std::mutex mu{};
std::condition_variable cv;
// TODO(tyler) handle simulator communication std::shared_ptr<std::atomic<int>> blocked;
// TODO(tyler) investigate using a priority queue that prioritizes messages in a way that
// improves overall QoS. For example, maybe we want to schedule raft Append messages
// ahead of Read messages or generally writes before reads for lowering the load on the
// overall system faster etc... When we do this, we need to make sure to avoid
// starvation by sometimes randomizing priorities, rather than following a strict
// prioritization.
std::deque<Message> queue;
};
/// There are two reasons to implement our own Queue instead of using
/// one off-the-shelf:
/// 1. we will need to know in the simulator when all threads are waiting
/// 2. we will want to implement our own priority queue within this for QoS
class Queue {
std::shared_ptr<QueueInner> inner_ = std::make_shared<QueueInner>();
public:
void Push(Message &&message) {
{
MG_ASSERT(inner_.use_count() > 0);
std::unique_lock<std::mutex> lock(inner_->mu);
inner_->queue.emplace_back(std::move(message));
} // lock dropped before notifying condition variable
inner_->cv.notify_all();
}
Message Pop() {
MG_ASSERT(inner_.use_count() > 0);
std::unique_lock<std::mutex> lock(inner_->mu);
while (inner_->queue.empty()) {
inner_->cv.wait(lock);
}
Message message = std::move(inner_->queue.front());
inner_->queue.pop_front();
return message;
}
};
/// A CoordinatorWorker owns Raft<CoordinatorRsm> instances. receives messages from the MachineManager.
template <typename IoImpl>
class CoordinatorWorker {
io::Io<IoImpl> io_;
Queue queue_;
CoordinatorRsm<IoImpl> coordinator_;
bool Process(ShutDown && /*shut_down*/) { return false; }
bool Process(Cron && /* cron */) {
coordinator_.Cron();
return true;
}
bool Process(RouteMessage &&route_message) {
coordinator_.Handle(std::move(route_message.message), route_message.request_id, route_message.from);
return true;
}
public:
CoordinatorWorker(io::Io<IoImpl> io, Queue queue, Coordinator coordinator)
: io_(std::move(io)),
queue_(std::move(queue)),
coordinator_{std::move(io_.ForkLocal()), {}, std::move(coordinator)} {}
CoordinatorWorker(CoordinatorWorker &&) noexcept = default;
CoordinatorWorker &operator=(CoordinatorWorker &&) noexcept = default;
CoordinatorWorker(const CoordinatorWorker &) = delete;
CoordinatorWorker &operator=(const CoordinatorWorker &) = delete;
~CoordinatorWorker() = default;
void Run() {
while (true) {
Message message = queue_.Pop();
const bool should_continue = std::visit(
[this](auto &&msg) { return this->Process(std::forward<decltype(msg)>(msg)); }, std::move(message));
if (!should_continue) {
return;
}
}
}
};
} // namespace memgraph::coordinator::coordinator_worker

View File

@@ -11,6 +11,11 @@
#pragma once
#include <chrono>
#include <compare>
#include <ctime>
#include <iomanip>
#include "io/time.hpp"
namespace memgraph::coordinator {
@@ -19,10 +24,25 @@ using Time = memgraph::io::Time;
/// Hybrid-logical clock
struct Hlc {
uint64_t logical_id;
Time coordinator_wall_clock;
uint64_t logical_id = 0;
Time coordinator_wall_clock = Time::min();
auto operator<=>(const Hlc &other) const { return logical_id <=> other.logical_id; }
bool operator==(const Hlc &other) const = default;
bool operator<(const Hlc &other) const = default;
bool operator==(const uint64_t other) const { return logical_id == other; }
bool operator<(const uint64_t other) const { return logical_id < other; }
bool operator>=(const uint64_t other) const { return logical_id >= other; }
friend std::ostream &operator<<(std::ostream &in, const Hlc &hlc) {
auto wall_clock = std::chrono::system_clock::to_time_t(hlc.coordinator_wall_clock);
in << "Hlc { logical_id: " << hlc.logical_id;
in << ", coordinator_wall_clock: " << std::put_time(std::localtime(&wall_clock), "%F %T");
in << " }";
return in;
}
};
} // namespace memgraph::coordinator

View File

@@ -0,0 +1,552 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#include <optional>
#include <unordered_map>
#include <vector>
#include "common/types.hpp"
#include "coordinator/shard_map.hpp"
#include "spdlog/spdlog.h"
#include "storage/v3/schemas.hpp"
#include "storage/v3/temporal.hpp"
#include "utils/cast.hpp"
#include "utils/exceptions.hpp"
#include "utils/string.hpp"
namespace memgraph::coordinator {
using memgraph::common::SchemaType;
using memgraph::storage::v3::TemporalData;
using memgraph::storage::v3::TemporalType;
PrimaryKey SchemaToMinKey(const std::vector<SchemaProperty> &schema) {
PrimaryKey ret{};
const int64_t min_int = std::numeric_limits<int64_t>::min();
const TemporalData date{TemporalType::Date, min_int};
const TemporalData local_time{TemporalType::LocalTime, min_int};
const TemporalData local_date_time{TemporalType::LocalDateTime, min_int};
const TemporalData duration{TemporalType::Duration, min_int};
for (const auto &schema_property : schema) {
switch (schema_property.type) {
case SchemaType::BOOL:
ret.emplace_back(PropertyValue(false));
break;
case SchemaType::INT:
ret.emplace_back(PropertyValue(min_int));
break;
case SchemaType::STRING:
ret.emplace_back(PropertyValue(""));
break;
case SchemaType::DATE:
ret.emplace_back(PropertyValue(date));
break;
case SchemaType::LOCALTIME:
ret.emplace_back(PropertyValue(local_time));
break;
case SchemaType::LOCALDATETIME:
ret.emplace_back(PropertyValue(local_date_time));
break;
case SchemaType::DURATION:
ret.emplace_back(PropertyValue(duration));
break;
}
}
return ret;
}
ShardMap ShardMap::Parse(std::istream &input_stream) {
ShardMap shard_map;
const auto read_size = [&input_stream] {
size_t size{0};
input_stream >> size;
return size;
};
// Reads a string until the next whitespace
const auto read_word = [&input_stream] {
std::string word;
input_stream >> word;
return word;
};
const auto read_names = [&read_size, &read_word] {
const auto number_of_names = read_size();
spdlog::trace("Reading {} names", number_of_names);
std::vector<std::string> names;
names.reserve(number_of_names);
for (auto name_index = 0; name_index < number_of_names; ++name_index) {
names.push_back(read_word());
spdlog::trace("Read '{}'", names.back());
}
return names;
};
const auto read_line = [&input_stream] {
std::string line;
std::getline(input_stream, line);
return line;
};
const auto parse_type = [](const std::string &type) {
static const auto type_map = std::unordered_map<std::string, common::SchemaType>{
{"string", common::SchemaType::STRING}, {"int", common::SchemaType::INT}, {"bool", common::SchemaType::BOOL}};
const auto lower_case_type = utils::ToLowerCase(type);
auto it = type_map.find(lower_case_type);
MG_ASSERT(it != type_map.end(), "Invalid type in split files: {}", type);
return it->second;
};
const auto parse_property_value = [](std::string text, const common::SchemaType type) {
if (type == common::SchemaType::STRING) {
return storage::v3::PropertyValue{std::move(text)};
}
if (type == common::SchemaType::INT) {
size_t processed{0};
int64_t value = std::stoll(text, &processed);
MG_ASSERT(processed == text.size() || text[processed] == ' ', "Invalid integer format: '{}'", text);
return storage::v3::PropertyValue{value};
}
LOG_FATAL("Not supported type: {}", utils::UnderlyingCast(type));
};
spdlog::debug("Reading properties");
const auto properties = read_names();
MG_ASSERT(shard_map.AllocatePropertyIds(properties).size() == properties.size(),
"Unexpected number of properties created!");
spdlog::debug("Reading edge types");
const auto edge_types = read_names();
MG_ASSERT(shard_map.AllocateEdgeTypeIds(edge_types).size() == edge_types.size(),
"Unexpected number of properties created!");
spdlog::debug("Reading primary labels");
const auto number_of_primary_labels = read_size();
spdlog::debug("Reading {} primary labels", number_of_primary_labels);
for (auto label_index = 0; label_index < number_of_primary_labels; ++label_index) {
const auto primary_label = read_word();
spdlog::debug("Reading primary label named '{}'", primary_label);
const auto number_of_primary_properties = read_size();
spdlog::debug("Reading {} primary properties", number_of_primary_properties);
std::vector<std::string> pp_names;
std::vector<common::SchemaType> pp_types;
pp_names.reserve(number_of_primary_properties);
pp_types.reserve(number_of_primary_properties);
for (auto property_index = 0; property_index < number_of_primary_properties; ++property_index) {
pp_names.push_back(read_word());
spdlog::debug("Reading primary property named '{}'", pp_names.back());
pp_types.push_back(parse_type(read_word()));
}
auto pp_mapping = shard_map.AllocatePropertyIds(pp_names);
std::vector<SchemaProperty> schema;
schema.reserve(number_of_primary_properties);
for (auto property_index = 0; property_index < number_of_primary_properties; ++property_index) {
schema.push_back(storage::v3::SchemaProperty{pp_mapping.at(pp_names[property_index]), pp_types[property_index]});
}
const auto hlc = shard_map.GetHlc();
MG_ASSERT(shard_map.InitializeNewLabel(primary_label, schema, 1, hlc).has_value(),
"Cannot initialize new label: {}", primary_label);
const auto number_of_split_points = read_size();
spdlog::debug("Reading {} split points", number_of_split_points);
[[maybe_unused]] const auto remainder_from_last_line = read_line();
for (auto split_point_index = 0; split_point_index < number_of_split_points; ++split_point_index) {
const auto line = read_line();
spdlog::debug("Read split point '{}'", line);
MG_ASSERT(line.front() == '[', "Invalid split file format!");
MG_ASSERT(line.back() == ']', "Invalid split file format!");
std::string_view line_view{line};
line_view.remove_prefix(1);
line_view.remove_suffix(1);
static constexpr std::string_view kDelimiter{","};
auto pk_values_as_text = utils::Split(line_view, kDelimiter);
std::vector<PropertyValue> pk;
pk.reserve(number_of_primary_properties);
MG_ASSERT(pk_values_as_text.size() == number_of_primary_properties,
"Split point contains invalid number of values '{}'", line);
for (auto property_index = 0; property_index < number_of_primary_properties; ++property_index) {
pk.push_back(parse_property_value(std::move(pk_values_as_text[property_index]), schema[property_index].type));
}
shard_map.SplitShard(shard_map.GetHlc(), shard_map.labels.at(primary_label), pk);
}
}
return shard_map;
}
std::ostream &operator<<(std::ostream &in, const ShardMap &shard_map) {
using utils::print_helpers::operator<<;
in << "ShardMap { shard_map_version: " << shard_map.shard_map_version;
in << ", max_property_id: " << shard_map.max_property_id;
in << ", max_edge_type_id: " << shard_map.max_edge_type_id;
in << ", properties: " << shard_map.properties;
in << ", edge_types: " << shard_map.edge_types;
in << ", max_label_id: " << shard_map.max_label_id;
in << ", labels: " << shard_map.labels;
in << ", label_spaces: " << shard_map.label_spaces;
in << ", schemas: " << shard_map.schemas;
in << "}";
return in;
}
Shards ShardMap::GetShardsForLabel(const LabelName &label) const {
const auto id = labels.at(label);
const auto &shards = label_spaces.at(id).shards;
return shards;
}
std::vector<Shards> ShardMap::GetAllShards() const {
std::vector<Shards> all_shards;
all_shards.reserve(label_spaces.size());
std::transform(label_spaces.begin(), label_spaces.end(), std::back_inserter(all_shards),
[](const auto &label_space) { return label_space.second.shards; });
return all_shards;
}
// TODO(gabor) later we will want to update the wallclock time with
// the given Io<impl>'s time as well
Hlc ShardMap::IncrementShardMapVersion() noexcept {
++shard_map_version.logical_id;
return shard_map_version;
}
// TODO(antaljanosbenjamin) use a single map for all name id
// mapping and a single counter to maintain the next id
std::unordered_map<uint64_t, std::string> ShardMap::IdToNames() {
std::unordered_map<uint64_t, std::string> id_to_names;
const auto map_type_ids = [&id_to_names](const auto &name_to_id_type) {
for (const auto &[name, id] : name_to_id_type) {
id_to_names.emplace(id.AsUint(), name);
}
};
map_type_ids(edge_types);
map_type_ids(labels);
map_type_ids(properties);
return id_to_names;
}
Hlc ShardMap::GetHlc() const noexcept { return shard_map_version; }
std::vector<ShardToInitialize> ShardMap::AssignShards(Address storage_manager,
std::set<boost::uuids::uuid> initialized) {
std::vector<ShardToInitialize> ret{};
bool mutated = false;
for (auto &[label_id, label_space] : label_spaces) {
for (auto it = label_space.shards.begin(); it != label_space.shards.end(); it++) {
auto &[low_key, shard] = *it;
std::optional<PrimaryKey> high_key;
if (const auto next_it = std::next(it); next_it != label_space.shards.end()) {
high_key = next_it->first;
}
// TODO(tyler) avoid these triple-nested loops by having the heartbeat include better info
bool machine_contains_shard = false;
for (auto &aas : shard) {
if (initialized.contains(aas.address.unique_id)) {
machine_contains_shard = true;
if (aas.status != Status::CONSENSUS_PARTICIPANT) {
spdlog::info("marking shard as full consensus participant: {}", aas.address.unique_id);
aas.status = Status::CONSENSUS_PARTICIPANT;
}
} else {
const bool same_machine = aas.address.last_known_ip == storage_manager.last_known_ip &&
aas.address.last_known_port == storage_manager.last_known_port;
if (same_machine) {
machine_contains_shard = true;
spdlog::info("reminding shard manager that they should begin participating in shard");
ret.push_back(ShardToInitialize{
.uuid = aas.address.unique_id,
.label_id = label_id,
.min_key = low_key,
.max_key = high_key,
.schema = schemas[label_id],
.config = Config{},
.id_to_names = IdToNames(),
});
}
}
}
if (!machine_contains_shard && shard.size() < label_space.replication_factor) {
Address address = storage_manager;
// TODO(tyler) use deterministic UUID so that coordinators don't diverge here
address.unique_id = boost::uuids::uuid{boost::uuids::random_generator()()},
spdlog::info("assigning shard manager to shard");
ret.push_back(ShardToInitialize{
.uuid = address.unique_id,
.label_id = label_id,
.min_key = low_key,
.max_key = high_key,
.schema = schemas[label_id],
.config = Config{},
.id_to_names = IdToNames(),
});
AddressAndStatus aas = {
.address = address,
.status = Status::INITIALIZING,
};
shard.emplace_back(aas);
}
}
}
if (mutated) {
IncrementShardMapVersion();
}
return ret;
}
bool ShardMap::SplitShard(Hlc previous_shard_map_version, LabelId label_id, const PrimaryKey &key) {
if (previous_shard_map_version != shard_map_version) {
return false;
}
auto &label_space = label_spaces.at(label_id);
auto &shards_in_map = label_space.shards;
MG_ASSERT(!shards_in_map.empty());
MG_ASSERT(!shards_in_map.contains(key));
MG_ASSERT(label_spaces.contains(label_id));
// Finding the Shard that the new PrimaryKey should map to.
auto prev = std::prev(shards_in_map.upper_bound(key));
Shard duplicated_shard = prev->second;
// Apply the split
shards_in_map[key] = duplicated_shard;
IncrementShardMapVersion();
return true;
}
std::optional<LabelId> ShardMap::InitializeNewLabel(std::string label_name, std::vector<SchemaProperty> schema,
size_t replication_factor, Hlc last_shard_map_version) {
if (shard_map_version != last_shard_map_version || labels.contains(label_name)) {
return std::nullopt;
}
const LabelId label_id = LabelId::FromUint(++max_label_id);
labels.emplace(std::move(label_name), label_id);
PrimaryKey initial_key = SchemaToMinKey(schema);
Shard empty_shard = {};
Shards shards = {
{initial_key, empty_shard},
};
LabelSpace label_space{
.schema = schema,
.shards = shards,
.replication_factor = replication_factor,
};
schemas[label_id] = std::move(schema);
label_spaces.emplace(label_id, label_space);
IncrementShardMapVersion();
return label_id;
}
void ShardMap::AddServer(Address server_address) {
// Find a random place for the server to plug in
}
std::optional<LabelId> ShardMap::GetLabelId(const std::string &label) const {
if (const auto it = labels.find(label); it != labels.end()) {
return it->second;
}
return std::nullopt;
}
const std::string &ShardMap::GetLabelName(const LabelId label) const {
if (const auto it =
std::ranges::find_if(labels, [label](const auto &name_id_pair) { return name_id_pair.second == label; });
it != labels.end()) {
return it->first;
}
throw utils::BasicException("GetLabelName fails on the given label id!");
}
std::optional<PropertyId> ShardMap::GetPropertyId(const std::string &property_name) const {
if (const auto it = properties.find(property_name); it != properties.end()) {
return it->second;
}
return std::nullopt;
}
const std::string &ShardMap::GetPropertyName(const PropertyId property) const {
if (const auto it = std::ranges::find_if(
properties, [property](const auto &name_id_pair) { return name_id_pair.second == property; });
it != properties.end()) {
return it->first;
}
throw utils::BasicException("PropertyId not found!");
}
std::optional<EdgeTypeId> ShardMap::GetEdgeTypeId(const std::string &edge_type) const {
if (const auto it = edge_types.find(edge_type); it != edge_types.end()) {
return it->second;
}
return std::nullopt;
}
const std::string &ShardMap::GetEdgeTypeName(const EdgeTypeId property) const {
if (const auto it = std::ranges::find_if(
edge_types, [property](const auto &name_id_pair) { return name_id_pair.second == property; });
it != edge_types.end()) {
return it->first;
}
throw utils::BasicException("EdgeTypeId not found!");
}
Shards ShardMap::GetShardsForRange(const LabelName &label_name, const PrimaryKey &start_key,
const PrimaryKey &end_key) const {
MG_ASSERT(start_key <= end_key);
MG_ASSERT(labels.contains(label_name));
LabelId label_id = labels.at(label_name);
const auto &label_space = label_spaces.at(label_id);
const auto &shards_for_label = label_space.shards;
MG_ASSERT(shards_for_label.begin()->first <= start_key,
"the ShardMap must always contain a minimal key that is less than or equal to any requested key");
auto it = std::prev(shards_for_label.upper_bound(start_key));
const auto end_it = shards_for_label.upper_bound(end_key);
Shards shards{};
std::copy(it, end_it, std::inserter(shards, shards.end()));
return shards;
}
const Shard &ShardMap::GetShardForKey(const LabelName &label_name, const PrimaryKey &key) const {
MG_ASSERT(labels.contains(label_name));
LabelId label_id = labels.at(label_name);
const auto &label_space = label_spaces.at(label_id);
MG_ASSERT(label_space.shards.begin()->first <= key,
"the ShardMap must always contain a minimal key that is less than or equal to any requested key");
return std::prev(label_space.shards.upper_bound(key))->second;
}
const Shard &ShardMap::GetShardForKey(const LabelId &label_id, const PrimaryKey &key) const {
MG_ASSERT(label_spaces.contains(label_id));
const auto &label_space = label_spaces.at(label_id);
MG_ASSERT(label_space.shards.begin()->first <= key,
"the ShardMap must always contain a minimal key that is less than or equal to any requested key");
return std::prev(label_space.shards.upper_bound(key))->second;
}
PropertyMap ShardMap::AllocatePropertyIds(const std::vector<PropertyName> &new_properties) {
PropertyMap ret{};
bool mutated = false;
for (const auto &property_name : new_properties) {
if (properties.contains(property_name)) {
auto property_id = properties.at(property_name);
ret.emplace(property_name, property_id);
} else {
mutated = true;
const PropertyId property_id = PropertyId::FromUint(++max_property_id);
ret.emplace(property_name, property_id);
properties.emplace(property_name, property_id);
}
}
if (mutated) {
IncrementShardMapVersion();
}
return ret;
}
EdgeTypeIdMap ShardMap::AllocateEdgeTypeIds(const std::vector<EdgeTypeName> &new_edge_types) {
EdgeTypeIdMap ret;
bool mutated = false;
for (const auto &edge_type_name : new_edge_types) {
if (edge_types.contains(edge_type_name)) {
auto edge_type_id = edge_types.at(edge_type_name);
ret.emplace(edge_type_name, edge_type_id);
} else {
mutated = true;
const EdgeTypeId edge_type_id = EdgeTypeId::FromUint(++max_edge_type_id);
ret.emplace(edge_type_name, edge_type_id);
edge_types.emplace(edge_type_name, edge_type_id);
}
}
if (mutated) {
IncrementShardMapVersion();
}
return ret;
}
bool ShardMap::ClusterInitialized() const {
for (const auto &[label_id, label_space] : label_spaces) {
for (const auto &[low_key, shard] : label_space.shards) {
if (shard.size() < label_space.replication_factor) {
spdlog::info("label_space below desired replication factor");
return false;
}
for (const auto &aas : shard) {
if (aas.status != Status::CONSENSUS_PARTICIPANT) {
spdlog::info("shard member not yet a CONSENSUS_PARTICIPANT");
return false;
}
}
}
}
return true;
}
} // namespace memgraph::coordinator

View File

@@ -11,20 +11,37 @@
#pragma once
#include <algorithm>
#include <limits>
#include <map>
#include <set>
#include <vector>
#include <boost/uuid/uuid.hpp>
#include <boost/uuid/uuid_generators.hpp>
#include "common/types.hpp"
#include "coordinator/hybrid_logical_clock.hpp"
#include "io/address.hpp"
#include "storage/v3/config.hpp"
#include "storage/v3/id_types.hpp"
#include "storage/v3/name_id_mapper.hpp"
#include "storage/v3/property_value.hpp"
#include "storage/v3/schemas.hpp"
#include "storage/v3/temporal.hpp"
#include "utils/exceptions.hpp"
#include "utils/print_helpers.hpp"
namespace memgraph::coordinator {
constexpr int64_t kNotExistingId{0};
using memgraph::io::Address;
using memgraph::storage::v3::Config;
using memgraph::storage::v3::EdgeTypeId;
using memgraph::storage::v3::LabelId;
using memgraph::storage::v3::PropertyId;
using memgraph::storage::v3::PropertyValue;
using memgraph::storage::v3::SchemaProperty;
enum class Status : uint8_t {
@@ -38,152 +55,123 @@ enum class Status : uint8_t {
struct AddressAndStatus {
memgraph::io::Address address;
Status status;
friend bool operator<(const AddressAndStatus &lhs, const AddressAndStatus &rhs) { return lhs.address < rhs.address; }
friend std::ostream &operator<<(std::ostream &in, const AddressAndStatus &address_and_status) {
in << "AddressAndStatus { address: ";
in << address_and_status.address;
if (address_and_status.status == Status::CONSENSUS_PARTICIPANT) {
in << ", status: CONSENSUS_PARTICIPANT }";
} else {
in << ", status: INITIALIZING }";
}
return in;
}
friend bool operator==(const AddressAndStatus &lhs, const AddressAndStatus &rhs) {
return lhs.address == rhs.address;
}
};
using CompoundKey = std::vector<memgraph::storage::v3::PropertyValue>;
using PrimaryKey = std::vector<PropertyValue>;
using Shard = std::vector<AddressAndStatus>;
using Shards = std::map<CompoundKey, Shard>;
using Shards = std::map<PrimaryKey, Shard>;
using LabelName = std::string;
using PropertyName = std::string;
using EdgeTypeName = std::string;
using PropertyMap = std::map<PropertyName, PropertyId>;
using EdgeTypeIdMap = std::map<EdgeTypeName, EdgeTypeId>;
struct ShardToInitialize {
boost::uuids::uuid uuid;
LabelId label_id;
PrimaryKey min_key;
std::optional<PrimaryKey> max_key;
std::vector<SchemaProperty> schema;
Config config;
std::unordered_map<uint64_t, std::string> id_to_names;
};
PrimaryKey SchemaToMinKey(const std::vector<SchemaProperty> &schema);
struct LabelSpace {
std::vector<SchemaProperty> schema;
std::map<CompoundKey, Shard> shards;
// Maps between the smallest primary key stored in the shard and the shard
std::map<PrimaryKey, Shard> shards;
size_t replication_factor;
friend std::ostream &operator<<(std::ostream &in, const LabelSpace &label_space) {
using utils::print_helpers::operator<<;
in << "LabelSpace { schema: ";
in << label_space.schema;
in << ", shards: ";
in << label_space.shards;
in << ", replication_factor: " << label_space.replication_factor << "}";
return in;
}
};
struct ShardMap {
Hlc shard_map_version;
uint64_t max_property_id;
uint64_t max_property_id{kNotExistingId};
uint64_t max_edge_type_id{kNotExistingId};
std::map<PropertyName, PropertyId> properties;
uint64_t max_label_id;
std::map<LabelName, LabelId> labels;
std::map<EdgeTypeName, EdgeTypeId> edge_types;
uint64_t max_label_id{kNotExistingId};
std::map<LabelId, LabelSpace> label_spaces;
std::map<LabelId, std::vector<SchemaProperty>> schemas;
std::map<LabelName, LabelId> labels;
[[nodiscard]] static ShardMap Parse(std::istream &input_stream);
friend std::ostream &operator<<(std::ostream &in, const ShardMap &shard_map);
Shards GetShardsForLabel(const LabelName &label) const;
std::vector<Shards> GetAllShards() const;
// TODO(gabor) later we will want to update the wallclock time with
// the given Io<impl>'s time as well
Hlc IncrementShardMapVersion() noexcept {
++shard_map_version.logical_id;
return shard_map_version;
}
Hlc IncrementShardMapVersion() noexcept;
Hlc GetHlc() const noexcept;
Hlc GetHlc() const noexcept { return shard_map_version; }
std::unordered_map<uint64_t, std::string> IdToNames();
bool SplitShard(Hlc previous_shard_map_version, LabelId label_id, const CompoundKey &key) {
if (previous_shard_map_version != shard_map_version) {
return false;
}
// Returns the shard UUIDs that have been assigned but not yet acknowledged for this storage manager
std::vector<ShardToInitialize> AssignShards(Address storage_manager, std::set<boost::uuids::uuid> initialized);
auto &label_space = label_spaces.at(label_id);
auto &shards_in_map = label_space.shards;
bool SplitShard(Hlc previous_shard_map_version, LabelId label_id, const PrimaryKey &key);
MG_ASSERT(!shards_in_map.contains(key));
MG_ASSERT(label_spaces.contains(label_id));
std::optional<LabelId> InitializeNewLabel(std::string label_name, std::vector<SchemaProperty> schema,
size_t replication_factor, Hlc last_shard_map_version);
// Finding the Shard that the new CompoundKey should map to.
auto prev = std::prev(shards_in_map.upper_bound(key));
Shard duplicated_shard = prev->second;
void AddServer(Address server_address);
// Apply the split
shards_in_map[key] = duplicated_shard;
std::optional<LabelId> GetLabelId(const std::string &label) const;
// TODO(antaljanosbenjamin): Remove this and instead use NameIdMapper
const std::string &GetLabelName(LabelId label) const;
std::optional<PropertyId> GetPropertyId(const std::string &property_name) const;
const std::string &GetPropertyName(PropertyId property) const;
std::optional<EdgeTypeId> GetEdgeTypeId(const std::string &edge_type) const;
const std::string &GetEdgeTypeName(EdgeTypeId property) const;
return true;
}
Shards GetShardsForRange(const LabelName &label_name, const PrimaryKey &start_key, const PrimaryKey &end_key) const;
bool InitializeNewLabel(std::string label_name, std::vector<SchemaProperty> schema, Hlc last_shard_map_version) {
if (shard_map_version != last_shard_map_version || labels.contains(label_name)) {
return false;
}
const Shard &GetShardForKey(const LabelName &label_name, const PrimaryKey &key) const;
const LabelId label_id = LabelId::FromUint(++max_label_id);
const Shard &GetShardForKey(const LabelId &label_id, const PrimaryKey &key) const;
labels.emplace(std::move(label_name), label_id);
PropertyMap AllocatePropertyIds(const std::vector<PropertyName> &new_properties);
LabelSpace label_space{
.schema = std::move(schema),
.shards = Shards{},
};
EdgeTypeIdMap AllocateEdgeTypeIds(const std::vector<EdgeTypeName> &new_edge_types);
label_spaces.emplace(label_id, label_space);
IncrementShardMapVersion();
return true;
}
void AddServer(Address server_address) {
// Find a random place for the server to plug in
}
Shards GetShardsForRange(const LabelName &label_name, const CompoundKey &start_key, const CompoundKey &end_key) const {
MG_ASSERT(start_key <= end_key);
MG_ASSERT(labels.contains(label_name));
LabelId label_id = labels.at(label_name);
const auto &label_space = label_spaces.at(label_id);
const auto &shards_for_label = label_space.shards;
MG_ASSERT(shards_for_label.begin()->first <= start_key,
"the ShardMap must always contain a minimal key that is less than or equal to any requested key");
auto it = std::prev(shards_for_label.upper_bound(start_key));
const auto end_it = shards_for_label.upper_bound(end_key);
Shards shards{};
std::copy(it, end_it, std::inserter(shards, shards.end()));
return shards;
}
Shard GetShardForKey(const LabelName &label_name, const CompoundKey &key) const {
MG_ASSERT(labels.contains(label_name));
LabelId label_id = labels.at(label_name);
const auto &label_space = label_spaces.at(label_id);
MG_ASSERT(label_space.shards.begin()->first <= key,
"the ShardMap must always contain a minimal key that is less than or equal to any requested key");
return std::prev(label_space.shards.upper_bound(key))->second;
}
PropertyMap AllocatePropertyIds(const std::vector<PropertyName> &new_properties) {
PropertyMap ret{};
bool mutated = false;
for (const auto &property_name : new_properties) {
if (properties.contains(property_name)) {
auto property_id = properties.at(property_name);
ret.emplace(property_name, property_id);
} else {
mutated = true;
const PropertyId property_id = PropertyId::FromUint(++max_property_id);
ret.emplace(property_name, property_id);
properties.emplace(property_name, property_id);
}
}
if (mutated) {
IncrementShardMapVersion();
}
return ret;
}
std::optional<PropertyId> GetPropertyId(const std::string &property_name) const {
if (properties.contains(property_name)) {
return properties.at(property_name);
}
return std::nullopt;
}
/// Returns true if all shards have the desired number of replicas and they are in
/// the CONSENSUS_PARTICIPANT state. Note that this does not necessarily mean that
/// there is also an active leader for each shard.
bool ClusterInitialized() const;
};
} // namespace memgraph::coordinator

View File

@@ -13,7 +13,7 @@
#ifndef MG_AST_INCLUDE_PATH
#ifdef MG_CLANG_TIDY_CHECK
// NOLINTNEXTLINE(cppcoreguidelines-macro-usage)
#define MG_AST_INCLUDE_PATH "query/v2/frontend/ast/ast.hpp"
#include "query/v2/bindings/bindings.hpp"
#else
#error Missing AST include path
#endif
@@ -21,8 +21,6 @@
#ifndef MG_INJECTED_NAMESPACE_NAME
#ifdef MG_CLANG_TIDY_CHECK
// NOLINTNEXTLINE(cppcoreguidelines-macro-usage)
#define MG_INJECTED_NAMESPACE_NAME memgraph::query::v2
#else
#error Missing AST namespace
#endif

View File

@@ -0,0 +1,279 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#pragma once
#include <iostream>
#include <type_traits>
#include "expr/ast.hpp"
#include "expr/typed_value.hpp"
#include "utils/algorithm.hpp"
#include "utils/logging.hpp"
#include "utils/string.hpp"
namespace memgraph::expr {
inline constexpr const char *identifier_node_symbol = "MG_SYMBOL_NODE";
inline constexpr const char *identifier_edge_symbol = "MG_SYMBOL_EDGE";
namespace detail {
template <typename T>
void PrintObject(std::ostream *out, const T &arg) {
static_assert(!std::is_convertible<T, Expression *>::value,
"This overload shouldn't be called with pointers convertible "
"to Expression *. This means your other PrintObject overloads aren't "
"being called for certain AST nodes when they should (or perhaps such "
"overloads don't exist yet).");
*out << arg;
}
inline void PrintObject(std::ostream *out, const std::string &str) { *out << str; }
inline void PrintObject(std::ostream * /*out*/, Aggregation::Op /*op*/) {
throw utils::NotYetImplemented("PrintObject: Aggregation::Op");
}
inline void PrintObject(std::ostream * /*out*/, Expression * /*expr*/);
inline void PrintObject(std::ostream *out, Identifier *expr) { PrintObject(out, static_cast<Expression *>(expr)); }
template <typename T>
void PrintObject(std::ostream * /*out*/, const std::vector<T> & /*vec*/) {
throw utils::NotYetImplemented("PrintObject: vector<T>");
}
template <typename T>
void PrintObject(std::ostream * /*out*/, const std::vector<T, utils::Allocator<T>> & /*vec*/) {
throw utils::NotYetImplemented("PrintObject: vector<T, utils::Allocator<T>>");
}
template <typename K, typename V>
void PrintObject(std::ostream * /*out*/, const std::map<K, V> & /*map*/) {
throw utils::NotYetImplemented("PrintObject: map<K, V>");
}
template <typename T>
void PrintObject(std::ostream * /*out*/, const utils::pmr::map<utils::pmr::string, T> & /*map*/) {
throw utils::NotYetImplemented("PrintObject: map<utils::pmr::string, V>");
}
template <typename T1, typename T2, typename T3>
inline void PrintObject(std::ostream *out, const TypedValueT<T1, T2, T3> &value) {
using TypedValue = TypedValueT<T1, T2, T3>;
switch (value.type()) {
case TypedValue::Type::Null:
*out << "null";
break;
case TypedValue::Type::String:
PrintObject(out, value.ValueString());
break;
case TypedValue::Type::Bool:
*out << (value.ValueBool() ? "true" : "false");
break;
case TypedValue::Type::Int:
PrintObject(out, value.ValueInt());
break;
case TypedValue::Type::Double:
PrintObject(out, value.ValueDouble());
break;
case TypedValue::Type::List:
PrintObject(out, value.ValueList());
break;
case TypedValue::Type::Map:
PrintObject(out, value.ValueMap());
break;
case TypedValue::Type::Date:
PrintObject(out, value.ValueDate());
break;
case TypedValue::Type::Duration:
PrintObject(out, value.ValueDuration());
break;
case TypedValue::Type::LocalTime:
PrintObject(out, value.ValueLocalTime());
break;
case TypedValue::Type::LocalDateTime:
PrintObject(out, value.ValueLocalDateTime());
break;
default:
MG_ASSERT(false, "PrintObject(std::ostream *out, const TypedValue &value) should not reach here");
}
}
template <typename T>
void PrintOperatorArgs(const std::string & /*name*/, std::ostream *out, bool with_parenthesis, const T &arg) {
PrintObject(out, arg);
if (with_parenthesis) {
*out << ")";
}
}
template <typename T, typename... Ts>
void PrintOperatorArgs(const std::string &name, std::ostream *out, bool with_parenthesis, const T &arg,
const Ts &...args) {
PrintObject(out, arg);
*out << " " << name << " ";
PrintOperatorArgs(name, out, with_parenthesis, args...);
}
template <typename... Ts>
void PrintOperator(const std::string &name, std::ostream *out, bool with_parenthesis, const Ts &...args) {
if (with_parenthesis) {
*out << "(";
}
PrintOperatorArgs(name, out, with_parenthesis, args...);
}
// new
template <typename T>
void PrintOperatorArgs(std::ostream *out, const T &arg) {
PrintObject(out, arg);
}
template <typename T, typename... Ts>
void PrintOperatorArgs(std::ostream *out, const T &arg, const Ts &...args) {
PrintObject(out, arg);
PrintOperatorArgs(out, args...);
}
template <typename... Ts>
void PrintOperator(std::ostream *out, const Ts &...args) {
PrintOperatorArgs(out, args...);
}
} // namespace detail
class ExpressionPrettyPrinter : public ExpressionVisitor<void> {
public:
explicit ExpressionPrettyPrinter(std::ostream *out) : out_(out) {}
// Unary operators
// NOLINTNEXTLINE(cppcoreguidelines-macro-usage)
#define UNARY_OPERATOR_VISIT(OP_NODE, OP_STR) \
/* NOLINTNEXTLINE(bugprone-macro-parentheses) */ \
void Visit(OP_NODE &op) override { detail::PrintOperator(OP_STR, out_, false /*with_parenthesis*/, op.expression_); }
UNARY_OPERATOR_VISIT(NotOperator, "Not");
UNARY_OPERATOR_VISIT(UnaryPlusOperator, "+");
UNARY_OPERATOR_VISIT(UnaryMinusOperator, "-");
UNARY_OPERATOR_VISIT(IsNullOperator, "IsNull");
#undef UNARY_OPERATOR_VISIT
// Binary operators
// NOLINTNEXTLINE(cppcoreguidelines-macro-usage)
#define BINARY_OPERATOR_VISIT(OP_NODE, OP_STR) \
/* NOLINTNEXTLINE(bugprone-macro-parentheses) */ \
void Visit(OP_NODE &op) override { \
detail::PrintOperator(OP_STR, out_, true /*with_parenthesis*/, op.expression1_, op.expression2_); \
}
// NOLINTNEXTLINE(cppcoreguidelines-macro-usage)
#define BINARY_OPERATOR_VISIT_NOT_IMPL(OP_NODE, OP_STR) \
/* NOLINTNEXTLINE(bugprone-macro-parentheses) */ \
void Visit(OP_NODE & /*op*/) override { throw utils::NotYetImplemented("OP_NODE"); }
BINARY_OPERATOR_VISIT(OrOperator, "Or");
BINARY_OPERATOR_VISIT(XorOperator, "Xor");
BINARY_OPERATOR_VISIT(AndOperator, "And");
BINARY_OPERATOR_VISIT(AdditionOperator, "+");
BINARY_OPERATOR_VISIT(SubtractionOperator, "-");
BINARY_OPERATOR_VISIT(MultiplicationOperator, "*");
BINARY_OPERATOR_VISIT(DivisionOperator, "/");
BINARY_OPERATOR_VISIT(ModOperator, "%");
BINARY_OPERATOR_VISIT(NotEqualOperator, "!=");
BINARY_OPERATOR_VISIT(EqualOperator, "=");
BINARY_OPERATOR_VISIT(LessOperator, "<");
BINARY_OPERATOR_VISIT(GreaterOperator, ">");
BINARY_OPERATOR_VISIT(LessEqualOperator, "<=");
BINARY_OPERATOR_VISIT(GreaterEqualOperator, ">=");
BINARY_OPERATOR_VISIT_NOT_IMPL(InListOperator, "In");
BINARY_OPERATOR_VISIT_NOT_IMPL(SubscriptOperator, "Subscript");
#undef BINARY_OPERATOR_VISIT
#undef BINARY_OPERATOR_VISIT_NOT_IMPL
// Other
void Visit(ListSlicingOperator & /*op*/) override { throw utils::NotYetImplemented("ListSlicingOperator"); }
void Visit(IfOperator & /*op*/) override { throw utils::NotYetImplemented("IfOperator"); }
void Visit(ListLiteral & /*op*/) override { throw utils::NotYetImplemented("ListLiteral"); }
void Visit(MapLiteral & /*op*/) override { throw utils::NotYetImplemented("MapLiteral"); }
void Visit(LabelsTest & /*op*/) override { throw utils::NotYetImplemented("LabelsTest"); }
void Visit(Aggregation & /*op*/) override { throw utils::NotYetImplemented("Aggregation"); }
void Visit(Function & /*op*/) override { throw utils::NotYetImplemented("Function"); }
void Visit(Reduce & /*op*/) override { throw utils::NotYetImplemented("Reduce"); }
void Visit(Coalesce & /*op*/) override { throw utils::NotYetImplemented("Coalesce"); }
void Visit(Extract & /*op*/) override { throw utils::NotYetImplemented("Extract"); }
void Visit(All & /*op*/) override { throw utils::NotYetImplemented("All"); }
void Visit(Single & /*op*/) override { throw utils::NotYetImplemented("Single"); }
void Visit(Any & /*op*/) override { throw utils::NotYetImplemented("Any"); }
void Visit(None & /*op*/) override { throw utils::NotYetImplemented("None"); }
void Visit(Identifier &op) override {
auto is_node = true;
auto is_edge = false;
auto is_other = false;
if (is_node) {
detail::PrintOperator(out_, identifier_node_symbol);
} else if (is_edge) {
detail::PrintOperator(out_, identifier_edge_symbol);
} else {
MG_ASSERT(is_other);
detail::PrintOperator(out_, op.name_);
}
}
void Visit(PrimitiveLiteral &op) override { detail::PrintObject(out_, op.value_); }
void Visit(PropertyLookup &op) override { detail::PrintOperator(out_, op.expression_, ".", op.property_.name); }
void Visit(ParameterLookup & /*op*/) override { throw utils::NotYetImplemented("ParameterLookup"); }
void Visit(NamedExpression & /*op*/) override { throw utils::NotYetImplemented("NamedExpression"); }
void Visit(RegexMatch & /*op*/) override { throw utils::NotYetImplemented("RegexMatch"); }
private:
std::ostream *out_;
};
namespace detail {
inline void PrintObject(std::ostream *out, Expression *expr) {
if (expr) {
ExpressionPrettyPrinter printer{out};
expr->Accept(printer);
} else {
*out << "<null>";
}
}
} // namespace detail
inline void PrintExpressionToOriginalAndReplaceNodeAndEdgeSymbols(Expression *expr, std::ostream *out) {
ExpressionPrettyPrinter printer{out};
expr->Accept(printer);
}
inline std::string ExpressiontoStringWhileReplacingNodeAndEdgeSymbols(Expression *expr) {
std::ostringstream ss;
expr::PrintExpressionToOriginalAndReplaceNodeAndEdgeSymbols(expr, &ss);
return ss.str();
}
} // namespace memgraph::expr

View File

@@ -32,4 +32,21 @@ class ExpressionRuntimeException : public utils::BasicException {
using utils::BasicException::BasicException;
};
class RedeclareVariableError : public SemanticException {
public:
explicit RedeclareVariableError(const std::string &name) : SemanticException("Redeclaring variable: " + name + ".") {}
};
class UnboundVariableError : public SemanticException {
public:
explicit UnboundVariableError(const std::string &name) : SemanticException("Unbound variable: " + name + ".") {}
};
class TypeMismatchError : public SemanticException {
public:
TypeMismatchError(const std::string &name, const std::string &datum, const std::string &expected)
: SemanticException(fmt::format("Type mismatch: {} already defined as {}, expected {}.", name, datum, expected)) {
}
};
} // namespace memgraph::expr

View File

@@ -17,6 +17,7 @@
#include <map>
#include <optional>
#include <regex>
#include <type_traits>
#include <vector>
#include "expr/ast.hpp"
@@ -27,8 +28,11 @@
namespace memgraph::expr {
struct StorageTag {};
struct QueryEngineTag {};
template <typename TypedValue, typename EvaluationContext, typename DbAccessor, typename StorageView, typename LabelId,
typename PropertyValue, typename ConvFunction, typename Error>
typename PropertyValue, typename ConvFunctor, typename Error, typename Tag = StorageTag>
class ExpressionEvaluator : public ExpressionVisitor<TypedValue> {
public:
ExpressionEvaluator(Frame<TypedValue> *frame, const SymbolTable &symbol_table, const EvaluationContext &ctx,
@@ -377,6 +381,44 @@ class ExpressionEvaluator : public ExpressionVisitor<TypedValue> {
}
}
template <typename VertexAccessor, typename TTag = Tag,
typename TReturnType = std::enable_if_t<std::is_same_v<TTag, StorageTag>, bool>>
TReturnType HasLabelImpl(const VertexAccessor &vertex, const LabelIx &label, StorageTag /*tag*/) {
auto has_label = vertex.HasLabel(view_, GetLabel(label));
if (has_label.HasError() && has_label.GetError() == Error::NONEXISTENT_OBJECT) {
// This is a very nasty and temporary hack in order to make MERGE
// work. The old storage had the following logic when returning an
// `OLD` view: `return old ? old : new`. That means that if the
// `OLD` view didn't exist, it returned the NEW view. With this hack
// we simulate that behavior.
// TODO (mferencevic, teon.banek): Remove once MERGE is
// reimplemented.
has_label = vertex.HasLabel(StorageView::NEW, GetLabel(label));
}
if (has_label.HasError()) {
switch (has_label.GetError()) {
case Error::DELETED_OBJECT:
throw ExpressionRuntimeException("Trying to access labels on a deleted node.");
case Error::NONEXISTENT_OBJECT:
throw ExpressionRuntimeException("Trying to access labels from a node that doesn't exist.");
case Error::SERIALIZATION_ERROR:
case Error::VERTEX_HAS_EDGES:
case Error::PROPERTIES_DISABLED:
case Error::VERTEX_ALREADY_INSERTED:
throw ExpressionRuntimeException("Unexpected error when accessing labels.");
}
}
return *has_label;
}
template <typename VertexAccessor, typename TTag = Tag,
typename TReturnType = std::enable_if_t<std::is_same_v<TTag, QueryEngineTag>, bool>>
TReturnType HasLabelImpl(const VertexAccessor &vertex, const LabelIx &label_ix, QueryEngineTag /*tag*/) {
auto label = typename VertexAccessor::Label{LabelId::FromUint(label_ix.ix)};
auto has_label = vertex.HasLabel(label);
return !has_label;
}
TypedValue Visit(LabelsTest &labels_test) override {
auto expression_result = labels_test.expression_->Accept(*this);
switch (expression_result.type()) {
@@ -384,35 +426,12 @@ class ExpressionEvaluator : public ExpressionVisitor<TypedValue> {
return TypedValue(ctx_->memory);
case TypedValue::Type::Vertex: {
const auto &vertex = expression_result.ValueVertex();
for (const auto &label : labels_test.labels_) {
auto has_label = vertex.HasLabel(view_, GetLabel(label));
if (has_label.HasError() && has_label.GetError() == Error::NONEXISTENT_OBJECT) {
// This is a very nasty and temporary hack in order to make MERGE
// work. The old storage had the following logic when returning an
// `OLD` view: `return old ? old : new`. That means that if the
// `OLD` view didn't exist, it returned the NEW view. With this hack
// we simulate that behavior.
// TODO (mferencevic, teon.banek): Remove once MERGE is
// reimplemented.
has_label = vertex.HasLabel(StorageView::NEW, GetLabel(label));
}
if (has_label.HasError()) {
switch (has_label.GetError()) {
case Error::DELETED_OBJECT:
throw ExpressionRuntimeException("Trying to access labels on a deleted node.");
case Error::NONEXISTENT_OBJECT:
throw ExpressionRuntimeException("Trying to access labels from a node that doesn't exist.");
case Error::SERIALIZATION_ERROR:
case Error::VERTEX_HAS_EDGES:
case Error::PROPERTIES_DISABLED:
throw ExpressionRuntimeException("Unexpected error when accessing labels.");
}
}
if (!*has_label) {
return TypedValue(false, ctx_->memory);
}
if (std::ranges::all_of(labels_test.labels_, [&vertex, this](const auto label_test) {
return this->HasLabelImpl(vertex, label_test, Tag{});
})) {
return TypedValue(true, ctx_->memory);
}
return TypedValue(true, ctx_->memory);
return TypedValue(false, ctx_->memory);
}
default:
throw ExpressionRuntimeException("Only nodes have labels.");
@@ -695,7 +714,24 @@ class ExpressionEvaluator : public ExpressionVisitor<TypedValue> {
}
private:
template <class TRecordAccessor>
template <class TRecordAccessor, class TTag = Tag,
class TReturnType = std::enable_if_t<std::is_same_v<TTag, QueryEngineTag>, TypedValue>>
TReturnType GetProperty(const TRecordAccessor &record_accessor, PropertyIx prop) {
auto maybe_prop = record_accessor.GetProperty(prop.name);
// Handler non existent property
return conv_(maybe_prop, dba_);
}
template <class TRecordAccessor, class TTag = Tag,
class TReturnType = std::enable_if_t<std::is_same_v<TTag, QueryEngineTag>, TypedValue>>
TReturnType GetProperty(const TRecordAccessor &record_accessor, const std::string_view name) {
auto maybe_prop = record_accessor.GetProperty(std::string(name));
// Handler non existent property
return conv_(maybe_prop, dba_);
}
template <class TRecordAccessor, class TTag = Tag,
class TReturnType = std::enable_if_t<std::is_same_v<TTag, StorageTag>, TypedValue>>
TypedValue GetProperty(const TRecordAccessor &record_accessor, PropertyIx prop) {
auto maybe_prop = record_accessor.GetProperty(view_, ctx_->properties[prop.ix]);
if (maybe_prop.HasError() && maybe_prop.GetError() == Error::NONEXISTENT_OBJECT) {
@@ -716,13 +752,15 @@ class ExpressionEvaluator : public ExpressionVisitor<TypedValue> {
case Error::SERIALIZATION_ERROR:
case Error::VERTEX_HAS_EDGES:
case Error::PROPERTIES_DISABLED:
case Error::VERTEX_ALREADY_INSERTED:
throw ExpressionRuntimeException("Unexpected error when getting a property.");
}
}
return conv_(*maybe_prop);
return conv_(*maybe_prop, ctx_->memory);
}
template <class TRecordAccessor>
template <class TRecordAccessor, class TTag = Tag,
class TReturnType = std::enable_if_t<std::is_same_v<TTag, StorageTag>, TypedValue>>
TypedValue GetProperty(const TRecordAccessor &record_accessor, const std::string_view name) {
auto maybe_prop = record_accessor.GetProperty(view_, dba_->NameToProperty(name));
if (maybe_prop.HasError() && maybe_prop.GetError() == Error::NONEXISTENT_OBJECT) {
@@ -743,10 +781,11 @@ class ExpressionEvaluator : public ExpressionVisitor<TypedValue> {
case Error::SERIALIZATION_ERROR:
case Error::VERTEX_HAS_EDGES:
case Error::PROPERTIES_DISABLED:
case Error::VERTEX_ALREADY_INSERTED:
throw ExpressionRuntimeException("Unexpected error when getting a property.");
}
}
return conv_(*maybe_prop);
return conv_(*maybe_prop, ctx_->memory);
}
LabelId GetLabel(LabelIx label) { return ctx_->labels[label.ix]; }
@@ -757,7 +796,7 @@ class ExpressionEvaluator : public ExpressionVisitor<TypedValue> {
DbAccessor *dba_;
// which switching approach should be used when evaluating
StorageView view_;
ConvFunction conv_;
ConvFunctor conv_;
};
/// A helper function for evaluating an expression that's an int.

View File

@@ -35,6 +35,7 @@ class Frame {
const TypedValue &at(const Symbol &symbol) const { return elems_.at(symbol.position()); }
auto &elems() { return elems_; }
auto &elems() const { return elems_; }
utils::MemoryResource *GetMemoryResource() const { return elems_.get_allocator().GetMemoryResource(); }

View File

@@ -21,6 +21,7 @@
#include <utility>
#include <vector>
#include "expr/typed_value_exception.hpp"
#include "utils/algorithm.hpp"
#include "utils/exceptions.hpp"
#include "utils/fnv.hpp"
@@ -32,16 +33,6 @@
namespace memgraph::expr {
/**
* An exception raised by the TypedValue system. Typically when
* trying to perform operations (such as addition) on TypedValues
* of incompatible Types.
*/
class TypedValueException : public utils::BasicException {
public:
using utils::BasicException::BasicException;
};
// TODO: Neo4j does overflow checking. Should we also implement it?
/**
* Stores a query runtime value and its type.
@@ -116,13 +107,15 @@ class TypedValueT {
return hash;
}
case TypedValueT::Type::Vertex:
return 34;
case TypedValueT::Type::Edge:
return 0;
return 35;
case TypedValueT::Type::Path: {
const auto &vertices = value.ValuePath().vertices();
const auto &edges = value.ValuePath().edges();
return utils::FnvCollection<decltype(vertices), TVertexAccessor>{}(vertices) ^
utils::FnvCollection<decltype(edges), TEdgeAccessor>{}(edges);
// const auto &vertices = value.ValuePath().vertices();
// const auto &edges = value.ValuePath().edges();
// return utils::FnvCollection<decltype(vertices), TVertexAccessor>{}(vertices) ^
// utils::FnvCollection<decltype(edges), TEdgeAccessor>{}(edges);
return 36;
}
case TypedValueT::Type::Date:
return utils::DateHash{}(value.ValueDate());

View File

@@ -13,11 +13,14 @@
#include "utils/exceptions.hpp"
namespace memgraph::storage::v3::durability {
/// Exception used to handle errors during recovery.
class RecoveryFailure : public utils::BasicException {
namespace memgraph::expr {
/**
* An exception raised by the TypedValue system. Typically when
* trying to perform operations (such as addition) on TypedValues
* of incompatible Types.
*/
class TypedValueException : public utils::BasicException {
public:
using utils::BasicException::BasicException;
};
} // namespace memgraph::storage::v3::durability
} // namespace memgraph::expr

View File

@@ -15,9 +15,17 @@
#include <string>
#include <vector>
#include "coordinator/shard_map.hpp"
#include "query/v2/accessors.hpp"
#include "query/v2/requests.hpp"
#include "query/v2/shard_request_manager.hpp"
#include "storage/v3/edge_accessor.hpp"
#include "storage/v3/id_types.hpp"
#include "storage/v3/result.hpp"
#include "storage/v3/shard.hpp"
#include "storage/v3/vertex_accessor.hpp"
#include "storage/v3/view.hpp"
#include "utils/exceptions.hpp"
#include "utils/temporal.hpp"
using memgraph::communication::bolt::Value;
@@ -63,17 +71,52 @@ query::v2::TypedValue ToTypedValue(const Value &value) {
}
}
storage::v3::Result<communication::bolt::Vertex> ToBoltVertex(const query::v2::VertexAccessor &vertex,
const storage::v3::Shard &db, storage::v3::View view) {
return ToBoltVertex(vertex.impl_, db, view);
storage::v3::Result<communication::bolt::Vertex> ToBoltVertex(
const query::v2::accessors::VertexAccessor &vertex, const msgs::ShardRequestManagerInterface *shard_request_manager,
storage::v3::View /*view*/) {
auto id = communication::bolt::Id::FromUint(0);
auto labels = vertex.Labels();
std::vector<std::string> new_labels;
new_labels.reserve(labels.size());
for (const auto &label : labels) {
new_labels.push_back(shard_request_manager->LabelToName(label.id));
}
auto properties = vertex.Properties();
std::map<std::string, Value> new_properties;
for (const auto &[prop, property_value] : properties) {
new_properties[shard_request_manager->PropertyToName(prop)] = ToBoltValue(property_value);
}
return communication::bolt::Vertex{id, new_labels, new_properties};
}
storage::v3::Result<communication::bolt::Edge> ToBoltEdge(const query::v2::EdgeAccessor &edge,
const storage::v3::Shard &db, storage::v3::View view) {
return ToBoltEdge(edge.impl_, db, view);
storage::v3::Result<communication::bolt::Edge> ToBoltEdge(
const query::v2::accessors::EdgeAccessor &edge, const msgs::ShardRequestManagerInterface *shard_request_manager,
storage::v3::View /*view*/) {
// TODO(jbajic) Fix bolt communication
auto id = communication::bolt::Id::FromUint(0);
auto from = communication::bolt::Id::FromUint(0);
auto to = communication::bolt::Id::FromUint(0);
const auto &type = shard_request_manager->EdgeTypeToName(edge.EdgeType());
auto properties = edge.Properties();
std::map<std::string, Value> new_properties;
for (const auto &[prop, property_value] : properties) {
new_properties[shard_request_manager->PropertyToName(prop)] = ToBoltValue(property_value);
}
return communication::bolt::Edge{id, from, to, type, new_properties};
}
storage::v3::Result<Value> ToBoltValue(const query::v2::TypedValue &value, const storage::v3::Shard &db,
storage::v3::Result<communication::bolt::Path> ToBoltPath(
const query::v2::accessors::Path & /*edge*/, const msgs::ShardRequestManagerInterface * /*shard_request_manager*/,
storage::v3::View /*view*/) {
// TODO(jbajic) Fix bolt communication
return {storage::v3::Error::DELETED_OBJECT};
}
storage::v3::Result<Value> ToBoltValue(const query::v2::TypedValue &value,
const msgs::ShardRequestManagerInterface *shard_request_manager,
storage::v3::View view) {
switch (value.type()) {
case query::v2::TypedValue::Type::Null:
@@ -90,7 +133,7 @@ storage::v3::Result<Value> ToBoltValue(const query::v2::TypedValue &value, const
std::vector<Value> values;
values.reserve(value.ValueList().size());
for (const auto &v : value.ValueList()) {
auto maybe_value = ToBoltValue(v, db, view);
auto maybe_value = ToBoltValue(v, shard_request_manager, view);
if (maybe_value.HasError()) return maybe_value.GetError();
values.emplace_back(std::move(*maybe_value));
}
@@ -99,24 +142,24 @@ storage::v3::Result<Value> ToBoltValue(const query::v2::TypedValue &value, const
case query::v2::TypedValue::Type::Map: {
std::map<std::string, Value> map;
for (const auto &kv : value.ValueMap()) {
auto maybe_value = ToBoltValue(kv.second, db, view);
auto maybe_value = ToBoltValue(kv.second, shard_request_manager, view);
if (maybe_value.HasError()) return maybe_value.GetError();
map.emplace(kv.first, std::move(*maybe_value));
}
return Value(std::move(map));
}
case query::v2::TypedValue::Type::Vertex: {
auto maybe_vertex = ToBoltVertex(value.ValueVertex(), db, view);
auto maybe_vertex = ToBoltVertex(value.ValueVertex(), shard_request_manager, view);
if (maybe_vertex.HasError()) return maybe_vertex.GetError();
return Value(std::move(*maybe_vertex));
}
case query::v2::TypedValue::Type::Edge: {
auto maybe_edge = ToBoltEdge(value.ValueEdge(), db, view);
auto maybe_edge = ToBoltEdge(value.ValueEdge(), shard_request_manager, view);
if (maybe_edge.HasError()) return maybe_edge.GetError();
return Value(std::move(*maybe_edge));
}
case query::v2::TypedValue::Type::Path: {
auto maybe_path = ToBoltPath(value.ValuePath(), db, view);
auto maybe_path = ToBoltPath(value.ValuePath(), shard_request_manager, view);
if (maybe_path.HasError()) return maybe_path.GetError();
return Value(std::move(*maybe_path));
}
@@ -131,59 +174,41 @@ storage::v3::Result<Value> ToBoltValue(const query::v2::TypedValue &value, const
}
}
storage::v3::Result<communication::bolt::Vertex> ToBoltVertex(const storage::v3::VertexAccessor &vertex,
const storage::v3::Shard &db, storage::v3::View view) {
// TODO(jbajic) Fix bolt communication
auto id = communication::bolt::Id::FromUint(0);
auto maybe_labels = vertex.Labels(view);
if (maybe_labels.HasError()) return maybe_labels.GetError();
std::vector<std::string> labels;
labels.reserve(maybe_labels->size());
for (const auto &label : *maybe_labels) {
labels.push_back(db.LabelToName(label));
Value ToBoltValue(msgs::Value value) {
switch (value.type) {
case msgs::Value::Type::Null:
return {};
case msgs::Value::Type::Bool:
return {value.bool_v};
case msgs::Value::Type::Int64:
return {value.int_v};
case msgs::Value::Type::Double:
return {value.double_v};
case msgs::Value::Type::String:
return {std::string(value.string_v)};
case msgs::Value::Type::List: {
std::vector<Value> values;
values.reserve(value.list_v.size());
for (const auto &v : value.list_v) {
auto maybe_value = ToBoltValue(v);
values.emplace_back(std::move(maybe_value));
}
return Value{std::move(values)};
}
case msgs::Value::Type::Map: {
std::map<std::string, Value> map;
for (const auto &kv : value.map_v) {
auto maybe_value = ToBoltValue(kv.second);
map.emplace(kv.first, std::move(maybe_value));
}
return Value{std::move(map)};
}
case msgs::Value::Type::Vertex:
case msgs::Value::Type::Edge: {
throw utils::BasicException("Vertex and Edge not supported!");
}
// TODO Value to Date types not supported
}
auto maybe_properties = vertex.Properties(view);
if (maybe_properties.HasError()) return maybe_properties.GetError();
std::map<std::string, Value> properties;
for (const auto &prop : *maybe_properties) {
properties[db.PropertyToName(prop.first)] = ToBoltValue(prop.second);
}
return communication::bolt::Vertex{id, labels, properties};
}
storage::v3::Result<communication::bolt::Edge> ToBoltEdge(const storage::v3::EdgeAccessor &edge,
const storage::v3::Shard &db, storage::v3::View view) {
// TODO(jbajic) Fix bolt communication
auto id = communication::bolt::Id::FromUint(0);
auto from = communication::bolt::Id::FromUint(0);
auto to = communication::bolt::Id::FromUint(0);
const auto &type = db.EdgeTypeToName(edge.EdgeType());
auto maybe_properties = edge.Properties(view);
if (maybe_properties.HasError()) return maybe_properties.GetError();
std::map<std::string, Value> properties;
for (const auto &prop : *maybe_properties) {
properties[db.PropertyToName(prop.first)] = ToBoltValue(prop.second);
}
return communication::bolt::Edge{id, from, to, type, properties};
}
storage::v3::Result<communication::bolt::Path> ToBoltPath(const query::v2::Path &path, const storage::v3::Shard &db,
storage::v3::View view) {
std::vector<communication::bolt::Vertex> vertices;
vertices.reserve(path.vertices().size());
for (const auto &v : path.vertices()) {
auto maybe_vertex = ToBoltVertex(v, db, view);
if (maybe_vertex.HasError()) return maybe_vertex.GetError();
vertices.emplace_back(std::move(*maybe_vertex));
}
std::vector<communication::bolt::Edge> edges;
edges.reserve(path.edges().size());
for (const auto &e : path.edges()) {
auto maybe_edge = ToBoltEdge(e, db, view);
if (maybe_edge.HasError()) return maybe_edge.GetError();
edges.emplace_back(std::move(*maybe_edge));
}
return communication::bolt::Path(vertices, edges);
}
storage::v3::PropertyValue ToPropertyValue(const Value &value) {

View File

@@ -13,9 +13,12 @@
#pragma once
#include "communication/bolt/v1/value.hpp"
#include "coordinator/shard_map.hpp"
#include "query/v2/bindings/typed_value.hpp"
#include "query/v2/shard_request_manager.hpp"
#include "storage/v3/property_value.hpp"
#include "storage/v3/result.hpp"
#include "storage/v3/shard.hpp"
#include "storage/v3/view.hpp"
namespace memgraph::storage::v3 {
@@ -28,36 +31,40 @@ namespace memgraph::glue::v2 {
/// @param storage::v3::VertexAccessor for converting to
/// communication::bolt::Vertex.
/// @param storage::v3::Shard for getting label and property names.
/// @param msgs::ShardRequestManagerInterface *shard_request_manager getting label and property names.
/// @param storage::v3::View for deciding which vertex attributes are visible.
///
/// @throw std::bad_alloc
storage::v3::Result<communication::bolt::Vertex> ToBoltVertex(const storage::v3::VertexAccessor &vertex,
const storage::v3::Shard &db, storage::v3::View view);
storage::v3::Result<communication::bolt::Vertex> ToBoltVertex(
const storage::v3::VertexAccessor &vertex, const msgs::ShardRequestManagerInterface *shard_request_manager,
storage::v3::View view);
/// @param storage::v3::EdgeAccessor for converting to communication::bolt::Edge.
/// @param storage::v3::Shard for getting edge type and property names.
/// @param msgs::ShardRequestManagerInterface *shard_request_manager getting edge type and property names.
/// @param storage::v3::View for deciding which edge attributes are visible.
///
/// @throw std::bad_alloc
storage::v3::Result<communication::bolt::Edge> ToBoltEdge(const storage::v3::EdgeAccessor &edge,
const storage::v3::Shard &db, storage::v3::View view);
storage::v3::Result<communication::bolt::Edge> ToBoltEdge(
const storage::v3::EdgeAccessor &edge, const msgs::ShardRequestManagerInterface *shard_request_manager,
storage::v3::View view);
/// @param query::v2::Path for converting to communication::bolt::Path.
/// @param storage::v3::Shard for ToBoltVertex and ToBoltEdge.
/// @param msgs::ShardRequestManagerInterface *shard_request_manager ToBoltVertex and ToBoltEdge.
/// @param storage::v3::View for ToBoltVertex and ToBoltEdge.
///
/// @throw std::bad_alloc
storage::v3::Result<communication::bolt::Path> ToBoltPath(const query::v2::Path &path, const storage::v3::Shard &db,
storage::v3::View view);
storage::v3::Result<communication::bolt::Path> ToBoltPath(
const query::v2::accessors::Path &path, const msgs::ShardRequestManagerInterface *shard_request_manager,
storage::v3::View view);
/// @param query::v2::TypedValue for converting to communication::bolt::Value.
/// @param storage::v3::Shard for ToBoltVertex and ToBoltEdge.
/// @param msgs::ShardRequestManagerInterface *shard_request_manager ToBoltVertex and ToBoltEdge.
/// @param storage::v3::View for ToBoltVertex and ToBoltEdge.
///
/// @throw std::bad_alloc
storage::v3::Result<communication::bolt::Value> ToBoltValue(const query::v2::TypedValue &value,
const storage::v3::Shard &db, storage::v3::View view);
storage::v3::Result<communication::bolt::Value> ToBoltValue(
const query::v2::TypedValue &value, const msgs::ShardRequestManagerInterface *shard_request_manager,
storage::v3::View view);
query::v2::TypedValue ToTypedValue(const communication::bolt::Value &value);
@@ -65,4 +72,10 @@ communication::bolt::Value ToBoltValue(const storage::v3::PropertyValue &value);
storage::v3::PropertyValue ToPropertyValue(const communication::bolt::Value &value);
communication::bolt::Value ToBoltValue(msgs::Value value);
communication::bolt::Value ToBoltValue(msgs::Value value,
const msgs::ShardRequestManagerInterface *shard_request_manager,
storage::v3::View view);
} // namespace memgraph::glue::v2

View File

@@ -15,11 +15,36 @@
#include <fmt/format.h>
#include <boost/asio/ip/tcp.hpp>
#include <boost/functional/hash.hpp>
#include <boost/uuid/uuid.hpp>
#include <boost/uuid/uuid_generators.hpp>
#include <boost/uuid/uuid_io.hpp>
namespace memgraph::io {
struct PartialAddress {
boost::asio::ip::address ip;
uint16_t port;
friend bool operator==(const PartialAddress &lhs, const PartialAddress &rhs) = default;
/// unique_id is most dominant for ordering, then ip, then port
friend bool operator<(const PartialAddress &lhs, const PartialAddress &rhs) {
if (lhs.ip != rhs.ip) {
return lhs.ip < rhs.ip;
}
return lhs.port < rhs.port;
}
std::string ToString() const { return fmt::format("PartialAddress {{ ip: {}, port: {} }}", ip.to_string(), port); }
friend std::ostream &operator<<(std::ostream &in, const PartialAddress &partial_address) {
in << partial_address.ToString();
return in;
}
};
struct Address {
// It's important for all participants to have a
// unique identifier - IP and port alone are not
@@ -54,6 +79,13 @@ struct Address {
};
}
PartialAddress ToPartialAddress() const {
return PartialAddress{
.ip = last_known_ip,
.port = last_known_port,
};
}
friend bool operator==(const Address &lhs, const Address &rhs) = default;
/// unique_id is most dominant for ordering, then last_known_ip, then last_known_port
@@ -73,5 +105,36 @@ struct Address {
return fmt::format("Address {{ unique_id: {}, last_known_ip: {}, last_known_port: {} }}",
boost::uuids::to_string(unique_id), last_known_ip.to_string(), last_known_port);
}
friend std::ostream &operator<<(std::ostream &in, const Address &address) {
in << address.ToString();
return in;
}
};
}; // namespace memgraph::io
namespace std {
template <>
struct hash<memgraph::io::PartialAddress> {
size_t operator()(const memgraph::io::PartialAddress &pa) const {
using boost::hash_combine;
using boost::hash_value;
// Start with a hash value of 0 .
std::size_t seed = 0;
if (pa.ip.is_v4()) {
auto h = std::hash<boost::asio::ip::address_v4>()(pa.ip.to_v4());
hash_combine(seed, h);
} else {
auto h = std::hash<boost::asio::ip::address_v6>()(pa.ip.to_v6());
hash_combine(seed, h);
}
hash_combine(seed, hash_value(pa.port));
// Return the result.
return seed;
}
};
} // namespace std

View File

@@ -99,7 +99,7 @@ class Shared {
bool IsReady() const {
std::unique_lock<std::mutex> lock(mu_);
return item_;
return item_.has_value();
}
std::optional<T> TryGet() {
@@ -151,7 +151,9 @@ class Future {
Future &operator=(Future &&old) noexcept {
MG_ASSERT(!old.consumed_or_moved_, "Future moved from after already being moved from or consumed.");
shared_ = std::move(old.shared_);
consumed_or_moved_ = old.consumed_or_moved_;
old.consumed_or_moved_ = true;
return *this;
}
Future(const Future &) = delete;
@@ -198,7 +200,7 @@ class Future {
template <typename T>
class Promise {
std::shared_ptr<details::Shared<T>> shared_;
bool filled_or_moved_ = false;
bool filled_or_moved_{false};
public:
explicit Promise(std::shared_ptr<details::Shared<T>> shared) : shared_(shared) {}
@@ -218,6 +220,7 @@ class Promise {
Promise(const Promise &) = delete;
Promise &operator=(const Promise &) = delete;
// NOLINTNEXTLINE(clang-analyzer-core.uninitialized.Branch)
~Promise() { MG_ASSERT(filled_or_moved_, "Promise destroyed before its associated Future was filled!"); }
// Fill the expected item into the Future.

View File

@@ -25,7 +25,7 @@ class LocalSystem {
public:
Io<LocalTransport> Register(Address address) {
LocalTransport local_transport(local_transport_handle_, address);
LocalTransport local_transport(local_transport_handle_);
return Io{local_transport, address};
}

View File

@@ -25,28 +25,20 @@ namespace memgraph::io::local_transport {
class LocalTransport {
std::shared_ptr<LocalTransportHandle> local_transport_handle_;
const Address address_;
public:
LocalTransport(std::shared_ptr<LocalTransportHandle> local_transport_handle, Address address)
: local_transport_handle_(std::move(local_transport_handle)), address_(address) {}
explicit LocalTransport(std::shared_ptr<LocalTransportHandle> local_transport_handle)
: local_transport_handle_(std::move(local_transport_handle)) {}
template <Message RequestT, Message ResponseT>
ResponseFuture<ResponseT> Request(Address to_address, RequestId request_id, RequestT request, Duration timeout) {
auto [future, promise] = memgraph::io::FuturePromisePair<ResponseResult<ResponseT>>();
Address from_address = address_;
local_transport_handle_->SubmitRequest(to_address, from_address, request_id, std::move(request), timeout,
std::move(promise));
return std::move(future);
ResponseFuture<ResponseT> Request(Address to_address, Address from_address, RequestT request, Duration timeout) {
return local_transport_handle_->template SubmitRequest<RequestT, ResponseT>(to_address, from_address,
std::move(request), timeout);
}
template <Message... Ms>
requires(sizeof...(Ms) > 0) RequestResult<Ms...> Receive(Duration timeout) {
Address from_address = address_;
return local_transport_handle_->template Receive<Ms...>(timeout);
requires(sizeof...(Ms) > 0) RequestResult<Ms...> Receive(Address receiver_address, Duration timeout) {
return local_transport_handle_->template Receive<Ms...>(receiver_address, timeout);
}
template <Message M>
@@ -63,5 +55,7 @@ class LocalTransport {
std::random_device rng;
return distrib(rng);
}
LatencyHistogramSummaries ResponseLatencies() { return local_transport_handle_->ResponseLatencies(); }
};
}; // namespace memgraph::io::local_transport

View File

@@ -19,6 +19,7 @@
#include "io/errors.hpp"
#include "io/message_conversion.hpp"
#include "io/message_histogram_collector.hpp"
#include "io/time.hpp"
#include "io/transport.hpp"
@@ -28,6 +29,8 @@ class LocalTransportHandle {
mutable std::mutex mu_{};
mutable std::condition_variable cv_;
bool should_shut_down_ = false;
MessageHistogramCollector histograms_;
RequestId request_id_counter_ = 0;
// the responses to requests that are being waited on
std::map<PromiseKey, DeadlineAndOpaquePromise> promises_;
@@ -36,6 +39,13 @@ class LocalTransportHandle {
std::vector<OpaqueMessage> can_receive_;
public:
~LocalTransportHandle() {
for (auto &&[pk, promise] : promises_) {
std::move(promise.promise).TimeOut();
}
promises_.clear();
}
void ShutDown() {
std::unique_lock<std::mutex> lock(mu_);
should_shut_down_ = true;
@@ -47,17 +57,24 @@ class LocalTransportHandle {
return should_shut_down_;
}
LatencyHistogramSummaries ResponseLatencies() {
std::unique_lock<std::mutex> lock(mu_);
return histograms_.ResponseLatencies();
}
static Time Now() {
auto nano_time = std::chrono::system_clock::now();
return std::chrono::time_point_cast<std::chrono::microseconds>(nano_time);
}
template <Message... Ms>
requires(sizeof...(Ms) > 0) RequestResult<Ms...> Receive(Duration timeout) {
requires(sizeof...(Ms) > 0) RequestResult<Ms...> Receive(Address /* receiver_address */, Duration timeout) {
std::unique_lock lock(mu_);
Time before = Now();
spdlog::info("can_receive_ size: {}", can_receive_.size());
while (can_receive_.empty()) {
Time now = Now();
@@ -88,24 +105,32 @@ class LocalTransportHandle {
template <Message M>
void Send(Address to_address, Address from_address, RequestId request_id, M &&message) {
std::any message_any(std::forward<M>(message));
OpaqueMessage opaque_message{
.from_address = from_address, .request_id = request_id, .message = std::move(message_any)};
auto type_info = TypeInfoFor(message);
PromiseKey promise_key{.requester_address = to_address,
.request_id = opaque_message.request_id,
.replier_address = opaque_message.from_address};
std::any message_any(std::forward<M>(message));
OpaqueMessage opaque_message{.to_address = to_address,
.from_address = from_address,
.request_id = request_id,
.message = std::move(message_any),
.type_info = type_info};
PromiseKey promise_key{.requester_address = to_address, .request_id = opaque_message.request_id};
{
std::unique_lock<std::mutex> lock(mu_);
if (promises_.contains(promise_key)) {
spdlog::info("using message to fill promise");
// complete waiting promise if it's there
DeadlineAndOpaquePromise dop = std::move(promises_.at(promise_key));
promises_.erase(promise_key);
dop.promise.Fill(std::move(opaque_message));
Duration response_latency = Now() - dop.requested_at;
dop.promise.Fill(std::move(opaque_message), response_latency);
histograms_.Measure(type_info, response_latency);
} else {
spdlog::info("placing message in can_receive_");
can_receive_.emplace_back(std::move(opaque_message));
}
} // lock dropped
@@ -114,25 +139,34 @@ class LocalTransportHandle {
}
template <Message RequestT, Message ResponseT>
void SubmitRequest(Address to_address, Address from_address, RequestId request_id, RequestT &&request,
Duration timeout, ResponsePromise<ResponseT> promise) {
ResponseFuture<ResponseT> SubmitRequest(Address to_address, Address from_address, RequestT &&request,
Duration timeout) {
auto [future, promise] = memgraph::io::FuturePromisePair<ResponseResult<ResponseT>>();
const bool port_matches = to_address.last_known_port == from_address.last_known_port;
const bool ip_matches = to_address.last_known_ip == from_address.last_known_ip;
MG_ASSERT(port_matches && ip_matches);
const Time deadline = Now() + timeout;
const auto now = Now();
const Time deadline = now + timeout;
RequestId request_id = 0;
{
std::unique_lock<std::mutex> lock(mu_);
PromiseKey promise_key{
.requester_address = from_address, .request_id = request_id, .replier_address = to_address};
request_id = ++request_id_counter_;
PromiseKey promise_key{.requester_address = from_address, .request_id = request_id};
OpaquePromise opaque_promise(std::move(promise).ToUnique());
DeadlineAndOpaquePromise dop{.deadline = deadline, .promise = std::move(opaque_promise)};
DeadlineAndOpaquePromise dop{.requested_at = now, .deadline = deadline, .promise = std::move(opaque_promise)};
MG_ASSERT(!promises_.contains(promise_key));
promises_.emplace(std::move(promise_key), std::move(dop));
} // lock dropped
Send(to_address, from_address, request_id, std::forward<RequestT>(request));
return std::move(future);
}
};

View File

@@ -11,20 +11,16 @@
#pragma once
#include <boost/core/demangle.hpp>
#include "io/transport.hpp"
#include "utils/type_info_ref.hpp"
namespace memgraph::io {
using memgraph::io::Duration;
using memgraph::io::Message;
using memgraph::io::Time;
struct PromiseKey {
Address requester_address;
uint64_t request_id;
// TODO(tyler) possibly remove replier_address from promise key
// once we want to support DSR.
Address replier_address;
public:
friend bool operator<(const PromiseKey &lhs, const PromiseKey &rhs) {
@@ -32,11 +28,7 @@ struct PromiseKey {
return lhs.requester_address < rhs.requester_address;
}
if (lhs.request_id != rhs.request_id) {
return lhs.request_id < rhs.request_id;
}
return lhs.replier_address < rhs.replier_address;
return lhs.request_id < rhs.request_id;
}
};
@@ -45,6 +37,7 @@ struct OpaqueMessage {
Address from_address;
uint64_t request_id;
std::any message;
utils::TypeInfoRef type_info;
/// Recursively tries to match a specific type from the outer
/// variant's parameter pack against the type of the std::any,
@@ -92,6 +85,10 @@ struct OpaqueMessage {
};
}
std::string demangled_name = "\"" + boost::core::demangle(message.type().name()) + "\"";
spdlog::error("failed to cast message of type {} to expected request type (probably in Receive argument types)",
demangled_name);
return std::nullopt;
}
};
@@ -100,7 +97,7 @@ class OpaquePromiseTraitBase {
public:
virtual const std::type_info *TypeInfo() const = 0;
virtual bool IsAwaited(void *ptr) const = 0;
virtual void Fill(void *ptr, OpaqueMessage &&) const = 0;
virtual void Fill(void *ptr, OpaqueMessage &&, Duration) const = 0;
virtual void TimeOut(void *ptr) const = 0;
virtual ~OpaquePromiseTraitBase() = default;
@@ -118,12 +115,13 @@ class OpaquePromiseTrait : public OpaquePromiseTraitBase {
bool IsAwaited(void *ptr) const override { return static_cast<ResponsePromise<T> *>(ptr)->IsAwaited(); };
void Fill(void *ptr, OpaqueMessage &&opaque_message) const override {
void Fill(void *ptr, OpaqueMessage &&opaque_message, Duration response_latency) const override {
T message = std::any_cast<T>(std::move(opaque_message.message));
auto response_envelope = ResponseEnvelope<T>{.message = std::move(message),
.request_id = opaque_message.request_id,
.to_address = opaque_message.to_address,
.from_address = opaque_message.from_address};
.from_address = opaque_message.from_address,
.response_latency = response_latency};
auto promise = static_cast<ResponsePromise<T> *>(ptr);
auto unique_promise = std::unique_ptr<ResponsePromise<T>>(promise);
unique_promise->Fill(std::move(response_envelope));
@@ -187,9 +185,9 @@ class OpaquePromise {
ptr_ = nullptr;
}
void Fill(OpaqueMessage &&opaque_message) {
void Fill(OpaqueMessage &&opaque_message, Duration response_latency) {
MG_ASSERT(ptr_ != nullptr);
trait_->Fill(ptr_, std::move(opaque_message));
trait_->Fill(ptr_, std::move(opaque_message), response_latency);
ptr_ = nullptr;
}
@@ -199,8 +197,40 @@ class OpaquePromise {
};
struct DeadlineAndOpaquePromise {
Time requested_at;
Time deadline;
OpaquePromise promise;
};
template <class From>
std::type_info const &TypeInfoForVariant(From const &from) {
return std::visit([](auto &&x) -> decltype(auto) { return typeid(x); }, from);
}
template <class T>
utils::TypeInfoRef TypeInfoFor(const T & /* t */) {
return typeid(T);
}
template <typename From, typename Return, typename Head, typename... Rest>
std::optional<Return> ConvertVariantInner(From &&a) {
if (typeid(Head) == TypeInfoForVariant(a)) {
Head concrete = std::get<Head>(std::forward<From>(a));
return concrete;
}
if constexpr (sizeof...(Rest) > 0) {
return ConvertVariantInner<From, Return, Rest...>(std::forward<From>(a));
} else {
return std::nullopt;
}
}
/// This function converts a variant to another variant holding a subset OR superset of
/// possible types.
template <class From, class... Ms>
requires(sizeof...(Ms) > 0) std::optional<std::variant<Ms...>> ConvertVariant(From &&from) {
return ConvertVariantInner<From, std::variant<Ms...>, Ms...>(std::forward<From>(from));
}
} // namespace memgraph::io

View File

@@ -0,0 +1,127 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#pragma once
#include <chrono>
#include <cmath>
#include <unordered_map>
#include <boost/core/demangle.hpp>
#include "io/time.hpp"
#include "utils/histogram.hpp"
#include "utils/logging.hpp"
#include "utils/print_helpers.hpp"
#include "utils/type_info_ref.hpp"
namespace memgraph::io {
struct LatencyHistogramSummary {
uint64_t count;
Duration p0;
Duration p50;
Duration p75;
Duration p90;
Duration p95;
Duration p975;
Duration p99;
Duration p999;
Duration p9999;
Duration p100;
Duration sum;
friend std::ostream &operator<<(std::ostream &in, const LatencyHistogramSummary &histo) {
in << "{ \"count\": " << histo.count;
in << ", \"p0\": " << histo.p0.count();
in << ", \"p50\": " << histo.p50.count();
in << ", \"p75\": " << histo.p75.count();
in << ", \"p90\": " << histo.p90.count();
in << ", \"p95\": " << histo.p95.count();
in << ", \"p975\": " << histo.p975.count();
in << ", \"p99\": " << histo.p99.count();
in << ", \"p999\": " << histo.p999.count();
in << ", \"p9999\": " << histo.p9999.count();
in << ", \"p100\": " << histo.p100.count();
in << ", \"sum\": " << histo.sum.count();
in << " }";
return in;
}
};
struct LatencyHistogramSummaries {
std::unordered_map<std::string, LatencyHistogramSummary> latencies;
std::string SummaryTable() {
std::string output;
const auto row = [&output](const auto &c1, const auto &c2, const auto &c3, const auto &c4, const auto &c5,
const auto &c6, const auto &c7) {
output +=
fmt::format("{: >50} | {: >8} | {: >8} | {: >8} | {: >8} | {: >8} | {: >8}\n", c1, c2, c3, c4, c5, c6, c7);
};
row("name", "count", "min (μs)", "med (μs)", "p99 (μs)", "max (μs)", "sum (ms)");
for (const auto &[name, histo] : latencies) {
row(name, histo.count, histo.p0.count(), histo.p50.count(), histo.p99.count(), histo.p100.count(),
histo.sum.count() / 1000);
}
output += "\n";
return output;
}
friend std::ostream &operator<<(std::ostream &in, const LatencyHistogramSummaries &histo) {
using memgraph::utils::print_helpers::operator<<;
in << histo.latencies;
return in;
}
};
class MessageHistogramCollector {
std::unordered_map<utils::TypeInfoRef, utils::Histogram, utils::TypeInfoHasher, utils::TypeInfoEqualTo> histograms_;
public:
void Measure(const std::type_info &type_info, const Duration &duration) {
auto &histo = histograms_[type_info];
histo.Measure(duration.count());
}
LatencyHistogramSummaries ResponseLatencies() {
std::unordered_map<std::string, LatencyHistogramSummary> ret{};
for (const auto &[type_id, histo] : histograms_) {
std::string demangled_name = "\"" + boost::core::demangle(type_id.get().name()) + "\"";
LatencyHistogramSummary latency_histogram_summary{
.count = histo.Count(),
.p0 = Duration(static_cast<int64_t>(histo.Percentile(0.0))),
.p50 = Duration(static_cast<int64_t>(histo.Percentile(50.0))),
.p75 = Duration(static_cast<int64_t>(histo.Percentile(75.0))),
.p90 = Duration(static_cast<int64_t>(histo.Percentile(90.0))),
.p95 = Duration(static_cast<int64_t>(histo.Percentile(95.0))),
.p975 = Duration(static_cast<int64_t>(histo.Percentile(97.5))),
.p99 = Duration(static_cast<int64_t>(histo.Percentile(99.0))),
.p999 = Duration(static_cast<int64_t>(histo.Percentile(99.9))),
.p9999 = Duration(static_cast<int64_t>(histo.Percentile(99.99))),
.p100 = Duration(static_cast<int64_t>(histo.Percentile(100.0))),
.sum = Duration(histo.Sum()),
};
ret.emplace(demangled_name, latency_histogram_summary);
}
return LatencyHistogramSummaries{.latencies = ret};
}
};
} // namespace memgraph::io

46
src/io/messages.hpp Normal file
View File

@@ -0,0 +1,46 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#pragma once
#include <variant>
#include <coordinator/coordinator.hpp>
#include <io/rsm/raft.hpp>
#include "query/v2/requests.hpp"
#include "utils/concepts.hpp"
namespace memgraph::io::messages {
using memgraph::coordinator::CoordinatorReadRequests;
using memgraph::coordinator::CoordinatorWriteRequests;
using memgraph::coordinator::CoordinatorWriteResponses;
using StorageReadRequest = msgs::ReadRequests;
using StorageWriteRequest = msgs::WriteRequests;
using memgraph::io::rsm::AppendRequest;
using memgraph::io::rsm::AppendResponse;
using memgraph::io::rsm::ReadRequest;
using memgraph::io::rsm::VoteRequest;
using memgraph::io::rsm::VoteResponse;
using memgraph::io::rsm::WriteRequest;
using memgraph::io::rsm::WriteResponse;
using CoordinatorMessages =
std::variant<ReadRequest<CoordinatorReadRequests>, AppendRequest<CoordinatorWriteRequests>, AppendResponse,
WriteRequest<CoordinatorWriteRequests>, VoteRequest, VoteResponse>;
using ShardMessages = std::variant<ReadRequest<StorageReadRequest>, AppendRequest<StorageWriteRequest>, AppendResponse,
WriteRequest<StorageWriteRequest>, VoteRequest, VoteResponse>;
using ShardManagerMessages = std::variant<WriteResponse<CoordinatorWriteResponses>>;
} // namespace memgraph::io::messages

View File

@@ -22,6 +22,9 @@
#include <unordered_map>
#include <vector>
#include <boost/core/demangle.hpp>
#include "io/message_conversion.hpp"
#include "io/simulator/simulator.hpp"
#include "io/transport.hpp"
#include "utils/concepts.hpp"
@@ -53,7 +56,6 @@ static constexpr size_t kMaximumAppendBatchSize = 1024;
using Term = uint64_t;
using LogIndex = uint64_t;
using LogSize = uint64_t;
using RequestId = uint64_t;
template <typename WriteOperation>
struct WriteRequest {
@@ -89,6 +91,36 @@ struct ReadResponse {
std::optional<Address> retry_leader;
};
template <class... ReadReturn>
utils::TypeInfoRef TypeInfoFor(const ReadResponse<std::variant<ReadReturn...>> &read_response) {
return TypeInfoForVariant(read_response.read_return);
}
template <class ReadReturn>
utils::TypeInfoRef TypeInfoFor(const ReadResponse<ReadReturn> & /* read_response */) {
return typeid(ReadReturn);
}
template <class... WriteReturn>
utils::TypeInfoRef TypeInfoFor(const WriteResponse<std::variant<WriteReturn...>> &write_response) {
return TypeInfoForVariant(write_response.write_return);
}
template <class WriteReturn>
utils::TypeInfoRef TypeInfoFor(const WriteResponse<WriteReturn> & /* write_response */) {
return typeid(WriteReturn);
}
template <class WriteOperation>
utils::TypeInfoRef TypeInfoFor(const WriteRequest<WriteOperation> & /* write_request */) {
return typeid(WriteOperation);
}
template <class... WriteOperations>
utils::TypeInfoRef TypeInfoFor(const WriteRequest<std::variant<WriteOperations...>> &write_request) {
return TypeInfoForVariant(write_request.operation);
}
/// AppendRequest is a raft-level message that the Leader
/// periodically broadcasts to all Follower peers. This
/// serves three main roles:
@@ -230,22 +262,60 @@ class Raft {
Io<IoImpl> io_;
std::vector<Address> peers_;
ReplicatedState replicated_state_;
Time next_cron_;
public:
Raft(Io<IoImpl> &&io, std::vector<Address> peers, ReplicatedState &&replicated_state)
: io_(std::forward<Io<IoImpl>>(io)),
peers_(peers),
replicated_state_(std::forward<ReplicatedState>(replicated_state)) {}
replicated_state_(std::forward<ReplicatedState>(replicated_state)) {
if (peers.empty()) {
role_ = Leader{};
}
}
/// Periodic protocol maintenance. Returns the time that Cron should be called again
/// in the future.
Time Cron() {
// dispatch periodic logic based on our role to a specific Cron method.
std::optional<Role> new_role = std::visit([&](auto &role) { return Cron(role); }, role_);
if (new_role) {
role_ = std::move(new_role).value();
}
const Duration random_cron_interval = RandomTimeout(kMinimumCronInterval, kMaximumCronInterval);
return io_.Now() + random_cron_interval;
}
/// Returns the Address for our underlying Io implementation
Address GetAddress() { return io_.GetAddress(); }
using ReceiveVariant = std::variant<ReadRequest<ReadOperation>, AppendRequest<WriteOperation>, AppendResponse,
WriteRequest<WriteOperation>, VoteRequest, VoteResponse>;
void Handle(ReceiveVariant &&message_variant, RequestId request_id, Address from_address) {
// dispatch the message to a handler based on our role,
// which can be specified in the Handle first argument,
// or it can be `auto` if it's a handler for several roles
// or messages.
std::optional<Role> new_role = std::visit(
[&](auto &&msg, auto &role) mutable {
return Handle(role, std::forward<decltype(msg)>(msg), request_id, from_address);
},
std::forward<ReceiveVariant>(message_variant), role_);
// TODO(tyler) (M3) maybe replace std::visit with get_if for explicit prioritized matching, [[likely]] etc...
if (new_role) {
role_ = std::move(new_role).value();
}
}
void Run() {
Time last_cron = io_.Now();
while (!io_.ShouldShutDown()) {
const auto now = io_.Now();
const Duration random_cron_interval = RandomTimeout(kMinimumCronInterval, kMaximumCronInterval);
if (now - last_cron > random_cron_interval) {
Cron();
last_cron = now;
if (now >= next_cron_) {
next_cron_ = Cron();
}
const Duration receive_timeout = RandomTimeout(kMinimumReceiveTimeout, kMaximumReceiveTimeout);
@@ -449,16 +519,6 @@ class Raft {
/// been received.
/////////////////////////////////////////////////////////////
/// Periodic protocol maintenance.
void Cron() {
// dispatch periodic logic based on our role to a specific Cron method.
std::optional<Role> new_role = std::visit([&](auto &role) { return Cron(role); }, role_);
if (new_role) {
role_ = std::move(new_role).value();
}
}
// Raft paper - 5.2
// Candidates keep sending Vote to peers until:
// 1. receiving Append with a higher term (become Follower)
@@ -521,7 +581,7 @@ class Raft {
const Time now = io_.Now();
const Duration broadcast_timeout = RandomTimeout(kMinimumBroadcastTimeout, kMaximumBroadcastTimeout);
if (now - leader.last_broadcast > broadcast_timeout) {
if (now > leader.last_broadcast + broadcast_timeout) {
BroadcastAppendEntries(leader.followers);
leader.last_broadcast = now;
}
@@ -541,26 +601,6 @@ class Raft {
/// message that has been received.
/////////////////////////////////////////////////////////////
using ReceiveVariant = std::variant<ReadRequest<ReadOperation>, AppendRequest<WriteOperation>, AppendResponse,
WriteRequest<WriteOperation>, VoteRequest, VoteResponse>;
void Handle(ReceiveVariant &&message_variant, RequestId request_id, Address from_address) {
// dispatch the message to a handler based on our role,
// which can be specified in the Handle first argument,
// or it can be `auto` if it's a handler for several roles
// or messages.
std::optional<Role> new_role = std::visit(
[&](auto &&msg, auto &role) mutable {
return Handle(role, std::forward<decltype(msg)>(msg), request_id, from_address);
},
std::forward<ReceiveVariant>(message_variant), role_);
// TODO(tyler) (M3) maybe replace std::visit with get_if for explicit prioritized matching, [[likely]] etc...
if (new_role) {
role_ = std::move(new_role).value();
}
}
// all roles can receive Vote and possibly become a follower
template <AllRoles ALL>
std::optional<Role> Handle(ALL & /* variable */, VoteRequest &&req, RequestId request_id, Address from_address) {
@@ -890,7 +930,9 @@ class Raft {
// only leaders actually handle replication requests from clients
std::optional<Role> Handle(Leader &leader, WriteRequest<WriteOperation> &&req, RequestId request_id,
Address from_address) {
Log("handling WriteRequest");
auto type_info = TypeInfoFor(req);
std::string demangled_name = boost::core::demangle(type_info.get().name());
Log("handling WriteRequest<" + demangled_name + ">");
// we are the leader. add item to log and send Append to peers
MG_ASSERT(state_.term >= LastLogTerm());
@@ -906,7 +948,11 @@ class Raft {
leader.pending_client_requests.emplace(log_index, pcr);
BroadcastAppendEntries(leader.followers);
if (peers_.empty()) {
BumpCommitIndexAndReplyToClients(leader);
} else {
BroadcastAppendEntries(leader.followers);
}
return std::nullopt;
}

View File

@@ -13,9 +13,11 @@
#include <iostream>
#include <optional>
#include <type_traits>
#include <vector>
#include "io/address.hpp"
#include "io/errors.hpp"
#include "io/rsm/raft.hpp"
#include "utils/result.hpp"
@@ -43,21 +45,36 @@ class RsmClient {
Address leader_;
ServerPool server_addrs_;
/// State for single async read/write operations. In the future this could become a map
/// of async operations that can be accessed via an ID etc...
std::optional<Time> async_read_before_;
std::optional<ResponseFuture<ReadResponse<ReadResponseT>>> async_read_;
ReadRequestT current_read_request_;
std::optional<Time> async_write_before_;
std::optional<ResponseFuture<WriteResponse<WriteResponseT>>> async_write_;
WriteRequestT current_write_request_;
void SelectRandomLeader() {
std::uniform_int_distribution<size_t> addr_distrib(0, (server_addrs_.size() - 1));
size_t addr_index = io_.Rand(addr_distrib);
leader_ = server_addrs_[addr_index];
spdlog::debug(
"client NOT redirected to leader server despite our success failing to be processed (it probably was sent to "
"a RSM Candidate) trying a random one at index {} with address {}",
addr_index, leader_.ToString());
}
template <typename ResponseT>
void PossiblyRedirectLeader(const ResponseT &response) {
if (response.retry_leader) {
MG_ASSERT(!response.success, "retry_leader should never be set for successful responses");
leader_ = response.retry_leader.value();
spdlog::debug("client redirected to leader server {}", leader_.ToString());
} else if (!response.success) {
std::uniform_int_distribution<size_t> addr_distrib(0, (server_addrs_.size() - 1));
size_t addr_index = io_.Rand(addr_distrib);
leader_ = server_addrs_[addr_index];
spdlog::debug(
"client NOT redirected to leader server despite our success failing to be processed (it probably was sent to "
"a RSM Candidate) trying a random one at index {} with address {}",
addr_index, leader_.ToString());
}
if (!response.success) {
SelectRandomLeader();
}
}
@@ -65,7 +82,13 @@ class RsmClient {
RsmClient(Io<IoImpl> io, Address leader, ServerPool server_addrs)
: io_{io}, leader_{leader}, server_addrs_{server_addrs} {}
RsmClient(const RsmClient &) = delete;
RsmClient &operator=(const RsmClient &) = delete;
RsmClient(RsmClient &&) noexcept = default;
RsmClient &operator=(RsmClient &&) noexcept = default;
RsmClient() = delete;
~RsmClient() = default;
BasicResult<TimedOut, WriteResponseT> SendWriteRequest(WriteRequestT req) {
WriteRequest<WriteRequestT> client_req;
@@ -131,6 +154,117 @@ class RsmClient {
return TimedOut{};
}
/// AsyncRead methods
void SendAsyncReadRequest(const ReadRequestT &req) {
MG_ASSERT(!async_read_);
ReadRequest<ReadRequestT> read_req = {.operation = req};
if (!async_read_before_) {
async_read_before_ = io_.Now();
}
current_read_request_ = std::move(req);
async_read_ = io_.template Request<ReadRequest<ReadRequestT>, ReadResponse<ReadResponseT>>(leader_, read_req);
}
std::optional<BasicResult<TimedOut, ReadResponseT>> PollAsyncReadRequest() {
MG_ASSERT(async_read_);
if (!async_read_->IsReady()) {
return std::nullopt;
}
return AwaitAsyncReadRequest();
}
std::optional<BasicResult<TimedOut, ReadResponseT>> AwaitAsyncReadRequest() {
ResponseResult<ReadResponse<ReadResponseT>> get_response_result = std::move(*async_read_).Wait();
async_read_.reset();
const Duration overall_timeout = io_.GetDefaultTimeout();
const bool past_time_out = io_.Now() < *async_read_before_ + overall_timeout;
const bool result_has_error = get_response_result.HasError();
if (result_has_error && past_time_out) {
// TODO static assert the exact type of error.
spdlog::debug("client timed out while trying to communicate with leader server {}", leader_.ToString());
async_read_before_ = std::nullopt;
return TimedOut{};
}
if (!result_has_error) {
ResponseEnvelope<ReadResponse<ReadResponseT>> &&get_response_envelope = std::move(get_response_result.GetValue());
ReadResponse<ReadResponseT> &&read_get_response = std::move(get_response_envelope.message);
PossiblyRedirectLeader(read_get_response);
if (read_get_response.success) {
async_read_before_ = std::nullopt;
return std::move(read_get_response.read_return);
}
SendAsyncReadRequest(current_read_request_);
} else if (result_has_error) {
SelectRandomLeader();
SendAsyncReadRequest(current_read_request_);
}
return std::nullopt;
}
/// AsyncWrite methods
void SendAsyncWriteRequest(const WriteRequestT &req) {
MG_ASSERT(!async_write_);
WriteRequest<WriteRequestT> write_req = {.operation = req};
if (!async_write_before_) {
async_write_before_ = io_.Now();
}
current_write_request_ = std::move(req);
async_write_ = io_.template Request<WriteRequest<WriteRequestT>, WriteResponse<WriteResponseT>>(leader_, write_req);
}
std::optional<BasicResult<TimedOut, WriteResponseT>> PollAsyncWriteRequest() {
MG_ASSERT(async_write_);
if (!async_write_->IsReady()) {
return std::nullopt;
}
return AwaitAsyncWriteRequest();
}
std::optional<BasicResult<TimedOut, WriteResponseT>> AwaitAsyncWriteRequest() {
ResponseResult<WriteResponse<WriteResponseT>> get_response_result = std::move(*async_write_).Wait();
async_write_.reset();
const Duration overall_timeout = io_.GetDefaultTimeout();
const bool past_time_out = io_.Now() < *async_write_before_ + overall_timeout;
const bool result_has_error = get_response_result.HasError();
if (result_has_error && past_time_out) {
// TODO static assert the exact type of error.
spdlog::debug("client timed out while trying to communicate with leader server {}", leader_.ToString());
async_write_before_ = std::nullopt;
return TimedOut{};
}
if (!result_has_error) {
ResponseEnvelope<WriteResponse<WriteResponseT>> &&get_response_envelope =
std::move(get_response_result.GetValue());
WriteResponse<WriteResponseT> &&write_get_response = std::move(get_response_envelope.message);
PossiblyRedirectLeader(write_get_response);
if (write_get_response.success) {
async_write_before_ = std::nullopt;
return std::move(write_get_response.write_return);
}
SendAsyncWriteRequest(current_write_request_);
} else if (result_has_error) {
SelectRandomLeader();
SendAsyncWriteRequest(current_write_request_);
}
return std::nullopt;
}
};
} // namespace memgraph::io::rsm

View File

@@ -29,6 +29,8 @@ class Simulator {
explicit Simulator(SimulatorConfig config)
: rng_(std::mt19937{config.rng_seed}), simulator_handle_{std::make_shared<SimulatorHandle>(config)} {}
~Simulator() { ShutDown(); }
void ShutDown() { simulator_handle_->ShutDown(); }
Io<SimulatorTransport> RegisterNew() {

View File

@@ -16,12 +16,10 @@
#include "io/simulator/simulator_stats.hpp"
#include "io/time.hpp"
#include "io/transport.hpp"
#include "utils/exceptions.hpp"
namespace memgraph::io::simulator {
using memgraph::io::Duration;
using memgraph::io::Time;
void SimulatorHandle::ShutDown() {
std::unique_lock<std::mutex> lock(mu_);
should_shut_down_ = true;
@@ -33,12 +31,17 @@ bool SimulatorHandle::ShouldShutDown() const {
return should_shut_down_;
}
LatencyHistogramSummaries SimulatorHandle::ResponseLatencies() {
std::unique_lock<std::mutex> lock(mu_);
return histograms_.ResponseLatencies();
}
void SimulatorHandle::IncrementServerCountAndWaitForQuiescentState(Address address) {
std::unique_lock<std::mutex> lock(mu_);
server_addresses_.insert(address);
while (true) {
const size_t blocked_servers = blocked_on_receive_;
const size_t blocked_servers = blocked_on_receive_.size();
const bool all_servers_blocked = blocked_servers == server_addresses_.size();
@@ -53,7 +56,7 @@ void SimulatorHandle::IncrementServerCountAndWaitForQuiescentState(Address addre
bool SimulatorHandle::MaybeTickSimulator() {
std::unique_lock<std::mutex> lock(mu_);
const size_t blocked_servers = blocked_on_receive_;
const size_t blocked_servers = blocked_on_receive_.size();
if (blocked_servers < server_addresses_.size()) {
// we only need to advance the simulator when all
@@ -73,13 +76,18 @@ bool SimulatorHandle::MaybeTickSimulator() {
// We tick the clock forward when all servers are blocked but
// there are no in-flight messages to schedule delivery of.
std::poisson_distribution<> time_distrib(50);
Duration clock_advance = std::chrono::microseconds{time_distrib(rng_)};
const Duration clock_advance = std::chrono::microseconds{time_distrib_(rng_)};
cluster_wide_time_microseconds_ += clock_advance;
MG_ASSERT(cluster_wide_time_microseconds_ < config_.abort_time,
"Cluster has executed beyond its configured abort_time, and something may be failing to make progress "
"in an expected amount of time.");
if (cluster_wide_time_microseconds_ >= config_.abort_time) {
if (should_shut_down_) {
return false;
}
spdlog::error(
"Cluster has executed beyond its configured abort_time, and something may be failing to make progress "
"in an expected amount of time.");
throw utils::BasicException{"Cluster has executed beyond its configured abort_time"};
}
return true;
}
@@ -93,17 +101,14 @@ bool SimulatorHandle::MaybeTickSimulator() {
auto [to_address, opaque_message] = std::move(in_flight_.back());
in_flight_.pop_back();
std::uniform_int_distribution<int> drop_distrib(0, 99);
const int drop_threshold = drop_distrib(rng_);
const int drop_threshold = drop_distrib_(rng_);
const bool should_drop = drop_threshold < config_.drop_percent;
if (should_drop) {
stats_.dropped_messages++;
}
PromiseKey promise_key{.requester_address = to_address,
.request_id = opaque_message.request_id,
.replier_address = opaque_message.from_address};
PromiseKey promise_key{.requester_address = to_address, .request_id = opaque_message.request_id};
if (promises_.contains(promise_key)) {
// complete waiting promise if it's there
@@ -117,13 +122,17 @@ bool SimulatorHandle::MaybeTickSimulator() {
dop.promise.TimeOut();
} else {
stats_.total_responses++;
dop.promise.Fill(std::move(opaque_message));
Duration response_latency = cluster_wide_time_microseconds_ - dop.requested_at;
auto type_info = opaque_message.type_info;
dop.promise.Fill(std::move(opaque_message), response_latency);
histograms_.Measure(type_info, response_latency);
}
} else if (should_drop) {
// don't add it anywhere, let it drop
} else {
// add to can_receive_ if not
const auto &[om_vec, inserted] = can_receive_.try_emplace(to_address, std::vector<OpaqueMessage>());
const auto &[om_vec, inserted] =
can_receive_.try_emplace(to_address.ToPartialAddress(), std::vector<OpaqueMessage>());
om_vec->second.emplace_back(std::move(opaque_message));
}

View File

@@ -25,6 +25,7 @@
#include "io/address.hpp"
#include "io/errors.hpp"
#include "io/message_conversion.hpp"
#include "io/message_histogram_collector.hpp"
#include "io/simulator/simulator_config.hpp"
#include "io/simulator/simulator_stats.hpp"
#include "io/time.hpp"
@@ -43,23 +44,26 @@ class SimulatorHandle {
std::map<PromiseKey, DeadlineAndOpaquePromise> promises_;
// messages that are sent to servers that may later receive them
std::map<Address, std::vector<OpaqueMessage>> can_receive_;
std::map<PartialAddress, std::vector<OpaqueMessage>> can_receive_;
Time cluster_wide_time_microseconds_;
bool should_shut_down_ = false;
SimulatorStats stats_;
size_t blocked_on_receive_ = 0;
std::set<Address> blocked_on_receive_;
std::set<Address> server_addresses_;
std::mt19937 rng_;
std::uniform_int_distribution<int> time_distrib_{5, 50};
std::uniform_int_distribution<int> drop_distrib_{0, 99};
SimulatorConfig config_;
MessageHistogramCollector histograms_;
RequestId request_id_counter_{0};
void TimeoutPromisesPastDeadline() {
const Time now = cluster_wide_time_microseconds_;
for (auto it = promises_.begin(); it != promises_.end();) {
auto &[promise_key, dop] = *it;
if (dop.deadline < now) {
spdlog::info("timing out request from requester {} to replier {}.", promise_key.requester_address.ToString(),
promise_key.replier_address.ToString());
if (dop.deadline < now && config_.perform_timeouts) {
spdlog::info("timing out request from requester {}.", promise_key.requester_address.ToString());
std::move(dop).promise.TimeOut();
it = promises_.erase(it);
@@ -74,6 +78,16 @@ class SimulatorHandle {
explicit SimulatorHandle(SimulatorConfig config)
: cluster_wide_time_microseconds_(config.start_time), rng_(config.rng_seed), config_(config) {}
LatencyHistogramSummaries ResponseLatencies();
~SimulatorHandle() {
for (auto it = promises_.begin(); it != promises_.end();) {
auto &[promise_key, dop] = *it;
std::move(dop).promise.TimeOut();
it = promises_.erase(it);
}
}
void IncrementServerCountAndWaitForQuiescentState(Address address);
/// This method causes most of the interesting simulation logic to happen, wrt network behavior.
@@ -87,41 +101,60 @@ class SimulatorHandle {
bool ShouldShutDown() const;
template <Message Request, Message Response>
void SubmitRequest(Address to_address, Address from_address, RequestId request_id, Request &&request,
Duration timeout, ResponsePromise<Response> &&promise) {
ResponseFuture<Response> SubmitRequest(Address to_address, Address from_address, Request &&request, Duration timeout,
std::function<bool()> &&maybe_tick_simulator) {
auto type_info = TypeInfoFor(request);
auto [future, promise] = memgraph::io::FuturePromisePairWithNotifier<ResponseResult<Response>>(
std::forward<std::function<bool()>>(maybe_tick_simulator));
std::unique_lock<std::mutex> lock(mu_);
RequestId request_id = ++request_id_counter_;
const Time deadline = cluster_wide_time_microseconds_ + timeout;
std::any message(request);
OpaqueMessage om{.to_address = to_address,
.from_address = from_address,
.request_id = request_id,
.message = std::move(message)};
.message = std::move(message),
.type_info = type_info};
in_flight_.emplace_back(std::make_pair(to_address, std::move(om)));
PromiseKey promise_key{.requester_address = from_address, .request_id = request_id, .replier_address = to_address};
PromiseKey promise_key{.requester_address = from_address, .request_id = request_id};
OpaquePromise opaque_promise(std::move(promise).ToUnique());
DeadlineAndOpaquePromise dop{.deadline = deadline, .promise = std::move(opaque_promise)};
DeadlineAndOpaquePromise dop{
.requested_at = cluster_wide_time_microseconds_,
.deadline = deadline,
.promise = std::move(opaque_promise),
};
MG_ASSERT(!promises_.contains(promise_key));
promises_.emplace(std::move(promise_key), std::move(dop));
stats_.total_messages++;
stats_.total_requests++;
cv_.notify_all();
return std::move(future);
}
template <Message... Ms>
requires(sizeof...(Ms) > 0) RequestResult<Ms...> Receive(const Address &receiver, Duration timeout) {
std::unique_lock<std::mutex> lock(mu_);
blocked_on_receive_ += 1;
blocked_on_receive_.emplace(receiver);
const Time deadline = cluster_wide_time_microseconds_ + timeout;
auto partial_address = receiver.ToPartialAddress();
while (!should_shut_down_ && (cluster_wide_time_microseconds_ < deadline)) {
if (can_receive_.contains(receiver)) {
std::vector<OpaqueMessage> &can_rx = can_receive_.at(receiver);
if (can_receive_.contains(partial_address)) {
std::vector<OpaqueMessage> &can_rx = can_receive_.at(partial_address);
if (!can_rx.empty()) {
OpaqueMessage message = std::move(can_rx.back());
can_rx.pop_back();
@@ -129,8 +162,9 @@ class SimulatorHandle {
// TODO(tyler) search for item in can_receive_ that matches the desired types, rather
// than asserting that the last item in can_rx matches.
auto m_opt = std::move(message).Take<Ms...>();
MG_ASSERT(m_opt.has_value(), "Wrong message type received compared to the expected type");
blocked_on_receive_ -= 1;
blocked_on_receive_.erase(receiver);
return std::move(m_opt).value();
}
@@ -144,19 +178,21 @@ class SimulatorHandle {
}
}
blocked_on_receive_ -= 1;
blocked_on_receive_.erase(receiver);
return TimedOut{};
}
template <Message M>
void Send(Address to_address, Address from_address, RequestId request_id, M message) {
auto type_info = TypeInfoFor(message);
std::unique_lock<std::mutex> lock(mu_);
std::any message_any(std::move(message));
OpaqueMessage om{.to_address = to_address,
.from_address = from_address,
.request_id = request_id,
.message = std::move(message_any)};
.message = std::move(message_any),
.type_info = type_info};
in_flight_.emplace_back(std::make_pair(std::move(to_address), std::move(om)));
stats_.total_messages++;

View File

@@ -33,19 +33,16 @@ class SimulatorTransport {
: simulator_handle_(simulator_handle), address_(address), rng_(std::mt19937{seed}) {}
template <Message RequestT, Message ResponseT>
ResponseFuture<ResponseT> Request(Address address, uint64_t request_id, RequestT request, Duration timeout) {
ResponseFuture<ResponseT> Request(Address to_address, Address from_address, RequestT request, Duration timeout) {
std::function<bool()> maybe_tick_simulator = [this] { return simulator_handle_->MaybeTickSimulator(); };
auto [future, promise] =
memgraph::io::FuturePromisePairWithNotifier<ResponseResult<ResponseT>>(maybe_tick_simulator);
simulator_handle_->SubmitRequest(address, address_, request_id, std::move(request), timeout, std::move(promise));
return std::move(future);
return simulator_handle_->template SubmitRequest<RequestT, ResponseT>(to_address, from_address, std::move(request),
timeout, std::move(maybe_tick_simulator));
}
template <Message... Ms>
requires(sizeof...(Ms) > 0) RequestResult<Ms...> Receive(Duration timeout) {
return simulator_handle_->template Receive<Ms...>(address_, timeout);
requires(sizeof...(Ms) > 0) RequestResult<Ms...> Receive(Address receiver_address, Duration timeout) {
return simulator_handle_->template Receive<Ms...>(receiver_address, timeout);
}
template <Message M>
@@ -61,5 +58,7 @@ class SimulatorTransport {
Return Rand(D distrib) {
return distrib(rng_);
}
LatencyHistogramSummaries ResponseLatencies() { return simulator_handle_->ResponseLatencies(); }
};
}; // namespace memgraph::io::simulator

View File

@@ -19,6 +19,7 @@
#include "io/address.hpp"
#include "io/errors.hpp"
#include "io/future.hpp"
#include "io/message_histogram_collector.hpp"
#include "io/time.hpp"
#include "utils/result.hpp"
@@ -40,6 +41,7 @@ struct ResponseEnvelope {
RequestId request_id;
Address to_address;
Address from_address;
Duration response_latency;
};
template <Message M>
@@ -66,7 +68,6 @@ template <typename I>
class Io {
I implementation_;
Address address_;
RequestId request_id_counter_ = 0;
Duration default_timeout_ = std::chrono::microseconds{100000};
public:
@@ -82,24 +83,24 @@ class Io {
/// Issue a request with an explicit timeout in microseconds provided. This tends to be used by clients.
template <Message RequestT, Message ResponseT>
ResponseFuture<ResponseT> RequestWithTimeout(Address address, RequestT request, Duration timeout) {
const RequestId request_id = ++request_id_counter_;
return implementation_.template Request<RequestT, ResponseT>(address, request_id, request, timeout);
const Address from_address = address_;
return implementation_.template Request<RequestT, ResponseT>(address, from_address, request, timeout);
}
/// Issue a request that times out after the default timeout. This tends
/// to be used by clients.
template <Message RequestT, Message ResponseT>
ResponseFuture<ResponseT> Request(Address address, RequestT request) {
const RequestId request_id = ++request_id_counter_;
ResponseFuture<ResponseT> Request(Address to_address, RequestT request) {
const Duration timeout = default_timeout_;
return implementation_.template Request<RequestT, ResponseT>(address, request_id, std::move(request), timeout);
const Address from_address = address_;
return implementation_.template Request<RequestT, ResponseT>(to_address, from_address, std::move(request), timeout);
}
/// Wait for an explicit number of microseconds for a request of one of the
/// provided types to arrive. This tends to be used by servers.
template <Message... Ms>
RequestResult<Ms...> ReceiveWithTimeout(Duration timeout) {
return implementation_.template Receive<Ms...>(timeout);
return implementation_.template Receive<Ms...>(address_, timeout);
}
/// Wait the default number of microseconds for a request of one of the
@@ -107,7 +108,7 @@ class Io {
template <Message... Ms>
requires(sizeof...(Ms) > 0) RequestResult<Ms...> Receive() {
const Duration timeout = default_timeout_;
return implementation_.template Receive<Ms...>(timeout);
return implementation_.template Receive<Ms...>(address_, timeout);
}
/// Send a message in a best-effort fashion. This is used for messaging where
@@ -134,5 +135,10 @@ class Io {
}
Address GetAddress() { return address_; }
void SetAddress(Address address) { address_ = address; }
Io<I> ForkLocal() { return Io(implementation_, address_.ForkUniqueAddress()); }
LatencyHistogramSummaries ResponseLatencies() { return implementation_.ResponseLatencies(); }
};
}; // namespace memgraph::io

View File

@@ -0,0 +1,47 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#pragma once
#include <algorithm>
#include <thread>
#include <boost/asio/ip/tcp.hpp>
#include "io/address.hpp"
#include "storage/v3/property_value.hpp"
#include "storage/v3/schemas.hpp"
namespace memgraph::machine_manager {
using memgraph::io::Address;
using memgraph::storage::v3::SchemaProperty;
using CompoundKey = std::vector<memgraph::storage::v3::PropertyValue>;
struct InitialLabelSpace {
std::string label_name;
std::vector<SchemaProperty> schema;
size_t replication_factor;
std::vector<CompoundKey> split_points;
};
struct MachineConfig {
std::vector<InitialLabelSpace> initial_label_spaces;
std::vector<Address> coordinator_addresses;
bool is_storage;
bool is_coordinator;
bool is_query_engine;
boost::asio::ip::address listen_ip;
uint16_t listen_port;
size_t shard_worker_threads = std::max(static_cast<unsigned int>(1), std::thread::hardware_concurrency());
};
} // namespace memgraph::machine_manager

View File

@@ -0,0 +1,204 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#pragma once
#include "coordinator/coordinator_rsm.hpp"
#include "coordinator/coordinator_worker.hpp"
#include "io/message_conversion.hpp"
#include "io/messages.hpp"
#include "io/rsm/rsm_client.hpp"
#include "io/time.hpp"
#include "machine_manager/machine_config.hpp"
#include "storage/v3/shard_manager.hpp"
namespace memgraph::machine_manager {
using coordinator::Coordinator;
using coordinator::CoordinatorReadRequests;
using coordinator::CoordinatorReadResponses;
using coordinator::CoordinatorRsm;
using coordinator::CoordinatorWriteRequests;
using coordinator::CoordinatorWriteResponses;
using coordinator::coordinator_worker::CoordinatorWorker;
using CoordinatorRouteMessage = coordinator::coordinator_worker::RouteMessage;
using CoordinatorQueue = coordinator::coordinator_worker::Queue;
using io::ConvertVariant;
using io::Duration;
using io::RequestId;
using io::Time;
using io::messages::CoordinatorMessages;
using io::messages::ShardManagerMessages;
using io::messages::ShardMessages;
using io::messages::StorageReadRequest;
using io::messages::StorageWriteRequest;
using io::rsm::AppendRequest;
using io::rsm::AppendResponse;
using io::rsm::ReadRequest;
using io::rsm::VoteRequest;
using io::rsm::VoteResponse;
using io::rsm::WriteRequest;
using io::rsm::WriteResponse;
using storage::v3::ShardManager;
/// The MachineManager is responsible for:
/// * starting the entire system and ensuring that high-level
/// operational requirements continue to be met
/// * acting as a machine's single caller of Io::Receive
/// * routing incoming messages from the Io interface to the
/// appropriate Coordinator or to the StorageManager
/// (it's not necessary to route anything in this layer
/// to the query engine because the query engine only
/// communicates using higher-level Futures that will
/// be filled immediately when the response comes in
/// at the lower-level transport layer.
///
/// Every storage engine has exactly one RsmEngine.
template <typename IoImpl>
class MachineManager {
io::Io<IoImpl> io_;
MachineConfig config_;
Address coordinator_address_;
CoordinatorQueue coordinator_queue_;
std::jthread coordinator_handle_;
ShardManager<IoImpl> shard_manager_;
Time next_cron_ = Time::min();
public:
// TODO initialize ShardManager with "real" coordinator addresses instead of io.GetAddress
// which is only true for single-machine config.
MachineManager(io::Io<IoImpl> io, MachineConfig config, Coordinator coordinator)
: io_(io),
config_(config),
coordinator_address_(io.GetAddress().ForkUniqueAddress()),
shard_manager_{io.ForkLocal(), config.shard_worker_threads, coordinator_address_} {
auto coordinator_io = io.ForkLocal();
coordinator_io.SetAddress(coordinator_address_);
CoordinatorWorker coordinator_worker{coordinator_io, coordinator_queue_, coordinator};
coordinator_handle_ = std::jthread([coordinator = std::move(coordinator_worker)]() mutable { coordinator.Run(); });
}
MachineManager(MachineManager &&) noexcept = default;
MachineManager &operator=(MachineManager &&) noexcept = default;
MachineManager(const MachineManager &) = delete;
MachineManager &operator=(const MachineManager &) = delete;
~MachineManager() {
if (coordinator_handle_.joinable()) {
coordinator_queue_.Push(coordinator::coordinator_worker::ShutDown{});
coordinator_handle_.join();
}
}
Address CoordinatorAddress() { return coordinator_address_; }
void Run() {
while (!io_.ShouldShutDown()) {
const auto now = io_.Now();
if (now >= next_cron_) {
next_cron_ = Cron();
}
Duration receive_timeout = std::max(next_cron_, now) - now;
// Note: this parameter pack must be kept in-sync with the ReceiveWithTimeout parameter pack below
using AllMessages =
std::variant<ReadRequest<CoordinatorReadRequests>, AppendRequest<CoordinatorWriteRequests>, AppendResponse,
WriteRequest<CoordinatorWriteRequests>, VoteRequest, VoteResponse,
WriteResponse<CoordinatorWriteResponses>, ReadRequest<StorageReadRequest>,
AppendRequest<StorageWriteRequest>, WriteRequest<StorageWriteRequest>>;
spdlog::info("MM waiting on Receive on address {}", io_.GetAddress().ToString());
// Note: this parameter pack must be kept in-sync with the AllMessages parameter pack above
auto request_result = io_.template ReceiveWithTimeout<
ReadRequest<CoordinatorReadRequests>, AppendRequest<CoordinatorWriteRequests>, AppendResponse,
WriteRequest<CoordinatorWriteRequests>, VoteRequest, VoteResponse, WriteResponse<CoordinatorWriteResponses>,
ReadRequest<StorageReadRequest>, AppendRequest<StorageWriteRequest>, WriteRequest<StorageWriteRequest>>(
receive_timeout);
if (request_result.HasError()) {
// time to do Cron
continue;
}
auto &&request_envelope = std::move(request_result.GetValue());
spdlog::info("MM got message to {}", request_envelope.to_address.ToString());
// If message is for the coordinator, cast it to subset and pass it to the coordinator
bool to_coordinator = coordinator_address_ == request_envelope.to_address;
if (to_coordinator) {
std::optional<CoordinatorMessages> conversion_attempt =
ConvertVariant<AllMessages, ReadRequest<CoordinatorReadRequests>, AppendRequest<CoordinatorWriteRequests>,
AppendResponse, WriteRequest<CoordinatorWriteRequests>, VoteRequest, VoteResponse>(
std::move(request_envelope.message));
MG_ASSERT(conversion_attempt.has_value(), "coordinator message conversion failed");
spdlog::info("got coordinator message");
CoordinatorMessages &&cm = std::move(conversion_attempt.value());
CoordinatorRouteMessage route_message{
.message = std::move(cm),
.request_id = request_envelope.request_id,
.to = request_envelope.to_address,
.from = request_envelope.from_address,
};
coordinator_queue_.Push(std::move(route_message));
continue;
}
bool to_sm = shard_manager_.GetAddress() == request_envelope.to_address;
spdlog::info("smm: {}", shard_manager_.GetAddress().ToString());
if (to_sm) {
std::optional<ShardManagerMessages> conversion_attempt =
ConvertVariant<AllMessages, WriteResponse<CoordinatorWriteResponses>>(std::move(request_envelope.message));
MG_ASSERT(conversion_attempt.has_value(), "shard manager message conversion failed");
spdlog::info("got shard manager message");
ShardManagerMessages &&smm = std::move(conversion_attempt.value());
shard_manager_.Receive(std::forward<ShardManagerMessages>(smm), request_envelope.request_id,
request_envelope.from_address);
continue;
}
// treat this as a message to a specific shard rsm and cast it accordingly
std::optional<ShardMessages> conversion_attempt =
ConvertVariant<AllMessages, ReadRequest<StorageReadRequest>, AppendRequest<StorageWriteRequest>,
AppendResponse, WriteRequest<StorageWriteRequest>, VoteRequest, VoteResponse>(
std::move(request_envelope.message));
MG_ASSERT(conversion_attempt.has_value(), "shard rsm message conversion failed for {} - incorrect message type",
request_envelope.to_address.ToString());
spdlog::info("got shard rsm message");
ShardMessages &&sm = std::move(conversion_attempt.value());
shard_manager_.Route(std::forward<ShardMessages>(sm), request_envelope.request_id, request_envelope.to_address,
request_envelope.from_address);
}
}
private:
Time Cron() {
spdlog::info("running MachineManager::Cron, address {}", io_.GetAddress().ToString());
coordinator_queue_.Push(coordinator::coordinator_worker::Cron{});
return shard_manager_.Cron();
}
};
} // namespace memgraph::machine_manager

File diff suppressed because it is too large Load Diff

View File

@@ -9,7 +9,6 @@ set(mg_query_v2_sources
${lcp_query_v2_cpp_files}
common.cpp
cypher_query_interpreter.cpp
dump.cpp
frontend/semantic/required_privileges.cpp
frontend/stripped.cpp
interpret/awesome_memgraph_functions.cpp
@@ -23,17 +22,9 @@ set(mg_query_v2_sources
plan/rewrite/index_lookup.cpp
plan/rule_based_planner.cpp
plan/variable_start_planner.cpp
procedure/mg_procedure_impl.cpp
procedure/mg_procedure_helpers.cpp
procedure/module.cpp
procedure/py_module.cpp
serialization/property_value.cpp
stream/streams.cpp
stream/sources.cpp
stream/common.cpp
trigger.cpp
trigger_context.cpp
bindings/typed_value.cpp)
bindings/typed_value.cpp
accessors.cpp)
find_package(Boost REQUIRED)
@@ -42,7 +33,7 @@ add_dependencies(mg-query-v2 generate_lcp_query_v2)
target_include_directories(mg-query-v2 PUBLIC ${CMAKE_SOURCE_DIR}/include)
target_include_directories(mg-query-v2 PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/bindings)
target_link_libraries(mg-query-v2 dl cppitertools Boost::headers)
target_link_libraries(mg-query-v2 mg-integrations-pulsar mg-integrations-kafka mg-storage-v3 mg-license mg-utils mg-kvstore mg-memory)
target_link_libraries(mg-query-v2 mg-integrations-pulsar mg-integrations-kafka mg-storage-v3 mg-license mg-utils mg-kvstore mg-memory mg-coordinator)
target_link_libraries(mg-query-v2 mg-expr)
if(NOT "${MG_PYTHON_PATH}" STREQUAL "")

View File

@@ -0,0 +1,96 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#include "query/v2/accessors.hpp"
#include "query/v2/requests.hpp"
#include "query/v2/shard_request_manager.hpp"
#include "storage/v3/id_types.hpp"
namespace memgraph::query::v2::accessors {
EdgeAccessor::EdgeAccessor(Edge edge, const msgs::ShardRequestManagerInterface *manager)
: edge(std::move(edge)), manager_(manager) {}
EdgeTypeId EdgeAccessor::EdgeType() const { return edge.type.id; }
const std::vector<std::pair<PropertyId, Value>> &EdgeAccessor::Properties() const { return edge.properties; }
Value EdgeAccessor::GetProperty(const std::string &prop_name) const {
auto prop_id = manager_->NameToProperty(prop_name);
auto it = std::find_if(edge.properties.begin(), edge.properties.end(), [&](auto &pr) { return prop_id == pr.first; });
if (it == edge.properties.end()) {
return {};
}
return it->second;
}
const Edge &EdgeAccessor::GetEdge() const { return edge; }
bool EdgeAccessor::IsCycle() const { return edge.src == edge.dst; };
VertexAccessor EdgeAccessor::To() const {
return VertexAccessor(Vertex{edge.dst}, std::vector<std::pair<PropertyId, msgs::Value>>{}, manager_);
}
VertexAccessor EdgeAccessor::From() const {
return VertexAccessor(Vertex{edge.src}, std::vector<std::pair<PropertyId, msgs::Value>>{}, manager_);
}
VertexAccessor::VertexAccessor(Vertex v, std::vector<std::pair<PropertyId, Value>> props,
const msgs::ShardRequestManagerInterface *manager)
: vertex(std::move(v)), properties(std::move(props)), manager_(manager) {}
VertexAccessor::VertexAccessor(Vertex v, std::map<PropertyId, Value> &&props,
const msgs::ShardRequestManagerInterface *manager)
: vertex(std::move(v)), manager_(manager) {
properties.reserve(props.size());
for (auto &[id, value] : props) {
properties.emplace_back(std::make_pair(id, std::move(value)));
}
}
VertexAccessor::VertexAccessor(Vertex v, const std::map<PropertyId, Value> &props,
const msgs::ShardRequestManagerInterface *manager)
: vertex(std::move(v)), manager_(manager) {
properties.reserve(props.size());
for (const auto &[id, value] : props) {
properties.emplace_back(std::make_pair(id, value));
}
}
Label VertexAccessor::PrimaryLabel() const { return vertex.id.first; }
const msgs::VertexId &VertexAccessor::Id() const { return vertex.id; }
std::vector<Label> VertexAccessor::Labels() const { return vertex.labels; }
bool VertexAccessor::HasLabel(Label &label) const {
return std::find_if(vertex.labels.begin(), vertex.labels.end(),
[label](const auto &l) { return l.id == label.id; }) != vertex.labels.end();
}
const std::vector<std::pair<PropertyId, Value>> &VertexAccessor::Properties() const { return properties; }
Value VertexAccessor::GetProperty(PropertyId prop_id) const {
auto it = std::find_if(properties.begin(), properties.end(), [&](auto &pr) { return prop_id == pr.first; });
if (it == properties.end()) {
return {};
}
return it->second;
}
// NOLINTNEXTLINE(readability-convert-member-functions-to-static)
Value VertexAccessor::GetProperty(const std::string &prop_name) const {
return GetProperty(manager_->NameToProperty(prop_name));
}
msgs::Vertex VertexAccessor::GetVertex() const { return vertex; }
} // namespace memgraph::query::v2::accessors

203
src/query/v2/accessors.hpp Normal file
View File

@@ -0,0 +1,203 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#pragma once
#include <map>
#include <optional>
#include <utility>
#include <vector>
#include "query/exceptions.hpp"
#include "query/v2/requests.hpp"
#include "storage/v3/view.hpp"
#include "utils/bound.hpp"
#include "utils/exceptions.hpp"
#include "utils/memory.hpp"
#include "utils/memory_tracker.hpp"
namespace memgraph::msgs {
class ShardRequestManagerInterface;
} // namespace memgraph::msgs
namespace memgraph::query::v2::accessors {
using Value = memgraph::msgs::Value;
using Edge = memgraph::msgs::Edge;
using Vertex = memgraph::msgs::Vertex;
using Label = memgraph::msgs::Label;
using PropertyId = memgraph::msgs::PropertyId;
using EdgeTypeId = memgraph::msgs::EdgeTypeId;
class VertexAccessor;
class EdgeAccessor final {
public:
explicit EdgeAccessor(Edge edge, const msgs::ShardRequestManagerInterface *manager);
[[nodiscard]] EdgeTypeId EdgeType() const;
[[nodiscard]] const std::vector<std::pair<PropertyId, Value>> &Properties() const;
[[nodiscard]] Value GetProperty(const std::string &prop_name) const;
[[nodiscard]] const Edge &GetEdge() const;
[[nodiscard]] bool IsCycle() const;
// Dummy function
// NOLINTNEXTLINE(readability-convert-member-functions-to-static)
[[nodiscard]] size_t CypherId() const { return 10; }
// bool HasSrcAccessor const { return src == nullptr; }
// bool HasDstAccessor const { return dst == nullptr; }
[[nodiscard]] VertexAccessor To() const;
[[nodiscard]] VertexAccessor From() const;
friend bool operator==(const EdgeAccessor &lhs, const EdgeAccessor &rhs) { return lhs.edge == rhs.edge; }
friend bool operator!=(const EdgeAccessor &lhs, const EdgeAccessor &rhs) { return !(lhs == rhs); }
private:
Edge edge;
const msgs::ShardRequestManagerInterface *manager_;
};
class VertexAccessor final {
public:
using PropertyId = msgs::PropertyId;
using Label = msgs::Label;
using VertexId = msgs::VertexId;
VertexAccessor(Vertex v, std::vector<std::pair<PropertyId, Value>> props,
const msgs::ShardRequestManagerInterface *manager);
VertexAccessor(Vertex v, std::map<PropertyId, Value> &&props, const msgs::ShardRequestManagerInterface *manager);
VertexAccessor(Vertex v, const std::map<PropertyId, Value> &props, const msgs::ShardRequestManagerInterface *manager);
[[nodiscard]] Label PrimaryLabel() const;
[[nodiscard]] const msgs::VertexId &Id() const;
[[nodiscard]] std::vector<Label> Labels() const;
[[nodiscard]] bool HasLabel(Label &label) const;
[[nodiscard]] const std::vector<std::pair<PropertyId, Value>> &Properties() const;
[[nodiscard]] Value GetProperty(PropertyId prop_id) const;
[[nodiscard]] Value GetProperty(const std::string &prop_name) const;
[[nodiscard]] msgs::Vertex GetVertex() const;
// Dummy function
// NOLINTNEXTLINE(readability-convert-member-functions-to-static)
[[nodiscard]] size_t CypherId() const { return 10; }
// auto InEdges(storage::View view, const std::vector<storage::EdgeTypeId> &edge_types) const
// -> storage::Result<decltype(iter::imap(MakeEdgeAccessor, *impl_.InEdges(view)))> {
// auto maybe_edges = impl_.InEdges(view, edge_types);
// if (maybe_edges.HasError()) return maybe_edges.GetError();
// return iter::imap(MakeEdgeAccessor, std::move(*maybe_edges));
// }
//
// auto InEdges(storage::View view) const { return InEdges(view, {}); }
//
// auto InEdges(storage::View view, const std::vector<storage::EdgeTypeId> &edge_types, const VertexAccessor &dest)
// const
// -> storage::Result<decltype(iter::imap(MakeEdgeAccessor, *impl_.InEdges(view)))> {
// auto maybe_edges = impl_.InEdges(view, edge_types, &dest.impl_);
// if (maybe_edges.HasError()) return maybe_edges.GetError();
// return iter::imap(MakeEdgeAccessor, std::move(*maybe_edges));
// }
//
// auto OutEdges(storage::View view, const std::vector<storage::EdgeTypeId> &edge_types) const
// -> storage::Result<decltype(iter::imap(MakeEdgeAccessor, *impl_.OutEdges(view)))> {
// auto maybe_edges = impl_.OutEdges(view, edge_types);
// if (maybe_edges.HasError()) return maybe_edges.GetError();
// return iter::imap(MakeEdgeAccessor, std::move(*maybe_edges));
// }
//
// auto OutEdges(storage::View view) const { return OutEdges(view, {}); }
//
// auto OutEdges(storage::View view, const std::vector<storage::EdgeTypeId> &edge_types,
// const VertexAccessor &dest) const
// -> storage::Result<decltype(iter::imap(MakeEdgeAccessor, *impl_.OutEdges(view)))> {
// auto maybe_edges = impl_.OutEdges(view, edge_types, &dest.impl_);
// if (maybe_edges.HasError()) return maybe_edges.GetError();
// return iter::imap(MakeEdgeAccessor, std::move(*maybe_edges));
// }
// storage::Result<size_t> InDegree(storage::View view) const { return impl_.InDegree(view); }
//
// storage::Result<size_t> OutDegree(storage::View view) const { return impl_.OutDegree(view); }
//
friend bool operator==(const VertexAccessor &lhs, const VertexAccessor &rhs) {
return lhs.vertex == rhs.vertex && lhs.properties == rhs.properties;
}
friend bool operator!=(const VertexAccessor &lhs, const VertexAccessor &rhs) { return !(lhs == rhs); }
private:
Vertex vertex;
std::vector<std::pair<PropertyId, Value>> properties;
const msgs::ShardRequestManagerInterface *manager_;
};
// inline VertexAccessor EdgeAccessor::To() const { return VertexAccessor(impl_.ToVertex()); }
// inline VertexAccessor EdgeAccessor::From() const { return VertexAccessor(impl_.FromVertex()); }
// Highly mocked interface. Won't work if used.
class Path {
public:
// Empty for now
explicit Path(const VertexAccessor & /*vertex*/, utils::MemoryResource *memory = utils::NewDeleteResource())
: mem(memory) {}
template <typename... TOthers>
explicit Path(const VertexAccessor &vertex, const TOthers &...others) {}
template <typename... TOthers>
Path(std::allocator_arg_t /*unused*/, utils::MemoryResource *memory, const VertexAccessor &vertex,
const TOthers &...others) {}
Path(const Path & /*other*/) {}
Path(const Path & /*other*/, utils::MemoryResource *memory) : mem(memory) {}
Path(Path && /*other*/) noexcept {}
Path(Path && /*other*/, utils::MemoryResource *memory) : mem(memory) {}
Path &operator=(const Path &path) {
if (this == &path) {
return *this;
}
return *this;
}
Path &operator=(Path &&path) noexcept {
if (this == &path) {
return *this;
}
return *this;
}
~Path() {}
friend bool operator==(const Path & /*lhs*/, const Path & /*rhs*/) { return true; };
utils::MemoryResource *GetMemoryResource() { return mem; }
private:
utils::MemoryResource *mem = utils::NewDeleteResource();
};
} // namespace memgraph::query::v2::accessors

View File

@@ -11,5 +11,9 @@
#pragma once
#ifdef MG_AST_INCLUDE_PATH
#error You are probably trying to include files from expr from both the storage and query engines! You will have a rough time kid!
#endif
#define MG_AST_INCLUDE_PATH "query/v2/frontend/ast/ast.hpp" // NOLINT(cppcoreguidelines-macro-usage)
#define MG_INJECTED_NAMESPACE_NAME memgraph::query::v2 // NOLINT(cppcoreguidelines-macro-usage)

View File

@@ -16,19 +16,34 @@
#include "expr/interpret/eval.hpp"
#include "query/v2/bindings/typed_value.hpp"
#include "query/v2/context.hpp"
#include "query/v2/conversions.hpp"
#include "query/v2/db_accessor.hpp"
#include "storage/v3/id_types.hpp"
#include "storage/v3/property_store.hpp"
#include "storage/v3/view.hpp"
#include "query/v2/requests.hpp"
#include "storage/v3/conversions.hpp"
#include "storage/v3/property_value.hpp"
#include "storage/v3/view.hpp"
namespace memgraph::msgs {
class ShardRequestManagerInterface;
} // namespace memgraph::msgs
namespace memgraph::query::v2 {
inline const auto lam = [](const auto &val) { return memgraph::storage::v3::PropertyToTypedValue<TypedValue>(val); };
inline const auto lam = [](const auto &val) { return ValueToTypedValue(val); };
namespace detail {
class Callable {
public:
auto operator()(const memgraph::storage::v3::PropertyValue &val) const {
return memgraph::storage::v3::PropertyToTypedValue<TypedValue>(val);
};
auto operator()(const msgs::Value &val, memgraph::msgs::ShardRequestManagerInterface *manager) const {
return ValueToTypedValue(val, manager);
};
};
using ExpressionEvaluator =
memgraph::expr::ExpressionEvaluator<TypedValue, EvaluationContext, DbAccessor, storage::v3::View,
storage::v3::LabelId, storage::v3::PropertyStore, decltype(lam),
memgraph::storage::v3::Error>;
} // namespace detail
using ExpressionEvaluator = memgraph::expr::ExpressionEvaluator<
TypedValue, memgraph::query::v2::EvaluationContext, memgraph::msgs::ShardRequestManagerInterface, storage::v3::View,
storage::v3::LabelId, msgs::Value, detail::Callable, memgraph::storage::v3::Error, memgraph::expr::QueryEngineTag>;
} // namespace memgraph::query::v2

View File

@@ -13,9 +13,15 @@
#include "query/v2/bindings/bindings.hpp"
#include "query/v2/bindings/typed_value.hpp"
#include "expr/interpret/frame.hpp"
#include "query/v2/bindings/typed_value.hpp"
#include "utils/pmr/vector.hpp"
namespace memgraph::query::v2 {
using Frame = memgraph::expr::Frame<TypedValue>;
} // namespace memgraph::query::v2
struct MultiFrame {
utils::pmr::vector<Frame> frames = utils::pmr::vector<Frame>(0, Frame{1}, utils::NewDeleteResource());
size_t valid_frames{0};
};
} // namespace memgraph::query::v2

View File

@@ -10,10 +10,10 @@
// licenses/APL.txt.
#include "expr/typed_value.hpp"
#include "query/v2/db_accessor.hpp"
#include "query/v2/accessors.hpp"
#include "query/v2/path.hpp"
namespace memgraph::expr {
namespace v2 = memgraph::query::v2;
template class TypedValueT<v2::VertexAccessor, v2::EdgeAccessor, v2::Path>;
template class TypedValueT<v2::accessors::VertexAccessor, v2::accessors::EdgeAccessor, v2::accessors::Path>;
} // namespace memgraph::expr

View File

@@ -14,13 +14,14 @@
#include "query/v2/bindings/bindings.hpp"
#include "expr/typed_value.hpp"
#include "query/v2/db_accessor.hpp"
#include "query/v2/path.hpp"
#include "query/v2/accessors.hpp"
namespace memgraph::expr {
namespace v2 = memgraph::query::v2;
extern template class memgraph::expr::TypedValueT<v2::VertexAccessor, v2::EdgeAccessor, v2::Path>;
extern template class memgraph::expr::TypedValueT<v2::accessors::VertexAccessor, v2::accessors::EdgeAccessor,
v2::accessors::Path>;
} // namespace memgraph::expr
namespace memgraph::query::v2 {
using TypedValue = memgraph::expr::TypedValueT<VertexAccessor, EdgeAccessor, Path>;
using TypedValue =
memgraph::expr::TypedValueT<v2::accessors::VertexAccessor, v2::accessors::EdgeAccessor, v2::accessors::Path>;
} // namespace memgraph::query::v2

View File

@@ -28,7 +28,7 @@
#include "storage/v3/id_types.hpp"
#include "storage/v3/property_value.hpp"
#include "storage/v3/result.hpp"
#include "storage/v3/schema_validator.hpp"
#include "storage/v3/shard_operation_result.hpp"
#include "storage/v3/view.hpp"
#include "utils/exceptions.hpp"
#include "utils/logging.hpp"
@@ -93,48 +93,13 @@ concept AccessorWithSetPropertyAndValidate = requires(T accessor, const storage:
const storage::v3::PropertyValue new_value) {
{
accessor.SetPropertyAndValidate(key, new_value)
} -> std::same_as<storage::v3::ResultSchema<storage::v3::PropertyValue>>;
} -> std::same_as<storage::v3::ShardOperationResult<storage::v3::PropertyValue>>;
};
template <typename TRecordAccessor>
concept RecordAccessor =
AccessorWithSetProperty<TRecordAccessor> || AccessorWithSetPropertyAndValidate<TRecordAccessor>;
inline void HandleSchemaViolation(const storage::v3::SchemaViolation &schema_violation, const DbAccessor &dba) {
switch (schema_violation.status) {
case storage::v3::SchemaViolation::ValidationStatus::VERTEX_HAS_NO_PRIMARY_PROPERTY: {
throw SchemaViolationException(
fmt::format("Primary key {} not defined on label :{}",
storage::v3::SchemaTypeToString(schema_violation.violated_schema_property->type),
dba.LabelToName(schema_violation.label)));
}
case storage::v3::SchemaViolation::ValidationStatus::NO_SCHEMA_DEFINED_FOR_LABEL: {
throw SchemaViolationException(
fmt::format("Label :{} is not a primary label", dba.LabelToName(schema_violation.label)));
}
case storage::v3::SchemaViolation::ValidationStatus::VERTEX_PROPERTY_WRONG_TYPE: {
throw SchemaViolationException(
fmt::format("Wrong type of property {} in schema :{}, should be of type {}",
*schema_violation.violated_property_value, dba.LabelToName(schema_violation.label),
storage::v3::SchemaTypeToString(schema_violation.violated_schema_property->type)));
}
case storage::v3::SchemaViolation::ValidationStatus::VERTEX_UPDATE_PRIMARY_KEY: {
throw SchemaViolationException(fmt::format("Updating of primary key {} on schema :{} not supported",
*schema_violation.violated_property_value,
dba.LabelToName(schema_violation.label)));
}
case storage::v3::SchemaViolation::ValidationStatus::VERTEX_UPDATE_PRIMARY_LABEL: {
throw SchemaViolationException(fmt::format(
"Adding primary label as secondary or removing primary label:", *schema_violation.violated_property_value,
dba.LabelToName(schema_violation.label)));
}
case storage::v3::SchemaViolation::ValidationStatus::VERTEX_SECONDARY_LABEL_IS_PRIMARY: {
throw SchemaViolationException(fmt::format("Cannot create vertex where primary label is secondary:{}",
dba.LabelToName(schema_violation.label)));
}
}
}
inline void HandleErrorOnPropertyUpdate(const storage::v3::Error error) {
switch (error) {
case storage::v3::Error::SERIALIZATION_ERROR:
@@ -145,37 +110,9 @@ inline void HandleErrorOnPropertyUpdate(const storage::v3::Error error) {
throw QueryRuntimeException("Can't set property because properties on edges are disabled.");
case storage::v3::Error::VERTEX_HAS_EDGES:
case storage::v3::Error::NONEXISTENT_OBJECT:
throw QueryRuntimeException("Unexpected error when setting a property.");
}
}
case storage::v3::Error::VERTEX_ALREADY_INSERTED:
/// Set a property `value` mapped with given `key` on a `record`.
///
/// @throw QueryRuntimeException if value cannot be set as a property value
template <RecordAccessor T>
storage::v3::PropertyValue PropsSetChecked(T *record, const DbAccessor &dba, const storage::v3::PropertyId &key,
const TypedValue &value) {
try {
if constexpr (std::is_same_v<T, VertexAccessor>) {
const auto maybe_old_value = record->SetPropertyAndValidate(key, storage::v3::TypedToPropertyValue(value));
if (maybe_old_value.HasError()) {
std::visit(utils::Overloaded{[](const storage::v3::Error error) { HandleErrorOnPropertyUpdate(error); },
[&dba](const storage::v3::SchemaViolation &schema_violation) {
HandleSchemaViolation(schema_violation, dba);
}},
maybe_old_value.GetError());
}
return std::move(*maybe_old_value);
} else {
// No validation on edge properties
const auto maybe_old_value = record->SetProperty(key, storage::v3::TypedToPropertyValue(value));
if (maybe_old_value.HasError()) {
HandleErrorOnPropertyUpdate(maybe_old_value.GetError());
}
return std::move(*maybe_old_value);
}
} catch (const expr::TypedValueException &) {
throw QueryRuntimeException("'{}' cannot be used as a property value.", value.type());
throw QueryRuntimeException("Unexpected error when setting a property.");
}
}

View File

@@ -13,16 +13,35 @@
#include <type_traits>
#include "io/local_transport/local_transport.hpp"
#include "query/v2/bindings/symbol_table.hpp"
#include "query/v2/common.hpp"
#include "query/v2/metadata.hpp"
#include "query/v2/parameters.hpp"
#include "query/v2/plan/profile.hpp"
#include "query/v2/trigger.hpp"
//#include "query/v2/trigger.hpp"
#include "query/v2/shard_request_manager.hpp"
#include "utils/async_timer.hpp"
namespace memgraph::query::v2 {
// Used to store range of ids that are available
// Used for edge id assignment
class IdAllocator {
public:
IdAllocator() = default;
IdAllocator(uint64_t low, uint64_t high) : current_edge_id_{low}, max_id_{high} {};
uint64_t AllocateId() {
MG_ASSERT(current_edge_id_ < max_id_, "Current Edge Id went above max id");
return current_edge_id_++;
}
private:
uint64_t current_edge_id_;
uint64_t max_id_;
};
struct EvaluationContext {
/// Memory for allocations during evaluation of a *single* Pull call.
///
@@ -41,21 +60,28 @@ struct EvaluationContext {
mutable std::unordered_map<std::string, int64_t> counters;
};
inline std::vector<storage::v3::PropertyId> NamesToProperties(const std::vector<std::string> &property_names,
DbAccessor *dba) {
inline std::vector<storage::v3::PropertyId> NamesToProperties(
const std::vector<std::string> &property_names, msgs::ShardRequestManagerInterface *shard_request_manager) {
std::vector<storage::v3::PropertyId> properties;
// TODO Fix by using reference
properties.reserve(property_names.size());
for (const auto &name : property_names) {
properties.push_back(dba->NameToProperty(name));
if (shard_request_manager != nullptr) {
for (const auto &name : property_names) {
properties.push_back(shard_request_manager->NameToProperty(name));
}
}
return properties;
}
inline std::vector<storage::v3::LabelId> NamesToLabels(const std::vector<std::string> &label_names, DbAccessor *dba) {
inline std::vector<storage::v3::LabelId> NamesToLabels(const std::vector<std::string> &label_names,
msgs::ShardRequestManagerInterface *shard_request_manager) {
std::vector<storage::v3::LabelId> labels;
labels.reserve(label_names.size());
for (const auto &name : label_names) {
labels.push_back(dba->NameToLabel(name));
// TODO Fix by using reference
if (shard_request_manager != nullptr) {
for (const auto &name : label_names) {
labels.push_back(shard_request_manager->NameToLabel(name));
}
}
return labels;
}
@@ -70,8 +96,9 @@ struct ExecutionContext {
plan::ProfilingStats stats;
plan::ProfilingStats *stats_root{nullptr};
ExecutionStats execution_stats;
TriggerContextCollector *trigger_context_collector{nullptr};
utils::AsyncTimer timer;
msgs::ShardRequestManagerInterface *shard_request_manager{nullptr};
IdAllocator *edge_ids_alloc;
};
static_assert(std::is_move_assignable_v<ExecutionContext>, "ExecutionContext must be move assignable!");

View File

@@ -0,0 +1,103 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#pragma once
#include "bindings/typed_value.hpp"
#include "query/v2/accessors.hpp"
#include "query/v2/requests.hpp"
#include "query/v2/shard_request_manager.hpp"
namespace memgraph::query::v2 {
inline TypedValue ValueToTypedValue(const msgs::Value &value, msgs::ShardRequestManagerInterface *manager) {
using Value = msgs::Value;
switch (value.type) {
case Value::Type::Null:
return {};
case Value::Type::Bool:
return TypedValue(value.bool_v);
case Value::Type::Int64:
return TypedValue(value.int_v);
case Value::Type::Double:
return TypedValue(value.double_v);
case Value::Type::String:
return TypedValue(value.string_v);
case Value::Type::List: {
const auto &lst = value.list_v;
std::vector<TypedValue> dst;
dst.reserve(lst.size());
for (const auto &elem : lst) {
dst.push_back(ValueToTypedValue(elem, manager));
}
return TypedValue(std::move(dst));
}
case Value::Type::Map: {
const auto &value_map = value.map_v;
std::map<std::string, TypedValue> dst;
for (const auto &[key, val] : value_map) {
dst[key] = ValueToTypedValue(val, manager);
}
return TypedValue(std::move(dst));
}
case Value::Type::Vertex:
return TypedValue(accessors::VertexAccessor(
value.vertex_v, std::vector<std::pair<storage::v3::PropertyId, msgs::Value>>{}, manager));
case Value::Type::Edge:
return TypedValue(accessors::EdgeAccessor(value.edge_v, manager));
}
throw std::runtime_error("Incorrect type in conversion");
}
inline msgs::Value TypedValueToValue(const TypedValue &value) {
using Value = msgs::Value;
switch (value.type()) {
case TypedValue::Type::Null:
return {};
case TypedValue::Type::Bool:
return Value(value.ValueBool());
case TypedValue::Type::Int:
return Value(value.ValueInt());
case TypedValue::Type::Double:
return Value(value.ValueDouble());
case TypedValue::Type::String:
return Value(std::string(value.ValueString()));
case TypedValue::Type::List: {
const auto &lst = value.ValueList();
std::vector<Value> dst;
dst.reserve(lst.size());
for (const auto &elem : lst) {
dst.push_back(TypedValueToValue(elem));
}
return Value(std::move(dst));
}
case TypedValue::Type::Map: {
const auto &value_map = value.ValueMap();
std::map<std::string, Value> dst;
for (const auto &[key, val] : value_map) {
dst[std::string(key)] = TypedValueToValue(val);
}
return Value(std::move(dst));
}
case TypedValue::Type::Vertex:
return Value(value.ValueVertex().GetVertex());
case TypedValue::Type::Edge:
return Value(value.ValueEdge().GetEdge());
case TypedValue::Type::Path:
case TypedValue::Type::LocalTime:
case TypedValue::Type::LocalDateTime:
case TypedValue::Type::Date:
case TypedValue::Type::Duration:
break;
}
throw std::runtime_error("Incorrect type in conversion");
}
} // namespace memgraph::query::v2

View File

@@ -11,6 +11,7 @@
#include "query/v2/cypher_query_interpreter.hpp"
#include "query/v2/bindings/symbol_generator.hpp"
#include "query/v2/shard_request_manager.hpp"
// NOLINTNEXTLINE (cppcoreguidelines-avoid-non-const-global-variables)
DEFINE_HIDDEN_bool(query_cost_planner, true, "Use the cost-estimating query planner.");
@@ -117,9 +118,9 @@ ParsedQuery ParseQuery(const std::string &query_string, const std::map<std::stri
}
std::unique_ptr<LogicalPlan> MakeLogicalPlan(AstStorage ast_storage, CypherQuery *query, const Parameters &parameters,
DbAccessor *db_accessor,
msgs::ShardRequestManagerInterface *shard_manager,
const std::vector<Identifier *> &predefined_identifiers) {
auto vertex_counts = plan::MakeVertexCountCache(db_accessor);
auto vertex_counts = plan::MakeVertexCountCache(shard_manager);
auto symbol_table = expr::MakeSymbolTable(query, predefined_identifiers);
auto planning_context = plan::MakePlanningContext(&ast_storage, &symbol_table, query, &vertex_counts);
auto [root, cost] = plan::MakeLogicalPlan(&planning_context, parameters, FLAGS_query_cost_planner);
@@ -129,7 +130,7 @@ std::unique_ptr<LogicalPlan> MakeLogicalPlan(AstStorage ast_storage, CypherQuery
std::shared_ptr<CachedPlan> CypherQueryToPlan(uint64_t hash, AstStorage ast_storage, CypherQuery *query,
const Parameters &parameters, utils::SkipList<PlanCacheEntry> *plan_cache,
DbAccessor *db_accessor,
msgs::ShardRequestManagerInterface *shard_manager,
const std::vector<Identifier *> &predefined_identifiers) {
std::optional<utils::SkipList<PlanCacheEntry>::Accessor> plan_cache_access;
if (plan_cache) {
@@ -145,7 +146,7 @@ std::shared_ptr<CachedPlan> CypherQueryToPlan(uint64_t hash, AstStorage ast_stor
}
auto plan = std::make_shared<CachedPlan>(
MakeLogicalPlan(std::move(ast_storage), query, parameters, db_accessor, predefined_identifiers));
MakeLogicalPlan(std::move(ast_storage), query, parameters, shard_manager, predefined_identifiers));
if (plan_cache_access) {
plan_cache_access->insert({hash, plan});
}

View File

@@ -132,7 +132,7 @@ class SingleNodeLogicalPlan final : public LogicalPlan {
};
std::unique_ptr<LogicalPlan> MakeLogicalPlan(AstStorage ast_storage, CypherQuery *query, const Parameters &parameters,
DbAccessor *db_accessor,
msgs::ShardRequestManagerInterface *shard_manager,
const std::vector<Identifier *> &predefined_identifiers);
/**
@@ -145,7 +145,7 @@ std::unique_ptr<LogicalPlan> MakeLogicalPlan(AstStorage ast_storage, CypherQuery
*/
std::shared_ptr<CachedPlan> CypherQueryToPlan(uint64_t hash, AstStorage ast_storage, CypherQuery *query,
const Parameters &parameters, utils::SkipList<PlanCacheEntry> *plan_cache,
DbAccessor *db_accessor,
msgs::ShardRequestManagerInterface *shard_manager,
const std::vector<Identifier *> &predefined_identifiers = {});
} // namespace memgraph::query::v2

View File

@@ -23,6 +23,7 @@
#include "storage/v3/key_store.hpp"
#include "storage/v3/property_value.hpp"
#include "storage/v3/result.hpp"
#include "storage/v3/shard_operation_result.hpp"
///////////////////////////////////////////////////////////
// Our communication layer and query engine don't mix
@@ -113,17 +114,19 @@ class VertexAccessor final {
auto PrimaryKey(storage::v3::View view) const { return impl_.PrimaryKey(view); }
storage::v3::ResultSchema<bool> AddLabel(storage::v3::LabelId label) { return impl_.AddLabelAndValidate(label); }
storage::v3::ResultSchema<bool> AddLabelAndValidate(storage::v3::LabelId label) {
storage::v3::ShardOperationResult<bool> AddLabel(storage::v3::LabelId label) {
return impl_.AddLabelAndValidate(label);
}
storage::v3::ResultSchema<bool> RemoveLabel(storage::v3::LabelId label) {
storage::v3::ShardOperationResult<bool> AddLabelAndValidate(storage::v3::LabelId label) {
return impl_.AddLabelAndValidate(label);
}
storage::v3::ShardOperationResult<bool> RemoveLabel(storage::v3::LabelId label) {
return impl_.RemoveLabelAndValidate(label);
}
storage::v3::ResultSchema<bool> RemoveLabelAndValidate(storage::v3::LabelId label) {
storage::v3::ShardOperationResult<bool> RemoveLabelAndValidate(storage::v3::LabelId label) {
return impl_.RemoveLabelAndValidate(label);
}
@@ -138,17 +141,17 @@ class VertexAccessor final {
return impl_.GetProperty(key, view);
}
storage::v3::ResultSchema<storage::v3::PropertyValue> SetProperty(storage::v3::PropertyId key,
const storage::v3::PropertyValue &value) {
storage::v3::ShardOperationResult<storage::v3::PropertyValue> SetProperty(storage::v3::PropertyId key,
const storage::v3::PropertyValue &value) {
return impl_.SetPropertyAndValidate(key, value);
}
storage::v3::ResultSchema<storage::v3::PropertyValue> SetPropertyAndValidate(
storage::v3::ShardOperationResult<storage::v3::PropertyValue> SetPropertyAndValidate(
storage::v3::PropertyId key, const storage::v3::PropertyValue &value) {
return impl_.SetPropertyAndValidate(key, value);
}
storage::v3::ResultSchema<storage::v3::PropertyValue> RemovePropertyAndValidate(storage::v3::PropertyId key) {
storage::v3::ShardOperationResult<storage::v3::PropertyValue> RemovePropertyAndValidate(storage::v3::PropertyId key) {
return SetPropertyAndValidate(key, storage::v3::PropertyValue{});
}
@@ -219,203 +222,7 @@ inline VertexAccessor EdgeAccessor::From() const { return *static_cast<VertexAcc
inline bool EdgeAccessor::IsCycle() const { return To() == From(); }
class DbAccessor final {
storage::v3::Shard::Accessor *accessor_;
class VerticesIterable final {
storage::v3::VerticesIterable iterable_;
public:
class Iterator final {
storage::v3::VerticesIterable::Iterator it_;
public:
explicit Iterator(storage::v3::VerticesIterable::Iterator it) : it_(it) {}
VertexAccessor operator*() const { return VertexAccessor(*it_); }
Iterator &operator++() {
++it_;
return *this;
}
bool operator==(const Iterator &other) const { return it_ == other.it_; }
bool operator!=(const Iterator &other) const { return !(other == *this); }
};
explicit VerticesIterable(storage::v3::VerticesIterable iterable) : iterable_(std::move(iterable)) {}
Iterator begin() { return Iterator(iterable_.begin()); }
Iterator end() { return Iterator(iterable_.end()); }
};
public:
explicit DbAccessor(storage::v3::Shard::Accessor *accessor) : accessor_(accessor) {}
// TODO(jbajic) Fix Remove Gid
// NOLINTNEXTLINE(readability-convert-member-functions-to-static)
std::optional<VertexAccessor> FindVertex(uint64_t /*unused*/) { return std::nullopt; }
std::optional<VertexAccessor> FindVertex(storage::v3::PrimaryKey &primary_key, storage::v3::View view) {
auto maybe_vertex = accessor_->FindVertex(primary_key, view);
if (maybe_vertex) return VertexAccessor(*maybe_vertex);
return std::nullopt;
}
void FinalizeTransaction() { accessor_->FinalizeTransaction(); }
VerticesIterable Vertices(storage::v3::View view) { return VerticesIterable(accessor_->Vertices(view)); }
VerticesIterable Vertices(storage::v3::View view, storage::v3::LabelId label) {
return VerticesIterable(accessor_->Vertices(label, view));
}
VerticesIterable Vertices(storage::v3::View view, storage::v3::LabelId label, storage::v3::PropertyId property) {
return VerticesIterable(accessor_->Vertices(label, property, view));
}
VerticesIterable Vertices(storage::v3::View view, storage::v3::LabelId label, storage::v3::PropertyId property,
const storage::v3::PropertyValue &value) {
return VerticesIterable(accessor_->Vertices(label, property, value, view));
}
VerticesIterable Vertices(storage::v3::View view, storage::v3::LabelId label, storage::v3::PropertyId property,
const std::optional<utils::Bound<storage::v3::PropertyValue>> &lower,
const std::optional<utils::Bound<storage::v3::PropertyValue>> &upper) {
return VerticesIterable(accessor_->Vertices(label, property, lower, upper, view));
}
storage::v3::ResultSchema<VertexAccessor> InsertVertexAndValidate(
const storage::v3::LabelId primary_label, const std::vector<storage::v3::LabelId> &labels,
const std::vector<std::pair<storage::v3::PropertyId, storage::v3::PropertyValue>> &properties) {
auto maybe_vertex_acc = accessor_->CreateVertexAndValidate(primary_label, labels, properties);
if (maybe_vertex_acc.HasError()) {
return {std::move(maybe_vertex_acc.GetError())};
}
return VertexAccessor{maybe_vertex_acc.GetValue()};
}
storage::v3::Result<EdgeAccessor> InsertEdge(VertexAccessor *from, VertexAccessor *to,
const storage::v3::EdgeTypeId &edge_type) {
static constexpr auto kDummyGid = storage::v3::Gid::FromUint(0);
auto maybe_edge = accessor_->CreateEdge(from->impl_.Id(storage::v3::View::NEW).GetValue(),
to->impl_.Id(storage::v3::View::NEW).GetValue(), edge_type, kDummyGid);
if (maybe_edge.HasError()) return storage::v3::Result<EdgeAccessor>(maybe_edge.GetError());
return EdgeAccessor(*maybe_edge);
}
storage::v3::Result<std::optional<EdgeAccessor>> RemoveEdge(EdgeAccessor *edge) {
auto res = accessor_->DeleteEdge(edge->impl_.FromVertex(), edge->impl_.ToVertex(), edge->impl_.Gid());
if (res.HasError()) {
return res.GetError();
}
const auto &value = res.GetValue();
if (!value) {
return std::optional<EdgeAccessor>{};
}
return std::make_optional<EdgeAccessor>(*value);
}
storage::v3::Result<std::optional<std::pair<VertexAccessor, std::vector<EdgeAccessor>>>> DetachRemoveVertex(
VertexAccessor *vertex_accessor) {
using ReturnType = std::pair<VertexAccessor, std::vector<EdgeAccessor>>;
auto res = accessor_->DetachDeleteVertex(&vertex_accessor->impl_);
if (res.HasError()) {
return res.GetError();
}
const auto &value = res.GetValue();
if (!value) {
return std::optional<ReturnType>{};
}
const auto &[vertex, edges] = *value;
std::vector<EdgeAccessor> deleted_edges;
deleted_edges.reserve(edges.size());
std::transform(edges.begin(), edges.end(), std::back_inserter(deleted_edges),
[](const auto &deleted_edge) { return EdgeAccessor{deleted_edge}; });
return std::make_optional<ReturnType>(vertex, std::move(deleted_edges));
}
storage::v3::Result<std::optional<VertexAccessor>> RemoveVertex(VertexAccessor *vertex_accessor) {
auto res = accessor_->DeleteVertex(&vertex_accessor->impl_);
if (res.HasError()) {
return res.GetError();
}
const auto &value = res.GetValue();
if (!value) {
return std::optional<VertexAccessor>{};
}
return {std::make_optional<VertexAccessor>(*value)};
}
// TODO(jbajic) Query engine should have a map of labels, properties and edge
// types
// NOLINTNEXTLINE(readability-convert-member-functions-to-static)
storage::v3::PropertyId NameToProperty(const std::string_view /*name*/) {
return storage::v3::PropertyId::FromUint(0);
}
// NOLINTNEXTLINE(readability-convert-member-functions-to-static)
storage::v3::LabelId NameToLabel(const std::string_view /*name*/) { return storage::v3::LabelId::FromUint(0); }
// NOLINTNEXTLINE(readability-convert-member-functions-to-static)
storage::v3::EdgeTypeId NameToEdgeType(const std::string_view /*name*/) {
return storage::v3::EdgeTypeId::FromUint(0);
}
const std::string &PropertyToName(storage::v3::PropertyId prop) const { return accessor_->PropertyToName(prop); }
const std::string &LabelToName(storage::v3::LabelId label) const { return accessor_->LabelToName(label); }
const std::string &EdgeTypeToName(storage::v3::EdgeTypeId type) const { return accessor_->EdgeTypeToName(type); }
void AdvanceCommand() { accessor_->AdvanceCommand(); }
utils::BasicResult<storage::v3::ConstraintViolation, void> Commit() { return accessor_->Commit(); }
void Abort() { accessor_->Abort(); }
bool LabelIndexExists(storage::v3::LabelId label) const { return accessor_->LabelIndexExists(label); }
bool LabelPropertyIndexExists(storage::v3::LabelId label, storage::v3::PropertyId prop) const {
return accessor_->LabelPropertyIndexExists(label, prop);
}
int64_t VerticesCount() const { return accessor_->ApproximateVertexCount(); }
int64_t VerticesCount(storage::v3::LabelId label) const { return accessor_->ApproximateVertexCount(label); }
int64_t VerticesCount(storage::v3::LabelId label, storage::v3::PropertyId property) const {
return accessor_->ApproximateVertexCount(label, property);
}
int64_t VerticesCount(storage::v3::LabelId label, storage::v3::PropertyId property,
const storage::v3::PropertyValue &value) const {
return accessor_->ApproximateVertexCount(label, property, value);
}
int64_t VerticesCount(storage::v3::LabelId label, storage::v3::PropertyId property,
const std::optional<utils::Bound<storage::v3::PropertyValue>> &lower,
const std::optional<utils::Bound<storage::v3::PropertyValue>> &upper) const {
return accessor_->ApproximateVertexCount(label, property, lower, upper);
}
storage::v3::IndicesInfo ListAllIndices() const { return accessor_->ListAllIndices(); }
storage::v3::ConstraintsInfo ListAllConstraints() const { return accessor_->ListAllConstraints(); }
const storage::v3::SchemaValidator &GetSchemaValidator() const { return accessor_->GetSchemaValidator(); }
storage::v3::SchemasInfo ListAllSchemas() const { return accessor_->ListAllSchemas(); }
};
} // namespace memgraph::query::v2

View File

@@ -1,541 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#include "query/v2/dump.hpp"
#include <iomanip>
#include <limits>
#include <map>
#include <optional>
#include <ostream>
#include <utility>
#include <vector>
#include <fmt/format.h>
#include "query/v2/bindings/typed_value.hpp"
#include "query/v2/db_accessor.hpp"
#include "query/v2/exceptions.hpp"
#include "query/v2/stream.hpp"
#include "storage/v3/property_value.hpp"
#include "storage/v3/storage.hpp"
#include "utils/algorithm.hpp"
#include "utils/logging.hpp"
#include "utils/string.hpp"
#include "utils/temporal.hpp"
namespace memgraph::query::v2 {
namespace {
// Property that is used to make a difference among vertices. It is added to
// property set of vertices to match edges and removed after the entire graph
// is built.
const char *kInternalPropertyId = "__mg_id__";
// Label that is attached to each vertex and is used for easier creation of
// index on internal property id.
const char *kInternalVertexLabel = "__mg_vertex__";
/// A helper function that escapes label, edge type and property names.
std::string EscapeName(const std::string_view value) {
std::string out;
out.reserve(value.size() + 2);
out.append(1, '`');
for (auto c : value) {
if (c == '`') {
out.append("``");
} else {
out.append(1, c);
}
}
out.append(1, '`');
return out;
}
void DumpPreciseDouble(std::ostream *os, double value) {
// A temporary stream is used to keep precision of the original output
// stream unchanged.
std::ostringstream temp_oss;
temp_oss << std::setprecision(std::numeric_limits<double>::max_digits10) << value;
*os << temp_oss.str();
}
namespace {
void DumpDate(std::ostream &os, const storage::v3::TemporalData &value) {
utils::Date date(value.microseconds);
os << "DATE(\"" << date << "\")";
}
void DumpLocalTime(std::ostream &os, const storage::v3::TemporalData &value) {
utils::LocalTime lt(value.microseconds);
os << "LOCALTIME(\"" << lt << "\")";
}
void DumpLocalDateTime(std::ostream &os, const storage::v3::TemporalData &value) {
utils::LocalDateTime ldt(value.microseconds);
os << "LOCALDATETIME(\"" << ldt << "\")";
}
void DumpDuration(std::ostream &os, const storage::v3::TemporalData &value) {
utils::Duration dur(value.microseconds);
os << "DURATION(\"" << dur << "\")";
}
void DumpTemporalData(std::ostream &os, const storage::v3::TemporalData &value) {
switch (value.type) {
case storage::v3::TemporalType::Date: {
DumpDate(os, value);
return;
}
case storage::v3::TemporalType::LocalTime: {
DumpLocalTime(os, value);
return;
}
case storage::v3::TemporalType::LocalDateTime: {
DumpLocalDateTime(os, value);
return;
}
case storage::v3::TemporalType::Duration: {
DumpDuration(os, value);
return;
}
}
}
} // namespace
void DumpPropertyValue(std::ostream *os, const storage::v3::PropertyValue &value) {
switch (value.type()) {
case storage::v3::PropertyValue::Type::Null:
*os << "Null";
return;
case storage::v3::PropertyValue::Type::Bool:
*os << (value.ValueBool() ? "true" : "false");
return;
case storage::v3::PropertyValue::Type::String:
*os << utils::Escape(value.ValueString());
return;
case storage::v3::PropertyValue::Type::Int:
*os << value.ValueInt();
return;
case storage::v3::PropertyValue::Type::Double:
DumpPreciseDouble(os, value.ValueDouble());
return;
case storage::v3::PropertyValue::Type::List: {
*os << "[";
const auto &list = value.ValueList();
utils::PrintIterable(*os, list, ", ", [](auto &os, const auto &item) { DumpPropertyValue(&os, item); });
*os << "]";
return;
}
case storage::v3::PropertyValue::Type::Map: {
*os << "{";
const auto &map = value.ValueMap();
utils::PrintIterable(*os, map, ", ", [](auto &os, const auto &kv) {
os << EscapeName(kv.first) << ": ";
DumpPropertyValue(&os, kv.second);
});
*os << "}";
return;
}
case storage::v3::PropertyValue::Type::TemporalData: {
DumpTemporalData(*os, value.ValueTemporalData());
return;
}
}
}
void DumpProperties(std::ostream *os, query::v2::DbAccessor *dba,
const std::map<storage::v3::PropertyId, storage::v3::PropertyValue> &store,
std::optional<int64_t> property_id = std::nullopt) {
*os << "{";
if (property_id) {
*os << kInternalPropertyId << ": " << *property_id;
if (store.size() > 0) *os << ", ";
}
utils::PrintIterable(*os, store, ", ", [&dba](auto &os, const auto &kv) {
os << EscapeName(dba->PropertyToName(kv.first)) << ": ";
DumpPropertyValue(&os, kv.second);
});
*os << "}";
}
void DumpVertex(std::ostream *os, query::v2::DbAccessor *dba, const query::v2::VertexAccessor &vertex) {
*os << "CREATE (";
*os << ":" << kInternalVertexLabel;
auto maybe_labels = vertex.Labels(storage::v3::View::OLD);
if (maybe_labels.HasError()) {
switch (maybe_labels.GetError()) {
case storage::v3::Error::DELETED_OBJECT:
throw query::v2::QueryRuntimeException("Trying to get labels from a deleted node.");
case storage::v3::Error::NONEXISTENT_OBJECT:
throw query::v2::QueryRuntimeException("Trying to get labels from a node that doesn't exist.");
case storage::v3::Error::SERIALIZATION_ERROR:
case storage::v3::Error::VERTEX_HAS_EDGES:
case storage::v3::Error::PROPERTIES_DISABLED:
throw query::v2::QueryRuntimeException("Unexpected error when getting labels.");
}
}
for (const auto &label : *maybe_labels) {
*os << ":" << EscapeName(dba->LabelToName(label));
}
*os << " ";
auto maybe_props = vertex.Properties(storage::v3::View::OLD);
if (maybe_props.HasError()) {
switch (maybe_props.GetError()) {
case storage::v3::Error::DELETED_OBJECT:
throw query::v2::QueryRuntimeException("Trying to get properties from a deleted object.");
case storage::v3::Error::NONEXISTENT_OBJECT:
throw query::v2::QueryRuntimeException("Trying to get properties from a node that doesn't exist.");
case storage::v3::Error::SERIALIZATION_ERROR:
case storage::v3::Error::VERTEX_HAS_EDGES:
case storage::v3::Error::PROPERTIES_DISABLED:
throw query::v2::QueryRuntimeException("Unexpected error when getting properties.");
}
}
DumpProperties(os, dba, *maybe_props, vertex.CypherId());
*os << ");";
}
void DumpEdge(std::ostream *os, query::v2::DbAccessor *dba, const query::v2::EdgeAccessor &edge) {
*os << "MATCH ";
*os << "(u:" << kInternalVertexLabel << "), ";
*os << "(v:" << kInternalVertexLabel << ")";
*os << " WHERE ";
*os << "u." << kInternalPropertyId << " = " << edge.From().CypherId();
*os << " AND ";
*os << "v." << kInternalPropertyId << " = " << edge.To().CypherId() << " ";
*os << "CREATE (u)-[";
*os << ":" << EscapeName(dba->EdgeTypeToName(edge.EdgeType()));
auto maybe_props = edge.Properties(storage::v3::View::OLD);
if (maybe_props.HasError()) {
switch (maybe_props.GetError()) {
case storage::v3::Error::DELETED_OBJECT:
throw query::v2::QueryRuntimeException("Trying to get properties from a deleted object.");
case storage::v3::Error::NONEXISTENT_OBJECT:
throw query::v2::QueryRuntimeException("Trying to get properties from an edge that doesn't exist.");
case storage::v3::Error::SERIALIZATION_ERROR:
case storage::v3::Error::VERTEX_HAS_EDGES:
case storage::v3::Error::PROPERTIES_DISABLED:
throw query::v2::QueryRuntimeException("Unexpected error when getting properties.");
}
}
if (maybe_props->size() > 0) {
*os << " ";
DumpProperties(os, dba, *maybe_props);
}
*os << "]->(v);";
}
void DumpLabelIndex(std::ostream *os, query::v2::DbAccessor *dba, const storage::v3::LabelId label) {
*os << "CREATE INDEX ON :" << EscapeName(dba->LabelToName(label)) << ";";
}
void DumpLabelPropertyIndex(std::ostream *os, query::v2::DbAccessor *dba, storage::v3::LabelId label,
storage::v3::PropertyId property) {
*os << "CREATE INDEX ON :" << EscapeName(dba->LabelToName(label)) << "(" << EscapeName(dba->PropertyToName(property))
<< ");";
}
void DumpExistenceConstraint(std::ostream *os, query::v2::DbAccessor *dba, storage::v3::LabelId label,
storage::v3::PropertyId property) {
*os << "CREATE CONSTRAINT ON (u:" << EscapeName(dba->LabelToName(label)) << ") ASSERT EXISTS (u."
<< EscapeName(dba->PropertyToName(property)) << ");";
}
void DumpUniqueConstraint(std::ostream *os, query::v2::DbAccessor *dba, storage::v3::LabelId label,
const std::set<storage::v3::PropertyId> &properties) {
*os << "CREATE CONSTRAINT ON (u:" << EscapeName(dba->LabelToName(label)) << ") ASSERT ";
utils::PrintIterable(*os, properties, ", ", [&dba](auto &stream, const auto &property) {
stream << "u." << EscapeName(dba->PropertyToName(property));
});
*os << " IS UNIQUE;";
}
} // namespace
PullPlanDump::PullPlanDump(DbAccessor *dba)
: dba_(dba),
vertices_iterable_(dba->Vertices(storage::v3::View::OLD)),
pull_chunks_{// Dump all label indices
CreateLabelIndicesPullChunk(),
// Dump all label property indices
CreateLabelPropertyIndicesPullChunk(),
// Dump all existence constraints
CreateExistenceConstraintsPullChunk(),
// Dump all unique constraints
CreateUniqueConstraintsPullChunk(),
// Create internal index for faster edge creation
CreateInternalIndexPullChunk(),
// Dump all vertices
CreateVertexPullChunk(),
// Dump all edges
CreateEdgePullChunk(),
// Drop the internal index
CreateDropInternalIndexPullChunk(),
// Internal index cleanup
CreateInternalIndexCleanupPullChunk()} {}
bool PullPlanDump::Pull(AnyStream *stream, std::optional<int> n) {
// Iterate all functions that stream some results.
// Each function should return number of results it streamed after it
// finishes. If the function did not finish streaming all the results,
// std::nullopt should be returned because n results have already been sent.
while (current_chunk_index_ < pull_chunks_.size() && (!n || *n > 0)) {
const auto maybe_streamed_count = pull_chunks_[current_chunk_index_](stream, n);
if (!maybe_streamed_count) {
// n wasn't large enough to stream all the results from the current chunk
break;
}
if (n) {
// chunk finished streaming its results
// subtract number of results streamed in current pull
// so we know how many results we need to stream from future
// chunks.
*n -= *maybe_streamed_count;
}
++current_chunk_index_;
}
return current_chunk_index_ == pull_chunks_.size();
}
PullPlanDump::PullChunk PullPlanDump::CreateLabelIndicesPullChunk() {
// Dump all label indices
return [this, global_index = 0U](AnyStream *stream, std::optional<int> n) mutable -> std::optional<size_t> {
// Delay the construction of indices vectors
if (!indices_info_) {
indices_info_.emplace(dba_->ListAllIndices());
}
const auto &label = indices_info_->label;
size_t local_counter = 0;
while (global_index < label.size() && (!n || local_counter < *n)) {
std::ostringstream os;
DumpLabelIndex(&os, dba_, label[global_index]);
stream->Result({TypedValue(os.str())});
++global_index;
++local_counter;
}
if (global_index == label.size()) {
return local_counter;
}
return std::nullopt;
};
}
PullPlanDump::PullChunk PullPlanDump::CreateLabelPropertyIndicesPullChunk() {
return [this, global_index = 0U](AnyStream *stream, std::optional<int> n) mutable -> std::optional<size_t> {
// Delay the construction of indices vectors
if (!indices_info_) {
indices_info_.emplace(dba_->ListAllIndices());
}
const auto &label_property = indices_info_->label_property;
size_t local_counter = 0;
while (global_index < label_property.size() && (!n || local_counter < *n)) {
std::ostringstream os;
const auto &label_property_index = label_property[global_index];
DumpLabelPropertyIndex(&os, dba_, label_property_index.first, label_property_index.second);
stream->Result({TypedValue(os.str())});
++global_index;
++local_counter;
}
if (global_index == label_property.size()) {
return local_counter;
}
return std::nullopt;
};
}
PullPlanDump::PullChunk PullPlanDump::CreateExistenceConstraintsPullChunk() {
return [this, global_index = 0U](AnyStream *stream, std::optional<int> n) mutable -> std::optional<size_t> {
// Delay the construction of constraint vectors
if (!constraints_info_) {
constraints_info_.emplace(dba_->ListAllConstraints());
}
const auto &existence = constraints_info_->existence;
size_t local_counter = 0;
while (global_index < existence.size() && (!n || local_counter < *n)) {
const auto &constraint = existence[global_index];
std::ostringstream os;
DumpExistenceConstraint(&os, dba_, constraint.first, constraint.second);
stream->Result({TypedValue(os.str())});
++global_index;
++local_counter;
}
if (global_index == existence.size()) {
return local_counter;
}
return std::nullopt;
};
}
PullPlanDump::PullChunk PullPlanDump::CreateUniqueConstraintsPullChunk() {
return [this, global_index = 0U](AnyStream *stream, std::optional<int> n) mutable -> std::optional<size_t> {
// Delay the construction of constraint vectors
if (!constraints_info_) {
constraints_info_.emplace(dba_->ListAllConstraints());
}
const auto &unique = constraints_info_->unique;
size_t local_counter = 0;
while (global_index < unique.size() && (!n || local_counter < *n)) {
const auto &constraint = unique[global_index];
std::ostringstream os;
DumpUniqueConstraint(&os, dba_, constraint.first, constraint.second);
stream->Result({TypedValue(os.str())});
++global_index;
++local_counter;
}
if (global_index == unique.size()) {
return local_counter;
}
return std::nullopt;
};
}
PullPlanDump::PullChunk PullPlanDump::CreateInternalIndexPullChunk() {
return [this](AnyStream *stream, std::optional<int>) mutable -> std::optional<size_t> {
if (vertices_iterable_.begin() != vertices_iterable_.end()) {
std::ostringstream os;
os << "CREATE INDEX ON :" << kInternalVertexLabel << "(" << kInternalPropertyId << ");";
stream->Result({TypedValue(os.str())});
internal_index_created_ = true;
return 1;
}
return 0;
};
}
PullPlanDump::PullChunk PullPlanDump::CreateVertexPullChunk() {
return [this, maybe_current_iter = std::optional<VertexAccessorIterableIterator>{}](
AnyStream *stream, std::optional<int> n) mutable -> std::optional<size_t> {
// Delay the call of begin() function
// If multiple begins are called before an iteration,
// one iteration will make the rest of iterators be in undefined
// states.
if (!maybe_current_iter) {
maybe_current_iter.emplace(vertices_iterable_.begin());
}
auto &current_iter{*maybe_current_iter};
size_t local_counter = 0;
while (current_iter != vertices_iterable_.end() && (!n || local_counter < *n)) {
std::ostringstream os;
DumpVertex(&os, dba_, *current_iter);
stream->Result({TypedValue(os.str())});
++local_counter;
++current_iter;
}
if (current_iter == vertices_iterable_.end()) {
return local_counter;
}
return std::nullopt;
};
}
PullPlanDump::PullChunk PullPlanDump::CreateEdgePullChunk() {
return [this, maybe_current_vertex_iter = std::optional<VertexAccessorIterableIterator>{},
// we need to save the iterable which contains list of accessor so
// our saved iterator is valid in the next run
maybe_edge_iterable = std::shared_ptr<EdgeAccessorIterable>{nullptr},
maybe_current_edge_iter = std::optional<EdgeAccessorIterableIterator>{}](
AnyStream *stream, std::optional<int> n) mutable -> std::optional<size_t> {
// Delay the call of begin() function
// If multiple begins are called before an iteration,
// one iteration will make the rest of iterators be in undefined
// states.
if (!maybe_current_vertex_iter) {
maybe_current_vertex_iter.emplace(vertices_iterable_.begin());
}
auto &current_vertex_iter{*maybe_current_vertex_iter};
size_t local_counter = 0U;
for (; current_vertex_iter != vertices_iterable_.end() && (!n || local_counter < *n); ++current_vertex_iter) {
const auto &vertex = *current_vertex_iter;
// If we have a saved iterable from a previous pull
// we need to use the same iterable
if (!maybe_edge_iterable) {
maybe_edge_iterable = std::make_shared<EdgeAccessorIterable>(vertex.OutEdges(storage::v3::View::OLD));
}
auto &maybe_edges = *maybe_edge_iterable;
MG_ASSERT(maybe_edges.HasValue(), "Invalid database state!");
auto current_edge_iter = maybe_current_edge_iter ? *maybe_current_edge_iter : maybe_edges->begin();
for (; current_edge_iter != maybe_edges->end() && (!n || local_counter < *n); ++current_edge_iter) {
std::ostringstream os;
DumpEdge(&os, dba_, *current_edge_iter);
stream->Result({TypedValue(os.str())});
++local_counter;
}
if (current_edge_iter != maybe_edges->end()) {
maybe_current_edge_iter.emplace(current_edge_iter);
return std::nullopt;
}
maybe_current_edge_iter = std::nullopt;
maybe_edge_iterable = nullptr;
}
if (current_vertex_iter == vertices_iterable_.end()) {
return local_counter;
}
return std::nullopt;
};
}
PullPlanDump::PullChunk PullPlanDump::CreateDropInternalIndexPullChunk() {
return [this](AnyStream *stream, std::optional<int>) {
if (internal_index_created_) {
std::ostringstream os;
os << "DROP INDEX ON :" << kInternalVertexLabel << "(" << kInternalPropertyId << ");";
stream->Result({TypedValue(os.str())});
return 1;
}
return 0;
};
}
PullPlanDump::PullChunk PullPlanDump::CreateInternalIndexCleanupPullChunk() {
return [this](AnyStream *stream, std::optional<int>) {
if (internal_index_created_) {
std::ostringstream os;
os << "MATCH (u) REMOVE u:" << kInternalVertexLabel << ", u." << kInternalPropertyId << ";";
stream->Result({TypedValue(os.str())});
return 1;
}
return 0;
};
}
void DumpDatabaseToCypherQueries(query::v2::DbAccessor *dba, AnyStream *stream) { PullPlanDump(dba).Pull(stream, {}); }
} // namespace memgraph::query::v2

View File

@@ -1,66 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#pragma once
#include <ostream>
#include "query/v2/db_accessor.hpp"
#include "query/v2/stream.hpp"
#include "storage/v3/storage.hpp"
namespace memgraph::query::v2 {
void DumpDatabaseToCypherQueries(query::v2::DbAccessor *dba, AnyStream *stream);
struct PullPlanDump {
explicit PullPlanDump(query::v2::DbAccessor *dba);
/// Pull the dump results lazily
/// @return true if all results were returned, false otherwise
bool Pull(AnyStream *stream, std::optional<int> n);
private:
query::v2::DbAccessor *dba_ = nullptr;
std::optional<storage::v3::IndicesInfo> indices_info_ = std::nullopt;
std::optional<storage::v3::ConstraintsInfo> constraints_info_ = std::nullopt;
using VertexAccessorIterable = decltype(std::declval<query::v2::DbAccessor>().Vertices(storage::v3::View::OLD));
using VertexAccessorIterableIterator = decltype(std::declval<VertexAccessorIterable>().begin());
using EdgeAccessorIterable = decltype(std::declval<VertexAccessor>().OutEdges(storage::v3::View::OLD));
using EdgeAccessorIterableIterator = decltype(std::declval<EdgeAccessorIterable>().GetValue().begin());
VertexAccessorIterable vertices_iterable_;
bool internal_index_created_ = false;
size_t current_chunk_index_ = 0;
using PullChunk = std::function<std::optional<size_t>(AnyStream *stream, std::optional<int> n)>;
// We define every part of the dump query in a self contained function.
// Each functions is responsible of keeping track of its execution status.
// If a function did finish its execution, it should return number of results
// it streamed so we know how many rows should be pulled from the next
// function, otherwise std::nullopt is returned.
std::vector<PullChunk> pull_chunks_;
PullChunk CreateLabelIndicesPullChunk();
PullChunk CreateLabelPropertyIndicesPullChunk();
PullChunk CreateExistenceConstraintsPullChunk();
PullChunk CreateUniqueConstraintsPullChunk();
PullChunk CreateInternalIndexPullChunk();
PullChunk CreateVertexPullChunk();
PullChunk CreateEdgePullChunk();
PullChunk CreateDropInternalIndexPullChunk();
PullChunk CreateInternalIndexCleanupPullChunk();
};
} // namespace memgraph::query::v2

View File

@@ -18,6 +18,7 @@
#include <vector>
#include "query/v2/bindings/ast_visitor.hpp"
#include "common/types.hpp"
#include "query/v2/bindings/symbol.hpp"
#include "query/v2/interpret/awesome_memgraph_functions.hpp"
#include "query/v2/bindings/typed_value.hpp"

View File

@@ -11,7 +11,6 @@
#include "query/v2/bindings/ast_visitor.hpp"
#include "query/v2/frontend/ast/ast.hpp"
#include "query/v2/procedure/module.hpp"
#include "utils/memory.hpp"
namespace memgraph::query::v2 {
@@ -86,12 +85,7 @@ class PrivilegeExtractor : public QueryVisitor<void>, public HierarchicalTreeVis
AddPrivilege(AuthQuery::Privilege::CREATE);
return false;
}
bool PreVisit(CallProcedure &procedure) override {
const auto maybe_proc =
procedure::FindProcedure(procedure::gModuleRegistry, procedure.procedure_name_, utils::NewDeleteResource());
if (maybe_proc && maybe_proc->second->info.required_privilege) {
AddPrivilege(*maybe_proc->second->info.required_privilege);
}
bool PreVisit(CallProcedure & /*procedure*/) override {
return false;
}
bool PreVisit(Delete & /*unused*/) override {

View File

@@ -21,11 +21,9 @@
#include <type_traits>
#include "query/v2/bindings/typed_value.hpp"
#include "query/v2/db_accessor.hpp"
#include "query/v2/conversions.hpp"
#include "query/v2/exceptions.hpp"
#include "query/v2/procedure/cypher_types.hpp"
#include "query/v2/procedure/mg_procedure_impl.hpp"
#include "query/v2/procedure/module.hpp"
#include "query/v2/shard_request_manager.hpp"
#include "storage/v3/conversions.hpp"
#include "utils/string.hpp"
#include "utils/temporal.hpp"
@@ -387,39 +385,7 @@ TypedValue Last(const TypedValue *args, int64_t nargs, const FunctionContext &ct
return TypedValue(list.back(), ctx.memory);
}
TypedValue Properties(const TypedValue *args, int64_t nargs, const FunctionContext &ctx) {
FType<Or<Null, Vertex, Edge>>("properties", args, nargs);
auto *dba = ctx.db_accessor;
auto get_properties = [&](const auto &record_accessor) {
TypedValue::TMap properties(ctx.memory);
auto maybe_props = record_accessor.Properties(ctx.view);
if (maybe_props.HasError()) {
switch (maybe_props.GetError()) {
case storage::v3::Error::DELETED_OBJECT:
throw QueryRuntimeException("Trying to get properties from a deleted object.");
case storage::v3::Error::NONEXISTENT_OBJECT:
throw query::v2::QueryRuntimeException("Trying to get properties from an object that doesn't exist.");
case storage::v3::Error::SERIALIZATION_ERROR:
case storage::v3::Error::VERTEX_HAS_EDGES:
case storage::v3::Error::PROPERTIES_DISABLED:
throw QueryRuntimeException("Unexpected error when getting properties.");
}
}
for (const auto &property : *maybe_props) {
properties.emplace(dba->PropertyToName(property.first),
storage::v3::PropertyToTypedValue<TypedValue>(property.second));
}
return TypedValue(std::move(properties));
};
const auto &value = args[0];
if (value.IsNull()) {
return TypedValue(ctx.memory);
} else if (value.IsVertex()) {
return get_properties(value.ValueVertex());
} else {
return get_properties(value.ValueEdge());
}
}
TypedValue Properties(const TypedValue * /*args*/, int64_t /*nargs*/, const FunctionContext & /*ctx*/) { return {}; }
TypedValue Size(const TypedValue *args, int64_t nargs, const FunctionContext &ctx) {
FType<Or<Null, List, String, Map, Path>>("size", args, nargs);
@@ -435,7 +401,8 @@ TypedValue Size(const TypedValue *args, int64_t nargs, const FunctionContext &ct
// to do it.
return TypedValue(static_cast<int64_t>(value.ValueMap().size()), ctx.memory);
} else {
return TypedValue(static_cast<int64_t>(value.ValuePath().edges().size()), ctx.memory);
// TODO(kostasrim) Fix the dummy return
return TypedValue(int64_t(0), ctx.memory);
}
}
@@ -445,51 +412,6 @@ TypedValue StartNode(const TypedValue *args, int64_t nargs, const FunctionContex
return TypedValue(args[0].ValueEdge().From(), ctx.memory);
}
namespace {
size_t UnwrapDegreeResult(storage::v3::Result<size_t> maybe_degree) {
if (maybe_degree.HasError()) {
switch (maybe_degree.GetError()) {
case storage::v3::Error::DELETED_OBJECT:
throw QueryRuntimeException("Trying to get degree of a deleted node.");
case storage::v3::Error::NONEXISTENT_OBJECT:
throw query::v2::QueryRuntimeException("Trying to get degree of a node that doesn't exist.");
case storage::v3::Error::SERIALIZATION_ERROR:
case storage::v3::Error::VERTEX_HAS_EDGES:
case storage::v3::Error::PROPERTIES_DISABLED:
throw QueryRuntimeException("Unexpected error when getting node degree.");
}
}
return *maybe_degree;
}
} // namespace
TypedValue Degree(const TypedValue *args, int64_t nargs, const FunctionContext &ctx) {
FType<Or<Null, Vertex>>("degree", args, nargs);
if (args[0].IsNull()) return TypedValue(ctx.memory);
const auto &vertex = args[0].ValueVertex();
size_t out_degree = UnwrapDegreeResult(vertex.OutDegree(ctx.view));
size_t in_degree = UnwrapDegreeResult(vertex.InDegree(ctx.view));
return TypedValue(static_cast<int64_t>(out_degree + in_degree), ctx.memory);
}
TypedValue InDegree(const TypedValue *args, int64_t nargs, const FunctionContext &ctx) {
FType<Or<Null, Vertex>>("inDegree", args, nargs);
if (args[0].IsNull()) return TypedValue(ctx.memory);
const auto &vertex = args[0].ValueVertex();
size_t in_degree = UnwrapDegreeResult(vertex.InDegree(ctx.view));
return TypedValue(static_cast<int64_t>(in_degree), ctx.memory);
}
TypedValue OutDegree(const TypedValue *args, int64_t nargs, const FunctionContext &ctx) {
FType<Or<Null, Vertex>>("outDegree", args, nargs);
if (args[0].IsNull()) return TypedValue(ctx.memory);
const auto &vertex = args[0].ValueVertex();
size_t out_degree = UnwrapDegreeResult(vertex.OutDegree(ctx.view));
return TypedValue(static_cast<int64_t>(out_degree), ctx.memory);
}
TypedValue ToBoolean(const TypedValue *args, int64_t nargs, const FunctionContext &ctx) {
FType<Or<Null, Bool, Integer, String>>("toBoolean", args, nargs);
const auto &value = args[0];
@@ -551,9 +473,8 @@ TypedValue ToInteger(const TypedValue *args, int64_t nargs, const FunctionContex
TypedValue Type(const TypedValue *args, int64_t nargs, const FunctionContext &ctx) {
FType<Or<Null, Edge>>("type", args, nargs);
auto *dba = ctx.db_accessor;
if (args[0].IsNull()) return TypedValue(ctx.memory);
return TypedValue(dba->EdgeTypeToName(args[0].ValueEdge().EdgeType()), ctx.memory);
return TypedValue(args[0].ValueEdge().EdgeType().AsInt(), ctx.memory);
}
TypedValue ValueType(const TypedValue *args, int64_t nargs, const FunctionContext &ctx) {
@@ -592,84 +513,6 @@ TypedValue ValueType(const TypedValue *args, int64_t nargs, const FunctionContex
}
}
// TODO: How is Keys different from Properties function?
TypedValue Keys(const TypedValue *args, int64_t nargs, const FunctionContext &ctx) {
FType<Or<Null, Vertex, Edge>>("keys", args, nargs);
auto *dba = ctx.db_accessor;
auto get_keys = [&](const auto &record_accessor) {
TypedValue::TVector keys(ctx.memory);
auto maybe_props = record_accessor.Properties(ctx.view);
if (maybe_props.HasError()) {
switch (maybe_props.GetError()) {
case storage::v3::Error::DELETED_OBJECT:
throw QueryRuntimeException("Trying to get keys from a deleted object.");
case storage::v3::Error::NONEXISTENT_OBJECT:
throw query::v2::QueryRuntimeException("Trying to get keys from an object that doesn't exist.");
case storage::v3::Error::SERIALIZATION_ERROR:
case storage::v3::Error::VERTEX_HAS_EDGES:
case storage::v3::Error::PROPERTIES_DISABLED:
throw QueryRuntimeException("Unexpected error when getting keys.");
}
}
for (const auto &property : *maybe_props) {
keys.emplace_back(dba->PropertyToName(property.first));
}
return TypedValue(std::move(keys));
};
const auto &value = args[0];
if (value.IsNull()) {
return TypedValue(ctx.memory);
} else if (value.IsVertex()) {
return get_keys(value.ValueVertex());
} else {
return get_keys(value.ValueEdge());
}
}
TypedValue Labels(const TypedValue *args, int64_t nargs, const FunctionContext &ctx) {
FType<Or<Null, Vertex>>("labels", args, nargs);
auto *dba = ctx.db_accessor;
if (args[0].IsNull()) return TypedValue(ctx.memory);
TypedValue::TVector labels(ctx.memory);
auto maybe_labels = args[0].ValueVertex().Labels(ctx.view);
if (maybe_labels.HasError()) {
switch (maybe_labels.GetError()) {
case storage::v3::Error::DELETED_OBJECT:
throw QueryRuntimeException("Trying to get labels from a deleted node.");
case storage::v3::Error::NONEXISTENT_OBJECT:
throw query::v2::QueryRuntimeException("Trying to get labels from a node that doesn't exist.");
case storage::v3::Error::SERIALIZATION_ERROR:
case storage::v3::Error::VERTEX_HAS_EDGES:
case storage::v3::Error::PROPERTIES_DISABLED:
throw QueryRuntimeException("Unexpected error when getting labels.");
}
}
for (const auto &label : *maybe_labels) {
labels.emplace_back(dba->LabelToName(label));
}
return TypedValue(std::move(labels));
}
TypedValue Nodes(const TypedValue *args, int64_t nargs, const FunctionContext &ctx) {
FType<Or<Null, Path>>("nodes", args, nargs);
if (args[0].IsNull()) return TypedValue(ctx.memory);
const auto &vertices = args[0].ValuePath().vertices();
TypedValue::TVector values(ctx.memory);
values.reserve(vertices.size());
for (const auto &v : vertices) values.emplace_back(v);
return TypedValue(std::move(values));
}
TypedValue Relationships(const TypedValue *args, int64_t nargs, const FunctionContext &ctx) {
FType<Or<Null, Path>>("relationships", args, nargs);
if (args[0].IsNull()) return TypedValue(ctx.memory);
const auto &edges = args[0].ValuePath().edges();
TypedValue::TVector values(ctx.memory);
values.reserve(edges.size());
for (const auto &e : edges) values.emplace_back(e);
return TypedValue(std::move(values));
}
TypedValue Range(const TypedValue *args, int64_t nargs, const FunctionContext &ctx) {
FType<Or<Null, Integer>, Or<Null, Integer>, Optional<Or<Null, NonZeroInteger>>>("range", args, nargs);
for (int64_t i = 0; i < nargs; ++i)
@@ -877,9 +720,9 @@ TypedValue Id(const TypedValue *args, int64_t nargs, const FunctionContext &ctx)
if (arg.IsNull()) {
return TypedValue(ctx.memory);
} else if (arg.IsVertex()) {
return TypedValue(arg.ValueVertex().CypherId(), ctx.memory);
return TypedValue(int64_t(arg.ValueVertex().CypherId()), ctx.memory);
} else {
return TypedValue(arg.ValueEdge().CypherId(), ctx.memory);
return TypedValue(int64_t(arg.ValueEdge().CypherId()), ctx.memory);
}
}
@@ -1180,57 +1023,11 @@ TypedValue Duration(const TypedValue *args, int64_t nargs, const FunctionContext
return TypedValue(utils::Duration(duration_parameters), ctx.memory);
}
std::function<TypedValue(const TypedValue *, const int64_t, const FunctionContext &)> UserFunction(
const mgp_func &func, const std::string &fully_qualified_name) {
return [func, fully_qualified_name](const TypedValue *args, int64_t nargs, const FunctionContext &ctx) -> TypedValue {
/// Find function is called to aquire the lock on Module pointer while user-defined function is executed
const auto &maybe_found =
procedure::FindFunction(procedure::gModuleRegistry, fully_qualified_name, utils::NewDeleteResource());
if (!maybe_found) {
throw QueryRuntimeException(
"Function '{}' has been unloaded. Please check query modules to confirm that function is loaded in Memgraph.",
fully_qualified_name);
}
const auto &func_cb = func.cb;
mgp_memory memory{ctx.memory};
mgp_func_context functx{ctx.db_accessor, ctx.view};
auto graph = mgp_graph::NonWritableGraph(*ctx.db_accessor, ctx.view);
std::vector<TypedValue> args_list;
args_list.reserve(nargs);
for (std::size_t i = 0; i < nargs; ++i) {
args_list.emplace_back(args[i]);
}
auto function_argument_list = mgp_list(ctx.memory);
procedure::ConstructArguments(args_list, func, fully_qualified_name, function_argument_list, graph);
mgp_func_result maybe_res;
func_cb(&function_argument_list, &functx, &maybe_res, &memory);
if (maybe_res.error_msg) {
throw QueryRuntimeException(*maybe_res.error_msg);
}
if (!maybe_res.value) {
throw QueryRuntimeException(
"Function '{}' didn't set the result nor the error message. Please either set the result by using "
"mgp_func_result_set_value or the error by using mgp_func_result_set_error_msg.",
fully_qualified_name);
}
return {*(maybe_res.value), ctx.memory};
};
}
} // namespace
std::function<TypedValue(const TypedValue *, int64_t, const FunctionContext &ctx)> NameToFunction(
const std::string &function_name) {
// Scalar functions
if (function_name == "DEGREE") return Degree;
if (function_name == "INDEGREE") return InDegree;
if (function_name == "OUTDEGREE") return OutDegree;
if (function_name == "ENDNODE") return EndNode;
if (function_name == "HEAD") return Head;
if (function_name == kId) return Id;
@@ -1246,11 +1043,7 @@ std::function<TypedValue(const TypedValue *, int64_t, const FunctionContext &ctx
if (function_name == "VALUETYPE") return ValueType;
// List functions
if (function_name == "KEYS") return Keys;
if (function_name == "LABELS") return Labels;
if (function_name == "NODES") return Nodes;
if (function_name == "RANGE") return Range;
if (function_name == "RELATIONSHIPS") return Relationships;
if (function_name == "TAIL") return Tail;
if (function_name == "UNIFORMSAMPLE") return UniformSample;
@@ -1309,14 +1102,6 @@ std::function<TypedValue(const TypedValue *, int64_t, const FunctionContext &ctx
if (function_name == "LOCALDATETIME") return LocalDateTime;
if (function_name == "DURATION") return Duration;
const auto &maybe_found =
procedure::FindFunction(procedure::gModuleRegistry, function_name, utils::NewDeleteResource());
if (maybe_found) {
const auto *func = (*maybe_found).second;
return UserFunction(*func, function_name);
}
return nullptr;
}

View File

@@ -16,12 +16,15 @@
#include <unordered_map>
#include "query/v2/bindings/typed_value.hpp"
#include "query/v2/db_accessor.hpp"
#include "storage/v3/view.hpp"
#include "utils/memory.hpp"
namespace memgraph::query::v2 {
namespace memgraph::msgs {
class ShardRequestManagerInterface;
} // namespace memgraph::msgs
class DbAccessor;
namespace memgraph::query::v2 {
namespace {
const char kStartsWith[] = "STARTSWITH";
@@ -31,7 +34,9 @@ const char kId[] = "ID";
} // namespace
struct FunctionContext {
DbAccessor *db_accessor;
// TODO(kostasrim) consider optional here. ShardRequestManager does not exist on the storage.
// DbAccessor *db_accessor;
msgs::ShardRequestManagerInterface *manager;
utils::MemoryResource *memory;
int64_t timestamp;
std::unordered_map<std::string, int64_t> *counters;

File diff suppressed because it is too large Load Diff

View File

@@ -12,7 +12,12 @@
#pragma once
#include <gflags/gflags.h>
#include <cstdint>
#include "coordinator/coordinator.hpp"
#include "coordinator/coordinator_client.hpp"
#include "io/local_transport/local_transport.hpp"
#include "io/transport.hpp"
#include "query/v2/auth_checker.hpp"
#include "query/v2/bindings/cypher_main_visitor.hpp"
#include "query/v2/bindings/typed_value.hpp"
@@ -26,9 +31,8 @@
#include "query/v2/metadata.hpp"
#include "query/v2/plan/operator.hpp"
#include "query/v2/plan/read_write_type_checker.hpp"
#include "query/v2/requests.hpp"
#include "query/v2/stream.hpp"
#include "query/v2/stream/streams.hpp"
#include "query/v2/trigger.hpp"
#include "storage/v3/isolation_level.hpp"
#include "storage/v3/name_id_mapper.hpp"
#include "utils/event_counter.hpp"
@@ -167,7 +171,8 @@ struct PreparedQuery {
*/
struct InterpreterContext {
explicit InterpreterContext(storage::v3::Shard *db, InterpreterConfig config,
const std::filesystem::path &data_directory);
const std::filesystem::path &data_directory,
io::Io<io::local_transport::LocalTransport> io, coordinator::Address coordinator_addr);
storage::v3::Shard *db;
@@ -180,12 +185,13 @@ struct InterpreterContext {
utils::SkipList<QueryCacheEntry> ast_cache;
utils::SkipList<PlanCacheEntry> plan_cache;
TriggerStore trigger_store;
utils::ThreadPool after_commit_trigger_pool{1};
const InterpreterConfig config;
IdAllocator edge_ids_alloc;
query::v2::stream::Streams streams;
// TODO (antaljanosbenjamin) Figure out an abstraction for io::Io to make it possible to construct an interpreter
// context with a simulator transport without templatizing it.
io::Io<io::local_transport::LocalTransport> io;
coordinator::Address coordinator_address;
storage::v3::LabelId NameToLabelId(std::string_view label_name) {
return storage::v3::LabelId::FromUint(query_id_mapper.NameToId(label_name));
@@ -290,6 +296,8 @@ class Interpreter final {
*/
void Abort();
const msgs::ShardRequestManagerInterface *GetShardRequestManager() const { return shard_request_manager_.get(); }
private:
struct QueryExecution {
std::optional<PreparedQuery> prepared_query;
@@ -331,10 +339,10 @@ class Interpreter final {
// This cannot be std::optional because we need to move this accessor later on into a lambda capture
// which is assigned to std::function. std::function requires every object to be copyable, so we
// move this unique_ptr into a shrared_ptr.
// move this unique_ptr into a shared_ptr.
std::unique_ptr<storage::v3::Shard::Accessor> db_accessor_;
std::optional<DbAccessor> execution_db_accessor_;
std::optional<TriggerContextCollector> trigger_context_collector_;
std::unique_ptr<msgs::ShardRequestManagerInterface> shard_request_manager_;
bool in_explicit_transaction_{false};
bool expect_rollback_{false};

View File

@@ -12,8 +12,8 @@
#pragma once
#include <algorithm>
#include <unordered_map>
#include <utility>
#include <vector>
#include "storage/v3/property_value.hpp"
#include "utils/logging.hpp"
@@ -32,7 +32,7 @@ struct Parameters {
* @param position Token position in query of value.
* @param value
*/
void Add(int position, const storage::v3::PropertyValue &value) { storage_.emplace_back(position, value); }
void Add(int position, const storage::v3::PropertyValue &value) { storage_.emplace(position, value); }
/**
* Returns the value found for the given token position.
@@ -41,23 +41,11 @@ struct Parameters {
* @return Value for the given token position.
*/
const storage::v3::PropertyValue &AtTokenPosition(int position) const {
auto found = std::find_if(storage_.begin(), storage_.end(), [&](const auto &a) { return a.first == position; });
auto found = storage_.find(position);
MG_ASSERT(found != storage_.end(), "Token position must be present in container");
return found->second;
}
/**
* Returns the position-th stripped value. Asserts that this
* container has at least (position + 1) elements.
*
* @param position Which stripped param is sought.
* @return Token position and value for sought param.
*/
const std::pair<int, storage::v3::PropertyValue> &At(int position) const {
MG_ASSERT(position < static_cast<int>(storage_.size()), "Invalid position");
return storage_[position];
}
/** Returns the number of arguments in this container */
auto size() const { return storage_.size(); }
@@ -65,7 +53,7 @@ struct Parameters {
auto end() const { return storage_.end(); }
private:
std::vector<std::pair<int, storage::v3::PropertyValue>> storage_;
std::unordered_map<int, storage::v3::PropertyValue> storage_;
};
} // namespace memgraph::query::v2

View File

@@ -14,6 +14,8 @@
#include <functional>
#include <utility>
#include "query/db_accessor.hpp"
#include "query/v2/accessors.hpp"
#include "query/v2/db_accessor.hpp"
#include "utils/logging.hpp"
#include "utils/memory.hpp"
@@ -28,6 +30,8 @@ namespace memgraph::query::v2 {
*/
class Path {
public:
using VertexAccessor = accessors::VertexAccessor;
using EdgeAccessor = accessors::EdgeAccessor;
/** Allocator type so that STL containers are aware that we need one */
using allocator_type = utils::Allocator<char>;

File diff suppressed because it is too large Load Diff

View File

@@ -22,18 +22,23 @@
#include <variant>
#include <vector>
#include "expr/semantic/symbol.hpp"
#include "query/v2/common.hpp"
#include "query/v2/frontend/ast/ast.hpp"
#include "expr/semantic/symbol.hpp"
#include "query/v2//bindings/typed_value.hpp"
#include "query/v2//bindings/frame.hpp"
#include "query/v2//bindings/symbol_table.hpp"
#include "query/v2/bindings/typed_value.hpp"
#include "query/v2/bindings/frame.hpp"
#include "query/v2/bindings/symbol_table.hpp"
#include "storage/v3/id_types.hpp"
#include "utils/bound.hpp"
#include "utils/fnv.hpp"
#include "utils/memory.hpp"
#include "utils/visitor.hpp"
#include "utils/logging.hpp"
#include "query/v2/accessors.hpp"
using VertexAccessor = memgraph::query::v2::accessors::VertexAccessor;
using EdgeAccessor = memgraph::query::v2::accessors::EdgeAccessor;
using Path = memgraph::query::v2::accessors::Path;
cpp<#
(lcp:namespace memgraph)
@@ -66,6 +71,8 @@ class Cursor {
/// @throws QueryRuntimeException if something went wrong with execution
virtual bool Pull(Frame &, ExecutionContext &) = 0;
virtual void PullMultiple(MultiFrame &, ExecutionContext &) { LOG_FATAL("PullMultipleIsNotImplemented"); }
/// Resets the Cursor to its initial state.
virtual void Reset() = 0;
@@ -328,11 +335,12 @@ and false on every following Pull.")
public:
OnceCursor() {}
bool Pull(Frame &, ExecutionContext &) override;
void PullMultiple(MultiFrame &, ExecutionContext &) override;
void Shutdown() override;
void Reset() override;
private:
bool did_pull_{false};
size_t pull_count_{false};
};
cpp<#)
(:serialize (:slk))
@@ -1202,6 +1210,7 @@ RETURN clause) the Produce's pull succeeds exactly once.")
public:
ProduceCursor(const Produce &, utils::MemoryResource *);
bool Pull(Frame &, ExecutionContext &) override;
void PullMultiple(MultiFrame &, ExecutionContext &) override;
void Shutdown() override;
void Reset() override;

View File

@@ -14,11 +14,13 @@
#include "query/v2/bindings/pretty_print.hpp"
#include "query/v2/db_accessor.hpp"
#include "query/v2/shard_request_manager.hpp"
#include "utils/string.hpp"
namespace memgraph::query::v2::plan {
PlanPrinter::PlanPrinter(const DbAccessor *dba, std::ostream *out) : dba_(dba), out_(out) {}
PlanPrinter::PlanPrinter(const msgs::ShardRequestManagerInterface *request_manager, std::ostream *out)
: request_manager_(request_manager), out_(out) {}
#define PRE_VISIT(TOp) \
bool PlanPrinter::PreVisit(TOp &) { \
@@ -32,7 +34,7 @@ bool PlanPrinter::PreVisit(CreateExpand &op) {
WithPrintLn([&](auto &out) {
out << "* CreateExpand (" << op.input_symbol_.name() << ")"
<< (op.edge_info_.direction == query::v2::EdgeAtom::Direction::IN ? "<-" : "-") << "["
<< op.edge_info_.symbol.name() << ":" << dba_->EdgeTypeToName(op.edge_info_.edge_type) << "]"
<< op.edge_info_.symbol.name() << ":" << request_manager_->EdgeTypeToName(op.edge_info_.edge_type) << "]"
<< (op.edge_info_.direction == query::v2::EdgeAtom::Direction::OUT ? "->" : "-") << "("
<< op.node_info_.symbol.name() << ")";
});
@@ -52,7 +54,7 @@ bool PlanPrinter::PreVisit(query::v2::plan::ScanAll &op) {
bool PlanPrinter::PreVisit(query::v2::plan::ScanAllByLabel &op) {
WithPrintLn([&](auto &out) {
out << "* ScanAllByLabel"
<< " (" << op.output_symbol_.name() << " :" << dba_->LabelToName(op.label_) << ")";
<< " (" << op.output_symbol_.name() << " :" << request_manager_->LabelToName(op.label_) << ")";
});
return true;
}
@@ -60,8 +62,8 @@ bool PlanPrinter::PreVisit(query::v2::plan::ScanAllByLabel &op) {
bool PlanPrinter::PreVisit(query::v2::plan::ScanAllByLabelPropertyValue &op) {
WithPrintLn([&](auto &out) {
out << "* ScanAllByLabelPropertyValue"
<< " (" << op.output_symbol_.name() << " :" << dba_->LabelToName(op.label_) << " {"
<< dba_->PropertyToName(op.property_) << "})";
<< " (" << op.output_symbol_.name() << " :" << request_manager_->LabelToName(op.label_) << " {"
<< request_manager_->PropertyToName(op.property_) << "})";
});
return true;
}
@@ -69,8 +71,8 @@ bool PlanPrinter::PreVisit(query::v2::plan::ScanAllByLabelPropertyValue &op) {
bool PlanPrinter::PreVisit(query::v2::plan::ScanAllByLabelPropertyRange &op) {
WithPrintLn([&](auto &out) {
out << "* ScanAllByLabelPropertyRange"
<< " (" << op.output_symbol_.name() << " :" << dba_->LabelToName(op.label_) << " {"
<< dba_->PropertyToName(op.property_) << "})";
<< " (" << op.output_symbol_.name() << " :" << request_manager_->LabelToName(op.label_) << " {"
<< request_manager_->PropertyToName(op.property_) << "})";
});
return true;
}
@@ -78,8 +80,8 @@ bool PlanPrinter::PreVisit(query::v2::plan::ScanAllByLabelPropertyRange &op) {
bool PlanPrinter::PreVisit(query::v2::plan::ScanAllByLabelProperty &op) {
WithPrintLn([&](auto &out) {
out << "* ScanAllByLabelProperty"
<< " (" << op.output_symbol_.name() << " :" << dba_->LabelToName(op.label_) << " {"
<< dba_->PropertyToName(op.property_) << "})";
<< " (" << op.output_symbol_.name() << " :" << request_manager_->LabelToName(op.label_) << " {"
<< request_manager_->PropertyToName(op.property_) << "})";
});
return true;
}
@@ -98,7 +100,7 @@ bool PlanPrinter::PreVisit(query::v2::plan::Expand &op) {
<< (op.common_.direction == query::v2::EdgeAtom::Direction::IN ? "<-" : "-") << "["
<< op.common_.edge_symbol.name();
utils::PrintIterable(*out_, op.common_.edge_types, "|", [this](auto &stream, const auto &edge_type) {
stream << ":" << dba_->EdgeTypeToName(edge_type);
stream << ":" << request_manager_->EdgeTypeToName(edge_type);
});
*out_ << "]" << (op.common_.direction == query::v2::EdgeAtom::Direction::OUT ? "->" : "-") << "("
<< op.common_.node_symbol.name() << ")";
@@ -127,7 +129,7 @@ bool PlanPrinter::PreVisit(query::v2::plan::ExpandVariable &op) {
<< (op.common_.direction == query::v2::EdgeAtom::Direction::IN ? "<-" : "-") << "["
<< op.common_.edge_symbol.name();
utils::PrintIterable(*out_, op.common_.edge_types, "|", [this](auto &stream, const auto &edge_type) {
stream << ":" << dba_->EdgeTypeToName(edge_type);
stream << ":" << request_manager_->EdgeTypeToName(edge_type);
});
*out_ << "]" << (op.common_.direction == query::v2::EdgeAtom::Direction::OUT ? "->" : "-") << "("
<< op.common_.node_symbol.name() << ")";
@@ -261,14 +263,15 @@ void PlanPrinter::Branch(query::v2::plan::LogicalOperator &op, const std::string
--depth_;
}
void PrettyPrint(const DbAccessor &dba, const LogicalOperator *plan_root, std::ostream *out) {
PlanPrinter printer(&dba, out);
void PrettyPrint(const msgs::ShardRequestManagerInterface &request_manager, const LogicalOperator *plan_root,
std::ostream *out) {
PlanPrinter printer(&request_manager, out);
// FIXME(mtomic): We should make visitors that take const arguments.
const_cast<LogicalOperator *>(plan_root)->Accept(printer);
}
nlohmann::json PlanToJson(const DbAccessor &dba, const LogicalOperator *plan_root) {
impl::PlanToJsonVisitor visitor(&dba);
nlohmann::json PlanToJson(const msgs::ShardRequestManagerInterface &request_manager, const LogicalOperator *plan_root) {
impl::PlanToJsonVisitor visitor(&request_manager);
// FIXME(mtomic): We should make visitors that take const arguments.
const_cast<LogicalOperator *>(plan_root)->Accept(visitor);
return visitor.output();
@@ -346,11 +349,17 @@ json ToJson(const utils::Bound<Expression *> &bound) {
json ToJson(const Symbol &symbol) { return symbol.name(); }
json ToJson(storage::v3::EdgeTypeId edge_type, const DbAccessor &dba) { return dba.EdgeTypeToName(edge_type); }
json ToJson(storage::v3::EdgeTypeId edge_type, const msgs::ShardRequestManagerInterface &request_manager) {
return request_manager.EdgeTypeToName(edge_type);
}
json ToJson(storage::v3::LabelId label, const DbAccessor &dba) { return dba.LabelToName(label); }
json ToJson(storage::v3::LabelId label, const msgs::ShardRequestManagerInterface &request_manager) {
return request_manager.LabelToName(label);
}
json ToJson(storage::v3::PropertyId property, const DbAccessor &dba) { return dba.PropertyToName(property); }
json ToJson(storage::v3::PropertyId property, const msgs::ShardRequestManagerInterface &request_manager) {
return request_manager.PropertyToName(property);
}
json ToJson(NamedExpression *nexpr) {
json json;
@@ -359,29 +368,30 @@ json ToJson(NamedExpression *nexpr) {
return json;
}
json ToJson(const std::vector<std::pair<storage::v3::PropertyId, Expression *>> &properties, const DbAccessor &dba) {
json ToJson(const std::vector<std::pair<storage::v3::PropertyId, Expression *>> &properties,
const msgs::ShardRequestManagerInterface &request_manager) {
json json;
for (const auto &prop_pair : properties) {
json.emplace(ToJson(prop_pair.first, dba), ToJson(prop_pair.second));
json.emplace(ToJson(prop_pair.first, request_manager), ToJson(prop_pair.second));
}
return json;
}
json ToJson(const NodeCreationInfo &node_info, const DbAccessor &dba) {
json ToJson(const NodeCreationInfo &node_info, const msgs::ShardRequestManagerInterface &request_manager) {
json self;
self["symbol"] = ToJson(node_info.symbol);
self["labels"] = ToJson(node_info.labels, dba);
self["labels"] = ToJson(node_info.labels, request_manager);
const auto *props = std::get_if<PropertiesMapList>(&node_info.properties);
self["properties"] = ToJson(props ? *props : PropertiesMapList{}, dba);
self["properties"] = ToJson(props ? *props : PropertiesMapList{}, request_manager);
return self;
}
json ToJson(const EdgeCreationInfo &edge_info, const DbAccessor &dba) {
json ToJson(const EdgeCreationInfo &edge_info, const msgs::ShardRequestManagerInterface &request_manager) {
json self;
self["symbol"] = ToJson(edge_info.symbol);
const auto *props = std::get_if<PropertiesMapList>(&edge_info.properties);
self["properties"] = ToJson(props ? *props : PropertiesMapList{}, dba);
self["edge_type"] = ToJson(edge_info.edge_type, dba);
self["properties"] = ToJson(props ? *props : PropertiesMapList{}, request_manager);
self["edge_type"] = ToJson(edge_info.edge_type, request_manager);
self["direction"] = ToString(edge_info.direction);
return self;
}
@@ -423,7 +433,7 @@ bool PlanToJsonVisitor::PreVisit(ScanAll &op) {
bool PlanToJsonVisitor::PreVisit(ScanAllByLabel &op) {
json self;
self["name"] = "ScanAllByLabel";
self["label"] = ToJson(op.label_, *dba_);
self["label"] = ToJson(op.label_, *request_manager_);
self["output_symbol"] = ToJson(op.output_symbol_);
op.input_->Accept(*this);
@@ -436,8 +446,8 @@ bool PlanToJsonVisitor::PreVisit(ScanAllByLabel &op) {
bool PlanToJsonVisitor::PreVisit(ScanAllByLabelPropertyRange &op) {
json self;
self["name"] = "ScanAllByLabelPropertyRange";
self["label"] = ToJson(op.label_, *dba_);
self["property"] = ToJson(op.property_, *dba_);
self["label"] = ToJson(op.label_, *request_manager_);
self["property"] = ToJson(op.property_, *request_manager_);
self["lower_bound"] = op.lower_bound_ ? ToJson(*op.lower_bound_) : json();
self["upper_bound"] = op.upper_bound_ ? ToJson(*op.upper_bound_) : json();
self["output_symbol"] = ToJson(op.output_symbol_);
@@ -452,8 +462,8 @@ bool PlanToJsonVisitor::PreVisit(ScanAllByLabelPropertyRange &op) {
bool PlanToJsonVisitor::PreVisit(ScanAllByLabelPropertyValue &op) {
json self;
self["name"] = "ScanAllByLabelPropertyValue";
self["label"] = ToJson(op.label_, *dba_);
self["property"] = ToJson(op.property_, *dba_);
self["label"] = ToJson(op.label_, *request_manager_);
self["property"] = ToJson(op.property_, *request_manager_);
self["expression"] = ToJson(op.expression_);
self["output_symbol"] = ToJson(op.output_symbol_);
@@ -467,8 +477,8 @@ bool PlanToJsonVisitor::PreVisit(ScanAllByLabelPropertyValue &op) {
bool PlanToJsonVisitor::PreVisit(ScanAllByLabelProperty &op) {
json self;
self["name"] = "ScanAllByLabelProperty";
self["label"] = ToJson(op.label_, *dba_);
self["property"] = ToJson(op.property_, *dba_);
self["label"] = ToJson(op.label_, *request_manager_);
self["property"] = ToJson(op.property_, *request_manager_);
self["output_symbol"] = ToJson(op.output_symbol_);
op.input_->Accept(*this);
@@ -491,7 +501,7 @@ bool PlanToJsonVisitor::PreVisit(ScanAllById &op) {
bool PlanToJsonVisitor::PreVisit(CreateNode &op) {
json self;
self["name"] = "CreateNode";
self["node_info"] = ToJson(op.node_info_, *dba_);
self["node_info"] = ToJson(op.node_info_, *request_manager_);
op.input_->Accept(*this);
self["input"] = PopOutput();
@@ -504,8 +514,8 @@ bool PlanToJsonVisitor::PreVisit(CreateExpand &op) {
json self;
self["name"] = "CreateExpand";
self["input_symbol"] = ToJson(op.input_symbol_);
self["node_info"] = ToJson(op.node_info_, *dba_);
self["edge_info"] = ToJson(op.edge_info_, *dba_);
self["node_info"] = ToJson(op.node_info_, *request_manager_);
self["edge_info"] = ToJson(op.edge_info_, *request_manager_);
self["existing_node"] = op.existing_node_;
op.input_->Accept(*this);
@@ -521,7 +531,7 @@ bool PlanToJsonVisitor::PreVisit(Expand &op) {
self["input_symbol"] = ToJson(op.input_symbol_);
self["node_symbol"] = ToJson(op.common_.node_symbol);
self["edge_symbol"] = ToJson(op.common_.edge_symbol);
self["edge_types"] = ToJson(op.common_.edge_types, *dba_);
self["edge_types"] = ToJson(op.common_.edge_types, *request_manager_);
self["direction"] = ToString(op.common_.direction);
self["existing_node"] = op.common_.existing_node;
@@ -538,7 +548,7 @@ bool PlanToJsonVisitor::PreVisit(ExpandVariable &op) {
self["input_symbol"] = ToJson(op.input_symbol_);
self["node_symbol"] = ToJson(op.common_.node_symbol);
self["edge_symbol"] = ToJson(op.common_.edge_symbol);
self["edge_types"] = ToJson(op.common_.edge_types, *dba_);
self["edge_types"] = ToJson(op.common_.edge_types, *request_manager_);
self["direction"] = ToString(op.common_.direction);
self["type"] = ToString(op.type_);
self["is_reverse"] = op.is_reverse_;
@@ -613,7 +623,7 @@ bool PlanToJsonVisitor::PreVisit(Delete &op) {
bool PlanToJsonVisitor::PreVisit(SetProperty &op) {
json self;
self["name"] = "SetProperty";
self["property"] = ToJson(op.property_, *dba_);
self["property"] = ToJson(op.property_, *request_manager_);
self["lhs"] = ToJson(op.lhs_);
self["rhs"] = ToJson(op.rhs_);
@@ -650,7 +660,7 @@ bool PlanToJsonVisitor::PreVisit(SetLabels &op) {
json self;
self["name"] = "SetLabels";
self["input_symbol"] = ToJson(op.input_symbol_);
self["labels"] = ToJson(op.labels_, *dba_);
self["labels"] = ToJson(op.labels_, *request_manager_);
op.input_->Accept(*this);
self["input"] = PopOutput();
@@ -662,7 +672,7 @@ bool PlanToJsonVisitor::PreVisit(SetLabels &op) {
bool PlanToJsonVisitor::PreVisit(RemoveProperty &op) {
json self;
self["name"] = "RemoveProperty";
self["property"] = ToJson(op.property_, *dba_);
self["property"] = ToJson(op.property_, *request_manager_);
self["lhs"] = ToJson(op.lhs_);
op.input_->Accept(*this);
@@ -676,7 +686,7 @@ bool PlanToJsonVisitor::PreVisit(RemoveLabels &op) {
json self;
self["name"] = "RemoveLabels";
self["input_symbol"] = ToJson(op.input_symbol_);
self["labels"] = ToJson(op.labels_, *dba_);
self["labels"] = ToJson(op.labels_, *request_manager_);
op.input_->Accept(*this);
self["input"] = PopOutput();

View File

@@ -18,28 +18,29 @@
#include "query/v2/frontend/ast/ast.hpp"
#include "query/v2/plan/operator.hpp"
#include "query/v2/shard_request_manager.hpp"
namespace memgraph::query::v2 {
class DbAccessor;
namespace plan {
class LogicalOperator;
/// Pretty print a `LogicalOperator` plan to a `std::ostream`.
/// DbAccessor is needed for resolving label and property names.
/// ShardRequestManager is needed for resolving label and property names.
/// Note that `plan_root` isn't modified, but we can't take it as a const
/// because we don't have support for visiting a const LogicalOperator.
void PrettyPrint(const DbAccessor &dba, const LogicalOperator *plan_root, std::ostream *out);
void PrettyPrint(const msgs::ShardRequestManagerInterface &request_manager, const LogicalOperator *plan_root,
std::ostream *out);
/// Overload of `PrettyPrint` which defaults the `std::ostream` to `std::cout`.
inline void PrettyPrint(const DbAccessor &dba, const LogicalOperator *plan_root) {
PrettyPrint(dba, plan_root, &std::cout);
inline void PrettyPrint(const msgs::ShardRequestManagerInterface &request_manager, const LogicalOperator *plan_root) {
PrettyPrint(request_manager, plan_root, &std::cout);
}
/// Convert a `LogicalOperator` plan to a JSON representation.
/// DbAccessor is needed for resolving label and property names.
nlohmann::json PlanToJson(const DbAccessor &dba, const LogicalOperator *plan_root);
nlohmann::json PlanToJson(const msgs::ShardRequestManagerInterface &request_manager, const LogicalOperator *plan_root);
class PlanPrinter : public virtual HierarchicalLogicalOperatorVisitor {
public:
@@ -47,7 +48,7 @@ class PlanPrinter : public virtual HierarchicalLogicalOperatorVisitor {
using HierarchicalLogicalOperatorVisitor::PreVisit;
using HierarchicalLogicalOperatorVisitor::Visit;
PlanPrinter(const DbAccessor *dba, std::ostream *out);
PlanPrinter(const msgs::ShardRequestManagerInterface *request_manager, std::ostream *out);
bool DefaultPreVisit() override;
@@ -114,7 +115,7 @@ class PlanPrinter : public virtual HierarchicalLogicalOperatorVisitor {
void Branch(LogicalOperator &op, const std::string &branch_name = "");
int64_t depth_{0};
const DbAccessor *dba_{nullptr};
const msgs::ShardRequestManagerInterface *request_manager_{nullptr};
std::ostream *out_{nullptr};
};
@@ -132,20 +133,20 @@ nlohmann::json ToJson(const utils::Bound<Expression *> &bound);
nlohmann::json ToJson(const Symbol &symbol);
nlohmann::json ToJson(storage::v3::EdgeTypeId edge_type, const DbAccessor &dba);
nlohmann::json ToJson(storage::v3::EdgeTypeId edge_type, const msgs::ShardRequestManagerInterface &request_manager);
nlohmann::json ToJson(storage::v3::LabelId label, const DbAccessor &dba);
nlohmann::json ToJson(storage::v3::LabelId label, const msgs::ShardRequestManagerInterface &request_manager);
nlohmann::json ToJson(storage::v3::PropertyId property, const DbAccessor &dba);
nlohmann::json ToJson(storage::v3::PropertyId property, const msgs::ShardRequestManagerInterface &request_manager);
nlohmann::json ToJson(NamedExpression *nexpr);
nlohmann::json ToJson(const std::vector<std::pair<storage::v3::PropertyId, Expression *>> &properties,
const DbAccessor &dba);
const msgs::ShardRequestManagerInterface &request_manager);
nlohmann::json ToJson(const NodeCreationInfo &node_info, const DbAccessor &dba);
nlohmann::json ToJson(const NodeCreationInfo &node_info, const msgs::ShardRequestManagerInterface &request_manager);
nlohmann::json ToJson(const EdgeCreationInfo &edge_info, const DbAccessor &dba);
nlohmann::json ToJson(const EdgeCreationInfo &edge_info, const msgs::ShardRequestManagerInterface &request_manager);
nlohmann::json ToJson(const Aggregate::Element &elem);
@@ -160,7 +161,8 @@ nlohmann::json ToJson(const std::vector<T> &items, Args &&...args) {
class PlanToJsonVisitor : public virtual HierarchicalLogicalOperatorVisitor {
public:
explicit PlanToJsonVisitor(const DbAccessor *dba) : dba_(dba) {}
explicit PlanToJsonVisitor(const msgs::ShardRequestManagerInterface *request_manager)
: request_manager_(request_manager) {}
using HierarchicalLogicalOperatorVisitor::PostVisit;
using HierarchicalLogicalOperatorVisitor::PreVisit;
@@ -216,7 +218,7 @@ class PlanToJsonVisitor : public virtual HierarchicalLogicalOperatorVisitor {
protected:
nlohmann::json output_;
const DbAccessor *dba_;
const msgs::ShardRequestManagerInterface *request_manager_;
nlohmann::json PopOutput() {
nlohmann::json tmp;

View File

@@ -24,18 +24,18 @@ namespace memgraph::query::v2::plan {
namespace {
unsigned long long IndividualCycles(const ProfilingStats &cumulative_stats) {
uint64_t IndividualCycles(const ProfilingStats &cumulative_stats) {
return cumulative_stats.num_cycles - std::accumulate(cumulative_stats.children.begin(),
cumulative_stats.children.end(), 0ULL,
[](auto acc, auto &stats) { return acc + stats.num_cycles; });
}
double RelativeTime(unsigned long long num_cycles, unsigned long long total_cycles) {
return static_cast<double>(num_cycles) / total_cycles;
double RelativeTime(const uint64_t num_cycles, const uint64_t total_cycles) {
return static_cast<double>(num_cycles) / static_cast<double>(total_cycles);
}
double AbsoluteTime(unsigned long long num_cycles, unsigned long long total_cycles,
std::chrono::duration<double> total_time) {
double AbsoluteTime(const uint64_t num_cycles, const uint64_t total_cycles,
const std::chrono::duration<double> total_time) {
return (RelativeTime(num_cycles, total_cycles) * static_cast<std::chrono::duration<double, std::milli>>(total_time))
.count();
}
@@ -50,26 +50,29 @@ namespace {
class ProfilingStatsToTableHelper {
public:
ProfilingStatsToTableHelper(unsigned long long total_cycles, std::chrono::duration<double> total_time)
ProfilingStatsToTableHelper(uint64_t total_cycles, std::chrono::duration<double> total_time)
: total_cycles_(total_cycles), total_time_(total_time) {}
void Output(const ProfilingStats &cumulative_stats) {
auto cycles = IndividualCycles(cumulative_stats);
auto custom_data_copy = cumulative_stats.custom_data;
ConvertCyclesToTime(custom_data_copy);
rows_.emplace_back(std::vector<TypedValue>{
TypedValue(FormatOperator(cumulative_stats.name)), TypedValue(cumulative_stats.actual_hits),
TypedValue(FormatRelativeTime(cycles)), TypedValue(FormatAbsoluteTime(cycles))});
rows_.emplace_back(
std::vector<TypedValue>{TypedValue(FormatOperator(cumulative_stats.name)),
TypedValue(cumulative_stats.actual_hits), TypedValue(FormatRelativeTime(cycles)),
TypedValue(FormatAbsoluteTime(cycles)), TypedValue(custom_data_copy.dump())});
for (size_t i = 1; i < cumulative_stats.children.size(); ++i) {
Branch(cumulative_stats.children[i]);
}
if (cumulative_stats.children.size() >= 1) {
if (!cumulative_stats.children.empty()) {
Output(cumulative_stats.children[0]);
}
}
std::vector<std::vector<TypedValue>> rows() { return rows_; }
std::vector<std::vector<TypedValue>> rows() const { return rows_; }
private:
void Branch(const ProfilingStats &cumulative_stats) {
@@ -80,7 +83,28 @@ class ProfilingStatsToTableHelper {
--depth_;
}
std::string Format(const char *str) {
double AbsoluteTime(const uint64_t cycles) const { return plan::AbsoluteTime(cycles, total_cycles_, total_time_); }
double RelativeTime(const uint64_t cycles) const { return plan::RelativeTime(cycles, total_cycles_); }
void ConvertCyclesToTime(nlohmann::json &custom_data) const {
const auto convert_cycles_in_json = [this](nlohmann::json &json) {
if (!json.is_object()) {
return;
}
if (auto it = json.find(ProfilingStats::kNumCycles); it != json.end()) {
auto num_cycles = it.value().get<uint64_t>();
json[ProfilingStats::kAbsoluteTime] = AbsoluteTime(num_cycles);
json[ProfilingStats::kRelativeTime] = RelativeTime(num_cycles);
}
};
for (auto &json : custom_data) {
convert_cycles_in_json(json);
}
}
std::string Format(const char *str) const {
std::ostringstream ss;
for (int64_t i = 0; i < depth_; ++i) {
ss << "| ";
@@ -89,21 +113,21 @@ class ProfilingStatsToTableHelper {
return ss.str();
}
std::string Format(const std::string &str) { return Format(str.c_str()); }
std::string Format(const std::string &str) const { return Format(str.c_str()); }
std::string FormatOperator(const char *str) { return Format(std::string("* ") + str); }
std::string FormatOperator(const char *str) const { return Format(std::string("* ") + str); }
std::string FormatRelativeTime(unsigned long long num_cycles) {
return fmt::format("{: 10.6f} %", RelativeTime(num_cycles, total_cycles_) * 100);
std::string FormatRelativeTime(uint64_t num_cycles) const {
return fmt::format("{: 10.6f} %", RelativeTime(num_cycles) * 100);
}
std::string FormatAbsoluteTime(unsigned long long num_cycles) {
return fmt::format("{: 10.6f} ms", AbsoluteTime(num_cycles, total_cycles_, total_time_));
std::string FormatAbsoluteTime(uint64_t num_cycles) const {
return fmt::format("{: 10.6f} ms", AbsoluteTime(num_cycles));
}
int64_t depth_{0};
std::vector<std::vector<TypedValue>> rows_;
unsigned long long total_cycles_;
uint64_t total_cycles_;
std::chrono::duration<double> total_time_;
};
@@ -126,7 +150,7 @@ class ProfilingStatsToJsonHelper {
using json = nlohmann::json;
public:
ProfilingStatsToJsonHelper(unsigned long long total_cycles, std::chrono::duration<double> total_time)
ProfilingStatsToJsonHelper(uint64_t total_cycles, std::chrono::duration<double> total_time)
: total_cycles_(total_cycles), total_time_(total_time) {}
void Output(const ProfilingStats &cumulative_stats) { return Output(cumulative_stats, &json_); }
@@ -151,7 +175,7 @@ class ProfilingStatsToJsonHelper {
}
json json_;
unsigned long long total_cycles_;
uint64_t total_cycles_;
std::chrono::duration<double> total_time_;
};

View File

@@ -26,10 +26,16 @@ namespace plan {
* Stores profiling statistics for a single logical operator.
*/
struct ProfilingStats {
static constexpr std::string_view kNumCycles{"num_cycles"};
static constexpr std::string_view kRelativeTime{"relative_time"};
static constexpr std::string_view kAbsoluteTime{"absolute_time"};
static constexpr std::string_view kActualHits{"actual_hits"};
int64_t actual_hits{0};
unsigned long long num_cycles{0};
uint64_t num_cycles{0};
uint64_t key{0};
const char *name{nullptr};
nlohmann::json custom_data;
// TODO: This should use the allocator for query execution
std::vector<ProfilingStats> children;
};

View File

@@ -272,6 +272,7 @@ class RuleBasedPlanner {
PropertiesMapList vector_props;
vector_props.reserve(node_properties->size());
for (const auto &kv : *node_properties) {
// TODO(kostasrim) GetProperty should be implemented in terms of ShardRequestManager NameToProperty
vector_props.push_back({GetProperty(kv.first), kv.second});
}
return std::move(vector_props);

View File

@@ -13,6 +13,8 @@
#include <cstdint>
#include <json/json.hpp>
#include "query/v2/context.hpp"
#include "query/v2/plan/profile.hpp"
#include "utils/likely.hpp"
@@ -20,6 +22,43 @@
namespace memgraph::query::v2::plan {
class ScopedCustomProfile {
public:
explicit ScopedCustomProfile(const std::string_view custom_data_name, ExecutionContext &context)
: custom_data_name_(custom_data_name), start_time_{utils::ReadTSC()}, context_{&context} {}
ScopedCustomProfile(const ScopedCustomProfile &) = delete;
ScopedCustomProfile(ScopedCustomProfile &&) = delete;
ScopedCustomProfile &operator=(const ScopedCustomProfile &) = delete;
ScopedCustomProfile &operator=(ScopedCustomProfile &&) = delete;
// If an exception is thrown in any of these functions that signals a problem that is much bigger than we could handle
// it here, thus we don't attempt to handle it.
// NOLINTNEXTLINE(bugprone-exception-escape)
~ScopedCustomProfile() {
if (nullptr != context_->stats_root) [[unlikely]] {
auto &custom_data = context_->stats_root->custom_data[custom_data_name_];
if (!custom_data.is_object()) {
custom_data = nlohmann::json::object();
}
const auto elapsed = utils::ReadTSC() - start_time_;
IncreaseCustomData(custom_data, ProfilingStats::kNumCycles, elapsed);
IncreaseCustomData(custom_data, ProfilingStats::kActualHits, 1);
}
}
private:
static void IncreaseCustomData(nlohmann::json &custom_data, const std::string_view key, const uint64_t increment) {
auto &json_data = custom_data[key];
const auto numerical_data = json_data.is_null() ? 0 : json_data.get<uint64_t>();
json_data = numerical_data + increment;
}
std::string_view custom_data_name_;
uint64_t start_time_;
ExecutionContext *context_;
};
/**
* A RAII class used for profiling logical operators. Instances of this class
* update the profiling data stored within the `ExecutionContext` object and build
@@ -29,7 +68,7 @@ namespace memgraph::query::v2::plan {
class ScopedProfile {
public:
ScopedProfile(uint64_t key, const char *name, query::v2::ExecutionContext *context) noexcept : context_(context) {
if (UNLIKELY(context_->is_profile_query)) {
if (IsProfiling()) [[unlikely]] {
root_ = context_->stats_root;
// Are we the root logical operator?
@@ -60,8 +99,13 @@ class ScopedProfile {
}
}
ScopedProfile(const ScopedProfile &) = delete;
ScopedProfile(ScopedProfile &&) = delete;
ScopedProfile &operator=(const ScopedProfile &) = delete;
ScopedProfile &operator=(ScopedProfile &&) = delete;
~ScopedProfile() noexcept {
if (UNLIKELY(context_->is_profile_query)) {
if (IsProfiling()) [[unlikely]] {
stats_->num_cycles += utils::ReadTSC() - start_time_;
// Restore the old root ("pop")
@@ -70,10 +114,12 @@ class ScopedProfile {
}
private:
[[nodiscard]] bool IsProfiling() const { return context_->is_profile_query; }
query::v2::ExecutionContext *context_;
ProfilingStats *root_{nullptr};
ProfilingStats *stats_{nullptr};
unsigned long long start_time_{0};
uint64_t start_time_{0};
};
} // namespace memgraph::query::v2::plan

View File

@@ -9,12 +9,12 @@
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
/// @file
#pragma once
#include <optional>
#include "query/v2/bindings/typed_value.hpp"
#include "query/v2/shard_request_manager.hpp"
#include "storage/v3/conversions.hpp"
#include "storage/v3/id_types.hpp"
#include "storage/v3/property_value.hpp"
@@ -28,110 +28,37 @@ namespace memgraph::query::v2::plan {
template <class TDbAccessor>
class VertexCountCache {
public:
VertexCountCache(TDbAccessor *db) : db_(db) {}
explicit VertexCountCache(TDbAccessor *shard_request_manager) : shard_request_manager_{shard_request_manager} {}
auto NameToLabel(const std::string &name) { return db_->NameToLabel(name); }
auto NameToProperty(const std::string &name) { return db_->NameToProperty(name); }
auto NameToEdgeType(const std::string &name) { return db_->NameToEdgeType(name); }
auto NameToLabel(const std::string &name) { return shard_request_manager_->NameToLabel(name); }
auto NameToProperty(const std::string &name) { return shard_request_manager_->NameToProperty(name); }
auto NameToEdgeType(const std::string &name) { return shard_request_manager_->NameToEdgeType(name); }
int64_t VerticesCount() {
if (!vertices_count_) vertices_count_ = db_->VerticesCount();
return *vertices_count_;
int64_t VerticesCount() { return 1; }
int64_t VerticesCount(storage::v3::LabelId /*label*/) { return 1; }
int64_t VerticesCount(storage::v3::LabelId /*label*/, storage::v3::PropertyId /*property*/) { return 1; }
int64_t VerticesCount(storage::v3::LabelId /*label*/, storage::v3::PropertyId /*property*/,
const storage::v3::PropertyValue & /*value*/) {
return 1;
}
int64_t VerticesCount(storage::v3::LabelId label) {
if (label_vertex_count_.find(label) == label_vertex_count_.end())
label_vertex_count_[label] = db_->VerticesCount(label);
return label_vertex_count_.at(label);
int64_t VerticesCount(storage::v3::LabelId /*label*/, storage::v3::PropertyId /*property*/,
const std::optional<utils::Bound<storage::v3::PropertyValue>> & /*lower*/,
const std::optional<utils::Bound<storage::v3::PropertyValue>> & /*upper*/) {
return 1;
}
int64_t VerticesCount(storage::v3::LabelId label, storage::v3::PropertyId property) {
auto key = std::make_pair(label, property);
if (label_property_vertex_count_.find(key) == label_property_vertex_count_.end())
label_property_vertex_count_[key] = db_->VerticesCount(label, property);
return label_property_vertex_count_.at(key);
}
int64_t VerticesCount(storage::v3::LabelId label, storage::v3::PropertyId property,
const storage::v3::PropertyValue &value) {
auto label_prop = std::make_pair(label, property);
auto &value_vertex_count = property_value_vertex_count_[label_prop];
// TODO: Why do we even need TypedValue in this whole file?
auto tv_value(storage::v3::PropertyToTypedValue<TypedValue>(value));
if (value_vertex_count.find(tv_value) == value_vertex_count.end())
value_vertex_count[tv_value] = db_->VerticesCount(label, property, value);
return value_vertex_count.at(tv_value);
}
int64_t VerticesCount(storage::v3::LabelId label, storage::v3::PropertyId property,
const std::optional<utils::Bound<storage::v3::PropertyValue>> &lower,
const std::optional<utils::Bound<storage::v3::PropertyValue>> &upper) {
auto label_prop = std::make_pair(label, property);
auto &bounds_vertex_count = property_bounds_vertex_count_[label_prop];
BoundsKey bounds = std::make_pair(lower, upper);
if (bounds_vertex_count.find(bounds) == bounds_vertex_count.end())
bounds_vertex_count[bounds] = db_->VerticesCount(label, property, lower, upper);
return bounds_vertex_count.at(bounds);
}
bool LabelIndexExists(storage::v3::LabelId label) { return db_->LabelIndexExists(label); }
// For now return true if label is primary label
bool LabelIndexExists(storage::v3::LabelId label) { return shard_request_manager_->IsPrimaryLabel(label); }
bool LabelPropertyIndexExists(storage::v3::LabelId label, storage::v3::PropertyId property) {
return db_->LabelPropertyIndexExists(label, property);
return shard_request_manager_->IsPrimaryKey(label, property);
}
private:
typedef std::pair<storage::v3::LabelId, storage::v3::PropertyId> LabelPropertyKey;
struct LabelPropertyHash {
size_t operator()(const LabelPropertyKey &key) const {
return utils::HashCombine<storage::v3::LabelId, storage::v3::PropertyId>{}(key.first, key.second);
}
};
typedef std::pair<std::optional<utils::Bound<storage::v3::PropertyValue>>,
std::optional<utils::Bound<storage::v3::PropertyValue>>>
BoundsKey;
struct BoundsHash {
size_t operator()(const BoundsKey &key) const {
const auto &maybe_lower = key.first;
const auto &maybe_upper = key.second;
query::v2::TypedValue lower;
query::v2::TypedValue upper;
if (maybe_lower) lower = storage::v3::PropertyToTypedValue<TypedValue>(maybe_lower->value());
if (maybe_upper) upper = storage::v3::PropertyToTypedValue<TypedValue>(maybe_upper->value());
query::v2::TypedValue::Hash hash;
return utils::HashCombine<size_t, size_t>{}(hash(lower), hash(upper));
}
};
struct BoundsEqual {
bool operator()(const BoundsKey &a, const BoundsKey &b) const {
auto bound_equal = [](const auto &maybe_bound_a, const auto &maybe_bound_b) {
if (maybe_bound_a && maybe_bound_b && maybe_bound_a->type() != maybe_bound_b->type()) return false;
query::v2::TypedValue bound_a;
query::v2::TypedValue bound_b;
if (maybe_bound_a) bound_a = storage::v3::PropertyToTypedValue<TypedValue>(maybe_bound_a->value());
if (maybe_bound_b) bound_b = storage::v3::PropertyToTypedValue<TypedValue>(maybe_bound_b->value());
return query::v2::TypedValue::BoolEqual{}(bound_a, bound_b);
};
return bound_equal(a.first, b.first) && bound_equal(a.second, b.second);
}
};
TDbAccessor *db_;
std::optional<int64_t> vertices_count_;
std::unordered_map<storage::v3::LabelId, int64_t> label_vertex_count_;
std::unordered_map<LabelPropertyKey, int64_t, LabelPropertyHash> label_property_vertex_count_;
std::unordered_map<
LabelPropertyKey,
std::unordered_map<query::v2::TypedValue, int64_t, query::v2::TypedValue::Hash, query::v2::TypedValue::BoolEqual>,
LabelPropertyHash>
property_value_vertex_count_;
std::unordered_map<LabelPropertyKey, std::unordered_map<BoundsKey, int64_t, BoundsHash, BoundsEqual>,
LabelPropertyHash>
property_bounds_vertex_count_;
msgs::ShardRequestManagerInterface *shard_request_manager_;
};
template <class TDbAccessor>

View File

@@ -1,293 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
/// @file
#pragma once
#include "mg_procedure.h"
#include <functional>
#include <memory>
#include <string_view>
#include "query/v2/bindings/typed_value.hpp"
#include "query/v2/procedure/cypher_type_ptr.hpp"
#include "query/v2/procedure/mg_procedure_impl.hpp"
#include "utils/memory.hpp"
#include "utils/pmr/string.hpp"
namespace memgraph::query::v2::procedure {
class ListType;
class NullableType;
/// Interface for all supported types in openCypher type system.
class CypherType {
public:
CypherType() = default;
virtual ~CypherType() = default;
CypherType(const CypherType &) = delete;
CypherType(CypherType &&) = delete;
CypherType &operator=(const CypherType &) = delete;
CypherType &operator=(CypherType &&) = delete;
/// Get name of the type as it should be presented to the user.
virtual std::string_view GetPresentableName() const = 0;
/// Return true if given mgp_value is of the type as described by `this`.
virtual bool SatisfiesType(const mgp_value &) const = 0;
/// Return true if given TypedValue is of the type as described by `this`.
virtual bool SatisfiesType(const query::v2::TypedValue &) const = 0;
// The following methods are a simple replacement for RTTI because we have
// some special cases we need to handle.
virtual const ListType *AsListType() const { return nullptr; }
virtual const NullableType *AsNullableType() const { return nullptr; }
};
// Simple Types
class AnyType : public CypherType {
public:
std::string_view GetPresentableName() const override { return "ANY"; }
bool SatisfiesType(const mgp_value &value) const override { return value.type != MGP_VALUE_TYPE_NULL; }
bool SatisfiesType(const query::v2::TypedValue &value) const override { return !value.IsNull(); }
};
class BoolType : public CypherType {
public:
std::string_view GetPresentableName() const override { return "BOOLEAN"; }
bool SatisfiesType(const mgp_value &value) const override { return value.type == MGP_VALUE_TYPE_BOOL; }
bool SatisfiesType(const query::v2::TypedValue &value) const override { return value.IsBool(); }
};
class StringType : public CypherType {
public:
std::string_view GetPresentableName() const override { return "STRING"; }
bool SatisfiesType(const mgp_value &value) const override { return value.type == MGP_VALUE_TYPE_STRING; }
bool SatisfiesType(const query::v2::TypedValue &value) const override { return value.IsString(); }
};
class IntType : public CypherType {
public:
std::string_view GetPresentableName() const override { return "INTEGER"; }
bool SatisfiesType(const mgp_value &value) const override { return value.type == MGP_VALUE_TYPE_INT; }
bool SatisfiesType(const query::v2::TypedValue &value) const override { return value.IsInt(); }
};
class FloatType : public CypherType {
public:
std::string_view GetPresentableName() const override { return "FLOAT"; }
bool SatisfiesType(const mgp_value &value) const override { return value.type == MGP_VALUE_TYPE_DOUBLE; }
bool SatisfiesType(const query::v2::TypedValue &value) const override { return value.IsDouble(); }
};
class NumberType : public CypherType {
public:
std::string_view GetPresentableName() const override { return "NUMBER"; }
bool SatisfiesType(const mgp_value &value) const override {
return value.type == MGP_VALUE_TYPE_INT || value.type == MGP_VALUE_TYPE_DOUBLE;
}
bool SatisfiesType(const query::v2::TypedValue &value) const override { return value.IsInt() || value.IsDouble(); }
};
class NodeType : public CypherType {
public:
std::string_view GetPresentableName() const override { return "NODE"; }
bool SatisfiesType(const mgp_value &value) const override { return value.type == MGP_VALUE_TYPE_VERTEX; }
bool SatisfiesType(const query::v2::TypedValue &value) const override { return value.IsVertex(); }
};
class RelationshipType : public CypherType {
public:
std::string_view GetPresentableName() const override { return "RELATIONSHIP"; }
bool SatisfiesType(const mgp_value &value) const override { return value.type == MGP_VALUE_TYPE_EDGE; }
bool SatisfiesType(const query::v2::TypedValue &value) const override { return value.IsEdge(); }
};
class PathType : public CypherType {
public:
std::string_view GetPresentableName() const override { return "PATH"; }
bool SatisfiesType(const mgp_value &value) const override { return value.type == MGP_VALUE_TYPE_PATH; }
bool SatisfiesType(const query::v2::TypedValue &value) const override { return value.IsPath(); }
};
// You'd think that MapType would be a composite type like ListType, but nope.
// Why? No-one really knows. It's defined like that in "CIP2015-09-16 Public
// Type System and Type Annotations"
// Additionally, MapType also covers NodeType and RelationshipType because
// values of that type have property *maps*.
class MapType : public CypherType {
public:
std::string_view GetPresentableName() const override { return "MAP"; }
bool SatisfiesType(const mgp_value &value) const override {
return value.type == MGP_VALUE_TYPE_MAP || value.type == MGP_VALUE_TYPE_VERTEX || value.type == MGP_VALUE_TYPE_EDGE;
}
bool SatisfiesType(const query::v2::TypedValue &value) const override {
return value.IsMap() || value.IsVertex() || value.IsEdge();
}
};
// Temporal Types
class DateType : public CypherType {
public:
std::string_view GetPresentableName() const override { return "DATE"; }
bool SatisfiesType(const mgp_value &value) const override { return value.type == MGP_VALUE_TYPE_DATE; }
bool SatisfiesType(const query::v2::TypedValue &value) const override { return value.IsDate(); }
};
class LocalTimeType : public CypherType {
public:
std::string_view GetPresentableName() const override { return "LOCAL_TIME"; }
bool SatisfiesType(const mgp_value &value) const override { return value.type == MGP_VALUE_TYPE_LOCAL_TIME; }
bool SatisfiesType(const query::v2::TypedValue &value) const override { return value.IsLocalTime(); }
};
class LocalDateTimeType : public CypherType {
public:
std::string_view GetPresentableName() const override { return "LOCAL_DATE_TIME"; }
bool SatisfiesType(const mgp_value &value) const override { return value.type == MGP_VALUE_TYPE_LOCAL_DATE_TIME; }
bool SatisfiesType(const query::v2::TypedValue &value) const override { return value.IsLocalDateTime(); }
};
class DurationType : public CypherType {
public:
std::string_view GetPresentableName() const override { return "DURATION"; }
bool SatisfiesType(const mgp_value &value) const override { return value.type == MGP_VALUE_TYPE_DURATION; }
bool SatisfiesType(const query::v2::TypedValue &value) const override { return value.IsDuration(); }
};
// Composite Types
class ListType : public CypherType {
public:
CypherTypePtr element_type_;
utils::pmr::string presentable_name_;
/// @throw std::bad_alloc
/// @throw std::length_error
explicit ListType(CypherTypePtr element_type, utils::MemoryResource *memory)
: element_type_(std::move(element_type)), presentable_name_("LIST OF ", memory) {
presentable_name_.append(element_type_->GetPresentableName());
}
std::string_view GetPresentableName() const override { return presentable_name_; }
bool SatisfiesType(const mgp_value &value) const override {
if (value.type != MGP_VALUE_TYPE_LIST) {
return false;
}
auto *list = value.list_v;
const auto list_size = list->elems.size();
for (size_t i = 0; i < list_size; ++i) {
if (!element_type_->SatisfiesType(list->elems[i])) {
return false;
};
}
return true;
}
bool SatisfiesType(const query::v2::TypedValue &value) const override {
if (!value.IsList()) return false;
for (const auto &elem : value.ValueList()) {
if (!element_type_->SatisfiesType(elem)) return false;
}
return true;
}
const ListType *AsListType() const override { return this; }
};
class NullableType : public CypherType {
CypherTypePtr type_;
utils::pmr::string presentable_name_;
// Constructor is private, because we use a factory method Create to prevent
// nesting NullableType on top of each other.
// @throw std::bad_alloc
// @throw std::length_error
explicit NullableType(CypherTypePtr type, utils::MemoryResource *memory)
: type_(std::move(type)), presentable_name_(memory) {
const auto *list_type = type_->AsListType();
// ListType is specially formatted
if (list_type) {
presentable_name_.assign("LIST? OF ").append(list_type->element_type_->GetPresentableName());
} else {
presentable_name_.assign(type_->GetPresentableName()).append("?");
}
}
public:
/// Create a NullableType of some CypherType.
/// If passed in `type` is already a NullableType, it is returned intact.
/// Otherwise, `type` is wrapped in a new instance of NullableType.
/// @throw std::bad_alloc
/// @throw std::length_error
static CypherTypePtr Create(CypherTypePtr type, utils::MemoryResource *memory) {
if (type->AsNullableType()) return type;
utils::Allocator<NullableType> alloc(memory);
auto *nullable = alloc.allocate(1);
try {
new (nullable) NullableType(std::move(type), memory);
} catch (...) {
alloc.deallocate(nullable, 1);
throw;
}
return CypherTypePtr(nullable, [alloc](CypherType *base_ptr) mutable {
alloc.delete_object(static_cast<NullableType *>(base_ptr));
});
}
std::string_view GetPresentableName() const override { return presentable_name_; }
bool SatisfiesType(const mgp_value &value) const override {
return value.type == MGP_VALUE_TYPE_NULL || type_->SatisfiesType(value);
}
bool SatisfiesType(const query::v2::TypedValue &value) const override {
return value.IsNull() || type_->SatisfiesType(value);
}
const NullableType *AsNullableType() const override { return this; }
};
} // namespace memgraph::query::v2::procedure

View File

@@ -1,36 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#include "query/v2/procedure/mg_procedure_helpers.hpp"
namespace memgraph::query::v2::procedure {
MgpUniquePtr<mgp_value> GetStringValueOrSetError(const char *string, mgp_memory *memory, mgp_result *result) {
procedure::MgpUniquePtr<mgp_value> value{nullptr, mgp_value_destroy};
const auto success =
TryOrSetError([&] { return procedure::CreateMgpObject(value, mgp_value_make_string, string, memory); }, result);
if (!success) {
value.reset();
}
return value;
}
bool InsertResultOrSetError(mgp_result *result, mgp_result_record *record, const char *result_name, mgp_value *value) {
if (const auto err = mgp_result_record_insert(record, result_name, value); err != mgp_error::MGP_ERROR_NO_ERROR) {
const auto error_msg = fmt::format("Unable to set the result for {}, error = {}", result_name, err);
static_cast<void>(mgp_result_set_error_msg(result, error_msg.c_str()));
return false;
}
return true;
}
} // namespace memgraph::query::v2::procedure

View File

@@ -1,69 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#pragma once
#include <memory>
#include <type_traits>
#include <utility>
#include <fmt/format.h>
#include "mg_procedure.h"
namespace memgraph::query::v2::procedure {
template <typename TResult, typename TFunc, typename... TArgs>
TResult Call(TFunc func, TArgs... args) {
static_assert(std::is_trivially_copyable_v<TFunc>);
static_assert((std::is_trivially_copyable_v<std::remove_reference_t<TArgs>> && ...));
TResult result{};
MG_ASSERT(func(args..., &result) == mgp_error::MGP_ERROR_NO_ERROR);
return result;
}
template <typename TFunc, typename... TArgs>
bool CallBool(TFunc func, TArgs... args) {
return Call<int>(func, args...) != 0;
}
template <typename TObj>
using MgpRawObjectDeleter = void (*)(TObj *);
template <typename TObj>
using MgpUniquePtr = std::unique_ptr<TObj, MgpRawObjectDeleter<TObj>>;
template <typename TObj, typename TFunc, typename... TArgs>
mgp_error CreateMgpObject(MgpUniquePtr<TObj> &obj, TFunc func, TArgs &&...args) {
TObj *raw_obj{nullptr};
const auto err = func(std::forward<TArgs>(args)..., &raw_obj);
obj.reset(raw_obj);
return err;
}
template <typename Fun>
[[nodiscard]] bool TryOrSetError(Fun &&func, mgp_result *result) {
if (const auto err = func(); err == mgp_error::MGP_ERROR_UNABLE_TO_ALLOCATE) {
static_cast<void>(mgp_result_set_error_msg(result, "Not enough memory!"));
return false;
} else if (err != mgp_error::MGP_ERROR_NO_ERROR) {
const auto error_msg = fmt::format("Unexpected error ({})!", err);
static_cast<void>(mgp_result_set_error_msg(result, error_msg.c_str()));
return false;
}
return true;
}
[[nodiscard]] MgpUniquePtr<mgp_value> GetStringValueOrSetError(const char *string, mgp_memory *memory,
mgp_result *result);
[[nodiscard]] bool InsertResultOrSetError(mgp_result *result, mgp_result_record *record, const char *result_name,
mgp_value *value);
} // namespace memgraph::query::v2::procedure

File diff suppressed because it is too large Load Diff

View File

@@ -1,927 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
/// @file
/// Contains private (implementation) declarations and definitions for
/// mg_procedure.h
#pragma once
#include "mg_procedure.h"
#include <optional>
#include <ostream>
#include "integrations/kafka/consumer.hpp"
#include "integrations/pulsar/consumer.hpp"
#include "query/v2/bindings/typed_value.hpp"
#include "query/v2/context.hpp"
#include "query/v2/db_accessor.hpp"
#include "query/v2/frontend/ast/ast.hpp"
#include "query/v2/procedure/cypher_type_ptr.hpp"
#include "storage/v3/view.hpp"
#include "utils/memory.hpp"
#include "utils/pmr/map.hpp"
#include "utils/pmr/string.hpp"
#include "utils/pmr/vector.hpp"
#include "utils/temporal.hpp"
/// Wraps memory resource used in custom procedures.
///
/// This should have been `using mgp_memory = memgraph::utils::MemoryResource`, but that's
/// not valid C++ because we have a forward declare `struct mgp_memory` in
/// mg_procedure.h
/// TODO: Make this extendable in C API, so that custom procedure writer can add
/// their own memory management wrappers.
struct mgp_memory {
memgraph::utils::MemoryResource *impl;
};
/// Immutable container of various values that appear in openCypher.
struct mgp_value {
/// Allocator type so that STL containers are aware that we need one.
using allocator_type = memgraph::utils::Allocator<mgp_value>;
// Construct MGP_VALUE_TYPE_NULL.
explicit mgp_value(memgraph::utils::MemoryResource *) noexcept;
mgp_value(bool, memgraph::utils::MemoryResource *) noexcept;
mgp_value(int64_t, memgraph::utils::MemoryResource *) noexcept;
mgp_value(double, memgraph::utils::MemoryResource *) noexcept;
/// @throw std::bad_alloc
mgp_value(const char *, memgraph::utils::MemoryResource *);
/// Take ownership of the mgp_list, MemoryResource must match.
mgp_value(mgp_list *, memgraph::utils::MemoryResource *) noexcept;
/// Take ownership of the mgp_map, MemoryResource must match.
mgp_value(mgp_map *, memgraph::utils::MemoryResource *) noexcept;
/// Take ownership of the mgp_vertex, MemoryResource must match.
mgp_value(mgp_vertex *, memgraph::utils::MemoryResource *) noexcept;
/// Take ownership of the mgp_edge, MemoryResource must match.
mgp_value(mgp_edge *, memgraph::utils::MemoryResource *) noexcept;
/// Take ownership of the mgp_path, MemoryResource must match.
mgp_value(mgp_path *, memgraph::utils::MemoryResource *) noexcept;
mgp_value(mgp_date *, memgraph::utils::MemoryResource *) noexcept;
mgp_value(mgp_local_time *, memgraph::utils::MemoryResource *) noexcept;
mgp_value(mgp_local_date_time *, memgraph::utils::MemoryResource *) noexcept;
mgp_value(mgp_duration *, memgraph::utils::MemoryResource *) noexcept;
/// Construct by copying memgraph::query::v2::TypedValue using memgraph::utils::MemoryResource.
/// mgp_graph is needed to construct mgp_vertex and mgp_edge.
/// @throw std::bad_alloc
mgp_value(const memgraph::query::v2::TypedValue &, mgp_graph *, memgraph::utils::MemoryResource *);
/// Construct by copying memgraph::storage::v3::PropertyValue using memgraph::utils::MemoryResource.
/// @throw std::bad_alloc
mgp_value(const memgraph::storage::v3::PropertyValue &, memgraph::utils::MemoryResource *);
/// Copy construction without memgraph::utils::MemoryResource is not allowed.
mgp_value(const mgp_value &) = delete;
/// Copy construct using given memgraph::utils::MemoryResource.
/// @throw std::bad_alloc
mgp_value(const mgp_value &, memgraph::utils::MemoryResource *);
/// Move construct using given memgraph::utils::MemoryResource.
/// @throw std::bad_alloc if MemoryResource is different, so we cannot move.
mgp_value(mgp_value &&, memgraph::utils::MemoryResource *);
/// Move construct, memgraph::utils::MemoryResource is inherited.
mgp_value(mgp_value &&other) noexcept : mgp_value(other, other.memory) {}
/// Copy-assignment is not allowed to preserve immutability.
mgp_value &operator=(const mgp_value &) = delete;
/// Move-assignment is not allowed to preserve immutability.
mgp_value &operator=(mgp_value &&) = delete;
~mgp_value() noexcept;
memgraph::utils::MemoryResource *GetMemoryResource() const noexcept { return memory; }
mgp_value_type type;
memgraph::utils::MemoryResource *memory;
union {
bool bool_v;
int64_t int_v;
double double_v;
memgraph::utils::pmr::string string_v;
// We use pointers so that taking ownership via C API is easier. Besides,
// mgp_map cannot use incomplete mgp_value type, because that would be
// undefined behaviour.
mgp_list *list_v;
mgp_map *map_v;
mgp_vertex *vertex_v;
mgp_edge *edge_v;
mgp_path *path_v;
mgp_date *date_v;
mgp_local_time *local_time_v;
mgp_local_date_time *local_date_time_v;
mgp_duration *duration_v;
};
};
inline memgraph::utils::DateParameters MapDateParameters(const mgp_date_parameters *parameters) {
return {.year = parameters->year, .month = parameters->month, .day = parameters->day};
}
struct mgp_date {
/// Allocator type so that STL containers are aware that we need one.
/// We don't actually need this, but it simplifies the C API, because we store
/// the allocator which was used to allocate `this`.
using allocator_type = memgraph::utils::Allocator<mgp_date>;
// Hopefully memgraph::utils::Date copy constructor remains noexcept, so that we can
// have everything noexcept here.
static_assert(std::is_nothrow_copy_constructible_v<memgraph::utils::Date>);
mgp_date(const memgraph::utils::Date &date, memgraph::utils::MemoryResource *memory) noexcept
: memory(memory), date(date) {}
mgp_date(const std::string_view string, memgraph::utils::MemoryResource *memory) noexcept
: memory(memory), date(memgraph::utils::ParseDateParameters(string).first) {}
mgp_date(const mgp_date_parameters *parameters, memgraph::utils::MemoryResource *memory) noexcept
: memory(memory), date(MapDateParameters(parameters)) {}
mgp_date(const int64_t microseconds, memgraph::utils::MemoryResource *memory) noexcept
: memory(memory), date(microseconds) {}
mgp_date(const mgp_date &other, memgraph::utils::MemoryResource *memory) noexcept
: memory(memory), date(other.date) {}
mgp_date(mgp_date &&other, memgraph::utils::MemoryResource *memory) noexcept : memory(memory), date(other.date) {}
mgp_date(mgp_date &&other) noexcept : memory(other.memory), date(other.date) {}
/// Copy construction without memgraph::utils::MemoryResource is not allowed.
mgp_date(const mgp_date &) = delete;
mgp_date &operator=(const mgp_date &) = delete;
mgp_date &operator=(mgp_date &&) = delete;
~mgp_date() = default;
memgraph::utils::MemoryResource *GetMemoryResource() const noexcept { return memory; }
memgraph::utils::MemoryResource *memory;
memgraph::utils::Date date;
};
inline memgraph::utils::LocalTimeParameters MapLocalTimeParameters(const mgp_local_time_parameters *parameters) {
return {.hour = parameters->hour,
.minute = parameters->minute,
.second = parameters->second,
.millisecond = parameters->millisecond,
.microsecond = parameters->microsecond};
}
struct mgp_local_time {
/// Allocator type so that STL containers are aware that we need one.
/// We don't actually need this, but it simplifies the C API, because we store
/// the allocator which was used to allocate `this`.
using allocator_type = memgraph::utils::Allocator<mgp_local_time>;
// Hopefully memgraph::utils::LocalTime copy constructor remains noexcept, so that we can
// have everything noexcept here.
static_assert(std::is_nothrow_copy_constructible_v<memgraph::utils::LocalTime>);
mgp_local_time(const std::string_view string, memgraph::utils::MemoryResource *memory) noexcept
: memory(memory), local_time(memgraph::utils::ParseLocalTimeParameters(string).first) {}
mgp_local_time(const mgp_local_time_parameters *parameters, memgraph::utils::MemoryResource *memory) noexcept
: memory(memory), local_time(MapLocalTimeParameters(parameters)) {}
mgp_local_time(const memgraph::utils::LocalTime &local_time, memgraph::utils::MemoryResource *memory) noexcept
: memory(memory), local_time(local_time) {}
mgp_local_time(const int64_t microseconds, memgraph::utils::MemoryResource *memory) noexcept
: memory(memory), local_time(microseconds) {}
mgp_local_time(const mgp_local_time &other, memgraph::utils::MemoryResource *memory) noexcept
: memory(memory), local_time(other.local_time) {}
mgp_local_time(mgp_local_time &&other, memgraph::utils::MemoryResource *memory) noexcept
: memory(memory), local_time(other.local_time) {}
mgp_local_time(mgp_local_time &&other) noexcept : memory(other.memory), local_time(other.local_time) {}
/// Copy construction without memgraph::utils::MemoryResource is not allowed.
mgp_local_time(const mgp_local_time &) = delete;
mgp_local_time &operator=(const mgp_local_time &) = delete;
mgp_local_time &operator=(mgp_local_time &&) = delete;
~mgp_local_time() = default;
memgraph::utils::MemoryResource *GetMemoryResource() const noexcept { return memory; }
memgraph::utils::MemoryResource *memory;
memgraph::utils::LocalTime local_time;
};
inline memgraph::utils::LocalDateTime CreateLocalDateTimeFromString(const std::string_view string) {
const auto &[date_parameters, local_time_parameters] = memgraph::utils::ParseLocalDateTimeParameters(string);
return memgraph::utils::LocalDateTime{date_parameters, local_time_parameters};
}
struct mgp_local_date_time {
/// Allocator type so that STL containers are aware that we need one.
/// We don't actually need this, but it simplifies the C API, because we store
/// the allocator which was used to allocate `this`.
using allocator_type = memgraph::utils::Allocator<mgp_local_date_time>;
// Hopefully memgraph::utils::LocalDateTime copy constructor remains noexcept, so that we can
// have everything noexcept here.
static_assert(std::is_nothrow_copy_constructible_v<memgraph::utils::LocalDateTime>);
mgp_local_date_time(const memgraph::utils::LocalDateTime &local_date_time,
memgraph::utils::MemoryResource *memory) noexcept
: memory(memory), local_date_time(local_date_time) {}
mgp_local_date_time(const std::string_view string, memgraph::utils::MemoryResource *memory) noexcept
: memory(memory), local_date_time(CreateLocalDateTimeFromString(string)) {}
mgp_local_date_time(const mgp_local_date_time_parameters *parameters,
memgraph::utils::MemoryResource *memory) noexcept
: memory(memory),
local_date_time(MapDateParameters(parameters->date_parameters),
MapLocalTimeParameters(parameters->local_time_parameters)) {}
mgp_local_date_time(const int64_t microseconds, memgraph::utils::MemoryResource *memory) noexcept
: memory(memory), local_date_time(microseconds) {}
mgp_local_date_time(const mgp_local_date_time &other, memgraph::utils::MemoryResource *memory) noexcept
: memory(memory), local_date_time(other.local_date_time) {}
mgp_local_date_time(mgp_local_date_time &&other, memgraph::utils::MemoryResource *memory) noexcept
: memory(memory), local_date_time(other.local_date_time) {}
mgp_local_date_time(mgp_local_date_time &&other) noexcept
: memory(other.memory), local_date_time(other.local_date_time) {}
/// Copy construction without memgraph::utils::MemoryResource is not allowed.
mgp_local_date_time(const mgp_local_date_time &) = delete;
mgp_local_date_time &operator=(const mgp_local_date_time &) = delete;
mgp_local_date_time &operator=(mgp_local_date_time &&) = delete;
~mgp_local_date_time() = default;
memgraph::utils::MemoryResource *GetMemoryResource() const noexcept { return memory; }
memgraph::utils::MemoryResource *memory;
memgraph::utils::LocalDateTime local_date_time;
};
inline memgraph::utils::DurationParameters MapDurationParameters(const mgp_duration_parameters *parameters) {
return {.day = parameters->day,
.hour = parameters->hour,
.minute = parameters->minute,
.second = parameters->second,
.millisecond = parameters->millisecond,
.microsecond = parameters->microsecond};
}
struct mgp_duration {
/// Allocator type so that STL containers are aware that we need one.
/// We don't actually need this, but it simplifies the C API, because we store
/// the allocator which was used to allocate `this`.
using allocator_type = memgraph::utils::Allocator<mgp_duration>;
// Hopefully memgraph::utils::Duration copy constructor remains noexcept, so that we can
// have everything noexcept here.
static_assert(std::is_nothrow_copy_constructible_v<memgraph::utils::Duration>);
mgp_duration(const std::string_view string, memgraph::utils::MemoryResource *memory) noexcept
: memory(memory), duration(memgraph::utils::ParseDurationParameters(string)) {}
mgp_duration(const mgp_duration_parameters *parameters, memgraph::utils::MemoryResource *memory) noexcept
: memory(memory), duration(MapDurationParameters(parameters)) {}
mgp_duration(const memgraph::utils::Duration &duration, memgraph::utils::MemoryResource *memory) noexcept
: memory(memory), duration(duration) {}
mgp_duration(const int64_t microseconds, memgraph::utils::MemoryResource *memory) noexcept
: memory(memory), duration(microseconds) {}
mgp_duration(const mgp_duration &other, memgraph::utils::MemoryResource *memory) noexcept
: memory(memory), duration(other.duration) {}
mgp_duration(mgp_duration &&other, memgraph::utils::MemoryResource *memory) noexcept
: memory(memory), duration(other.duration) {}
mgp_duration(mgp_duration &&other) noexcept : memory(other.memory), duration(other.duration) {}
/// Copy construction without memgraph::utils::MemoryResource is not allowed.
mgp_duration(const mgp_duration &) = delete;
mgp_duration &operator=(const mgp_duration &) = delete;
mgp_duration &operator=(mgp_duration &&) = delete;
~mgp_duration() = default;
memgraph::utils::MemoryResource *GetMemoryResource() const noexcept { return memory; }
memgraph::utils::MemoryResource *memory;
memgraph::utils::Duration duration;
};
struct mgp_list {
/// Allocator type so that STL containers are aware that we need one.
using allocator_type = memgraph::utils::Allocator<mgp_list>;
explicit mgp_list(memgraph::utils::MemoryResource *memory) : elems(memory) {}
mgp_list(memgraph::utils::pmr::vector<mgp_value> &&elems, memgraph::utils::MemoryResource *memory)
: elems(std::move(elems), memory) {}
mgp_list(const mgp_list &other, memgraph::utils::MemoryResource *memory) : elems(other.elems, memory) {}
mgp_list(mgp_list &&other, memgraph::utils::MemoryResource *memory) : elems(std::move(other.elems), memory) {}
mgp_list(mgp_list &&other) noexcept : elems(std::move(other.elems)) {}
/// Copy construction without memgraph::utils::MemoryResource is not allowed.
mgp_list(const mgp_list &) = delete;
mgp_list &operator=(const mgp_list &) = delete;
mgp_list &operator=(mgp_list &&) = delete;
~mgp_list() = default;
memgraph::utils::MemoryResource *GetMemoryResource() const noexcept {
return elems.get_allocator().GetMemoryResource();
}
// C++17 vector can work with incomplete type.
memgraph::utils::pmr::vector<mgp_value> elems;
};
struct mgp_map {
/// Allocator type so that STL containers are aware that we need one.
using allocator_type = memgraph::utils::Allocator<mgp_map>;
explicit mgp_map(memgraph::utils::MemoryResource *memory) : items(memory) {}
mgp_map(memgraph::utils::pmr::map<memgraph::utils::pmr::string, mgp_value> &&items,
memgraph::utils::MemoryResource *memory)
: items(std::move(items), memory) {}
mgp_map(const mgp_map &other, memgraph::utils::MemoryResource *memory) : items(other.items, memory) {}
mgp_map(mgp_map &&other, memgraph::utils::MemoryResource *memory) : items(std::move(other.items), memory) {}
mgp_map(mgp_map &&other) noexcept : items(std::move(other.items)) {}
/// Copy construction without memgraph::utils::MemoryResource is not allowed.
mgp_map(const mgp_map &) = delete;
mgp_map &operator=(const mgp_map &) = delete;
mgp_map &operator=(mgp_map &&) = delete;
~mgp_map() = default;
memgraph::utils::MemoryResource *GetMemoryResource() const noexcept {
return items.get_allocator().GetMemoryResource();
}
// Unfortunately using incomplete type with map is undefined, so mgp_map
// needs to be defined after mgp_value.
memgraph::utils::pmr::map<memgraph::utils::pmr::string, mgp_value> items;
};
struct mgp_map_item {
const char *key;
mgp_value *value;
};
struct mgp_map_items_iterator {
using allocator_type = memgraph::utils::Allocator<mgp_map_items_iterator>;
mgp_map_items_iterator(mgp_map *map, memgraph::utils::MemoryResource *memory)
: memory(memory), map(map), current_it(map->items.begin()) {
if (current_it != map->items.end()) {
current.key = current_it->first.c_str();
current.value = &current_it->second;
}
}
mgp_map_items_iterator(const mgp_map_items_iterator &) = delete;
mgp_map_items_iterator(mgp_map_items_iterator &&) = delete;
mgp_map_items_iterator &operator=(const mgp_map_items_iterator &) = delete;
mgp_map_items_iterator &operator=(mgp_map_items_iterator &&) = delete;
~mgp_map_items_iterator() = default;
memgraph::utils::MemoryResource *GetMemoryResource() const { return memory; }
memgraph::utils::MemoryResource *memory;
mgp_map *map;
decltype(map->items.begin()) current_it;
mgp_map_item current;
};
struct mgp_vertex {
/// Allocator type so that STL containers are aware that we need one.
/// We don't actually need this, but it simplifies the C API, because we store
/// the allocator which was used to allocate `this`.
using allocator_type = memgraph::utils::Allocator<mgp_vertex>;
// Hopefully VertexAccessor copy constructor remains noexcept, so that we can
// have everything noexcept here.
static_assert(std::is_nothrow_copy_constructible_v<memgraph::query::v2::VertexAccessor>);
mgp_vertex(memgraph::query::v2::VertexAccessor v, mgp_graph *graph, memgraph::utils::MemoryResource *memory) noexcept
: memory(memory), impl(v), graph(graph) {}
mgp_vertex(const mgp_vertex &other, memgraph::utils::MemoryResource *memory) noexcept
: memory(memory), impl(other.impl), graph(other.graph) {}
mgp_vertex(mgp_vertex &&other, memgraph::utils::MemoryResource *memory) noexcept
: memory(memory), impl(other.impl), graph(other.graph) {}
mgp_vertex(mgp_vertex &&other) noexcept : memory(other.memory), impl(other.impl), graph(other.graph) {}
/// Copy construction without memgraph::utils::MemoryResource is not allowed.
mgp_vertex(const mgp_vertex &) = delete;
mgp_vertex &operator=(const mgp_vertex &) = delete;
mgp_vertex &operator=(mgp_vertex &&) = delete;
bool operator==(const mgp_vertex &other) const noexcept { return this->impl == other.impl; }
bool operator!=(const mgp_vertex &other) const noexcept { return !(*this == other); };
~mgp_vertex() = default;
memgraph::utils::MemoryResource *GetMemoryResource() const noexcept { return memory; }
memgraph::utils::MemoryResource *memory;
memgraph::query::v2::VertexAccessor impl;
mgp_graph *graph;
};
struct mgp_edge {
/// Allocator type so that STL containers are aware that we need one.
/// We don't actually need this, but it simplifies the C API, because we store
/// the allocator which was used to allocate `this`.
using allocator_type = memgraph::utils::Allocator<mgp_edge>;
// TODO(antaljanosbenjamin): Handle this static assert failure when we will support procedures again
// Hopefully EdgeAccessor copy constructor remains noexcept, so that we can
// have everything noexcept here.
// static_assert(std::is_nothrow_copy_constructible_v<memgraph::query::v2::EdgeAccessor>);
static mgp_edge *Copy(const mgp_edge &edge, mgp_memory &memory);
mgp_edge(const memgraph::query::v2::EdgeAccessor &impl, mgp_graph *graph,
memgraph::utils::MemoryResource *memory) noexcept
: memory(memory), impl(impl), from(impl.From(), graph, memory), to(impl.To(), graph, memory) {}
mgp_edge(const mgp_edge &other, memgraph::utils::MemoryResource *memory) noexcept
: memory(memory), impl(other.impl), from(other.from, memory), to(other.to, memory) {}
mgp_edge(mgp_edge &&other, memgraph::utils::MemoryResource *memory) noexcept
: memory(other.memory), impl(other.impl), from(std::move(other.from), memory), to(std::move(other.to), memory) {}
mgp_edge(mgp_edge &&other) noexcept
: memory(other.memory), impl(other.impl), from(std::move(other.from)), to(std::move(other.to)) {}
/// Copy construction without memgraph::utils::MemoryResource is not allowed.
mgp_edge(const mgp_edge &) = delete;
mgp_edge &operator=(const mgp_edge &) = delete;
mgp_edge &operator=(mgp_edge &&) = delete;
~mgp_edge() = default;
bool operator==(const mgp_edge &other) const noexcept { return this->impl == other.impl; }
bool operator!=(const mgp_edge &other) const noexcept { return !(*this == other); };
memgraph::utils::MemoryResource *GetMemoryResource() const noexcept { return memory; }
memgraph::utils::MemoryResource *memory;
memgraph::query::v2::EdgeAccessor impl;
mgp_vertex from;
mgp_vertex to;
};
struct mgp_path {
/// Allocator type so that STL containers are aware that we need one.
using allocator_type = memgraph::utils::Allocator<mgp_path>;
explicit mgp_path(memgraph::utils::MemoryResource *memory) : vertices(memory), edges(memory) {}
mgp_path(const mgp_path &other, memgraph::utils::MemoryResource *memory)
: vertices(other.vertices, memory), edges(other.edges, memory) {}
mgp_path(mgp_path &&other, memgraph::utils::MemoryResource *memory)
: vertices(std::move(other.vertices), memory), edges(std::move(other.edges), memory) {}
mgp_path(mgp_path &&other) noexcept : vertices(std::move(other.vertices)), edges(std::move(other.edges)) {}
/// Copy construction without memgraph::utils::MemoryResource is not allowed.
mgp_path(const mgp_path &) = delete;
mgp_path &operator=(const mgp_path &) = delete;
mgp_path &operator=(mgp_path &&) = delete;
~mgp_path() = default;
memgraph::utils::MemoryResource *GetMemoryResource() const noexcept {
return vertices.get_allocator().GetMemoryResource();
}
memgraph::utils::pmr::vector<mgp_vertex> vertices;
memgraph::utils::pmr::vector<mgp_edge> edges;
};
struct mgp_result_record {
/// Result record signature as defined for mgp_proc.
const memgraph::utils::pmr::map<memgraph::utils::pmr::string,
std::pair<const memgraph::query::v2::procedure::CypherType *, bool>> *signature;
memgraph::utils::pmr::map<memgraph::utils::pmr::string, memgraph::query::v2::TypedValue> values;
};
struct mgp_result {
explicit mgp_result(
const memgraph::utils::pmr::map<memgraph::utils::pmr::string,
std::pair<const memgraph::query::v2::procedure::CypherType *, bool>> *signature,
memgraph::utils::MemoryResource *mem)
: signature(signature), rows(mem) {}
/// Result record signature as defined for mgp_proc.
const memgraph::utils::pmr::map<memgraph::utils::pmr::string,
std::pair<const memgraph::query::v2::procedure::CypherType *, bool>> *signature;
memgraph::utils::pmr::vector<mgp_result_record> rows;
std::optional<memgraph::utils::pmr::string> error_msg;
};
struct mgp_func_result {
mgp_func_result() {}
/// Return Magic function result. If user forgets it, the error is raised
std::optional<memgraph::query::v2::TypedValue> value;
/// Return Magic function result with potential error
std::optional<memgraph::utils::pmr::string> error_msg;
};
struct mgp_graph {
memgraph::query::v2::DbAccessor *impl;
memgraph::storage::v3::View view;
// TODO: Merge `mgp_graph` and `mgp_memory` into a single `mgp_context`. The
// `ctx` field is out of place here.
memgraph::query::v2::ExecutionContext *ctx;
static mgp_graph WritableGraph(memgraph::query::v2::DbAccessor &acc, memgraph::storage::v3::View view,
memgraph::query::v2::ExecutionContext &ctx) {
return mgp_graph{&acc, view, &ctx};
}
static mgp_graph NonWritableGraph(memgraph::query::v2::DbAccessor &acc, memgraph::storage::v3::View view) {
return mgp_graph{&acc, view, nullptr};
}
};
// Prevents user to use ExecutionContext in writable callables
struct mgp_func_context {
memgraph::query::v2::DbAccessor *impl;
memgraph::storage::v3::View view;
};
struct mgp_properties_iterator {
using allocator_type = memgraph::utils::Allocator<mgp_properties_iterator>;
// Define members at the start because we use decltype a lot here, so members
// need to be visible in method definitions.
memgraph::utils::MemoryResource *memory;
mgp_graph *graph;
std::remove_reference_t<decltype(*std::declval<memgraph::query::v2::VertexAccessor>().Properties(graph->view))> pvs;
decltype(pvs.begin()) current_it;
std::optional<std::pair<memgraph::utils::pmr::string, mgp_value>> current;
mgp_property property{nullptr, nullptr};
// Construct with no properties.
explicit mgp_properties_iterator(mgp_graph *graph, memgraph::utils::MemoryResource *memory)
: memory(memory), graph(graph), current_it(pvs.begin()) {}
// May throw who the #$@! knows what because PropertyValueStore doesn't
// document what it throws, and it may surely throw some piece of !@#$
// exception because it's built on top of STL and other libraries.
mgp_properties_iterator(mgp_graph *graph, decltype(pvs) pvs, memgraph::utils::MemoryResource *memory)
: memory(memory), graph(graph), pvs(std::move(pvs)), current_it(this->pvs.begin()) {
if (current_it != this->pvs.end()) {
current.emplace(memgraph::utils::pmr::string(graph->impl->PropertyToName(current_it->first), memory),
mgp_value(current_it->second, memory));
property.name = current->first.c_str();
property.value = &current->second;
}
}
mgp_properties_iterator(const mgp_properties_iterator &) = delete;
mgp_properties_iterator(mgp_properties_iterator &&) = delete;
mgp_properties_iterator &operator=(const mgp_properties_iterator &) = delete;
mgp_properties_iterator &operator=(mgp_properties_iterator &&) = delete;
~mgp_properties_iterator() = default;
memgraph::utils::MemoryResource *GetMemoryResource() const { return memory; }
};
struct mgp_edges_iterator {
using allocator_type = memgraph::utils::Allocator<mgp_edges_iterator>;
// Hopefully mgp_vertex copy constructor remains noexcept, so that we can
// have everything noexcept here.
static_assert(std::is_nothrow_constructible_v<mgp_vertex, const mgp_vertex &, memgraph::utils::MemoryResource *>);
mgp_edges_iterator(const mgp_vertex &v, memgraph::utils::MemoryResource *memory) noexcept
: memory(memory), source_vertex(v, memory) {}
mgp_edges_iterator(mgp_edges_iterator &&other) noexcept
: memory(other.memory),
source_vertex(std::move(other.source_vertex)),
in(std::move(other.in)),
in_it(std::move(other.in_it)),
out(std::move(other.out)),
out_it(std::move(other.out_it)),
current_e(std::move(other.current_e)) {}
mgp_edges_iterator(const mgp_edges_iterator &) = delete;
mgp_edges_iterator &operator=(const mgp_edges_iterator &) = delete;
mgp_edges_iterator &operator=(mgp_edges_iterator &&) = delete;
~mgp_edges_iterator() = default;
memgraph::utils::MemoryResource *GetMemoryResource() const { return memory; }
memgraph::utils::MemoryResource *memory;
mgp_vertex source_vertex;
std::optional<std::remove_reference_t<decltype(*source_vertex.impl.InEdges(source_vertex.graph->view))>> in;
std::optional<decltype(in->begin())> in_it;
std::optional<std::remove_reference_t<decltype(*source_vertex.impl.OutEdges(source_vertex.graph->view))>> out;
std::optional<decltype(out->begin())> out_it;
std::optional<mgp_edge> current_e;
};
struct mgp_vertices_iterator {
using allocator_type = memgraph::utils::Allocator<mgp_vertices_iterator>;
/// @throw anything VerticesIterable may throw
mgp_vertices_iterator(mgp_graph *graph, memgraph::utils::MemoryResource *memory)
: memory(memory), graph(graph), vertices(graph->impl->Vertices(graph->view)), current_it(vertices.begin()) {
if (current_it != vertices.end()) {
current_v.emplace(*current_it, graph, memory);
}
}
memgraph::utils::MemoryResource *GetMemoryResource() const { return memory; }
memgraph::utils::MemoryResource *memory;
mgp_graph *graph;
decltype(graph->impl->Vertices(graph->view)) vertices;
decltype(vertices.begin()) current_it;
std::optional<mgp_vertex> current_v;
};
struct mgp_type {
memgraph::query::v2::procedure::CypherTypePtr impl;
};
struct ProcedureInfo {
bool is_write = false;
std::optional<memgraph::query::v2::AuthQuery::Privilege> required_privilege = std::nullopt;
};
struct mgp_proc {
using allocator_type = memgraph::utils::Allocator<mgp_proc>;
/// @throw std::bad_alloc
/// @throw std::length_error
mgp_proc(const char *name, mgp_proc_cb cb, memgraph::utils::MemoryResource *memory, const ProcedureInfo &info = {})
: name(name, memory), cb(cb), args(memory), opt_args(memory), results(memory), info(info) {}
/// @throw std::bad_alloc
/// @throw std::length_error
mgp_proc(const char *name, std::function<void(mgp_list *, mgp_graph *, mgp_result *, mgp_memory *)> cb,
memgraph::utils::MemoryResource *memory, const ProcedureInfo &info = {})
: name(name, memory), cb(cb), args(memory), opt_args(memory), results(memory), info(info) {}
/// @throw std::bad_alloc
/// @throw std::length_error
mgp_proc(const std::string_view name, std::function<void(mgp_list *, mgp_graph *, mgp_result *, mgp_memory *)> cb,
memgraph::utils::MemoryResource *memory, const ProcedureInfo &info = {})
: name(name, memory), cb(cb), args(memory), opt_args(memory), results(memory), info(info) {}
/// @throw std::bad_alloc
/// @throw std::length_error
mgp_proc(const mgp_proc &other, memgraph::utils::MemoryResource *memory)
: name(other.name, memory),
cb(other.cb),
args(other.args, memory),
opt_args(other.opt_args, memory),
results(other.results, memory),
info(other.info) {}
mgp_proc(mgp_proc &&other, memgraph::utils::MemoryResource *memory)
: name(std::move(other.name), memory),
cb(std::move(other.cb)),
args(std::move(other.args), memory),
opt_args(std::move(other.opt_args), memory),
results(std::move(other.results), memory),
info(other.info) {}
mgp_proc(const mgp_proc &other) = default;
mgp_proc(mgp_proc &&other) = default;
mgp_proc &operator=(const mgp_proc &) = delete;
mgp_proc &operator=(mgp_proc &&) = delete;
~mgp_proc() = default;
/// Name of the procedure.
memgraph::utils::pmr::string name;
/// Entry-point for the procedure.
std::function<void(mgp_list *, mgp_graph *, mgp_result *, mgp_memory *)> cb;
/// Required, positional arguments as a (name, type) pair.
memgraph::utils::pmr::vector<
std::pair<memgraph::utils::pmr::string, const memgraph::query::v2::procedure::CypherType *>>
args;
/// Optional positional arguments as a (name, type, default_value) tuple.
memgraph::utils::pmr::vector<
std::tuple<memgraph::utils::pmr::string, const memgraph::query::v2::procedure::CypherType *,
memgraph::query::v2::TypedValue>>
opt_args;
/// Fields this procedure returns, as a (name -> (type, is_deprecated)) map.
memgraph::utils::pmr::map<memgraph::utils::pmr::string,
std::pair<const memgraph::query::v2::procedure::CypherType *, bool>>
results;
ProcedureInfo info;
};
struct mgp_trans {
using allocator_type = memgraph::utils::Allocator<mgp_trans>;
/// @throw std::bad_alloc
/// @throw std::length_error
mgp_trans(const char *name, mgp_trans_cb cb, memgraph::utils::MemoryResource *memory)
: name(name, memory), cb(cb), results(memory) {}
/// @throw std::bad_alloc
/// @throw std::length_error
mgp_trans(const char *name, std::function<void(mgp_messages *, mgp_graph *, mgp_result *, mgp_memory *)> cb,
memgraph::utils::MemoryResource *memory)
: name(name, memory), cb(cb), results(memory) {}
/// @throw std::bad_alloc
/// @throw std::length_error
mgp_trans(const mgp_trans &other, memgraph::utils::MemoryResource *memory)
: name(other.name, memory), cb(other.cb), results(other.results) {}
mgp_trans(mgp_trans &&other, memgraph::utils::MemoryResource *memory)
: name(std::move(other.name), memory), cb(std::move(other.cb)), results(std::move(other.results)) {}
mgp_trans(const mgp_trans &other) = default;
mgp_trans(mgp_trans &&other) = default;
mgp_trans &operator=(const mgp_trans &) = delete;
mgp_trans &operator=(mgp_trans &&) = delete;
~mgp_trans() = default;
/// Name of the transformation.
memgraph::utils::pmr::string name;
/// Entry-point for the transformation.
std::function<void(mgp_messages *, mgp_graph *, mgp_result *, mgp_memory *)> cb;
/// Fields this transformation returns.
memgraph::utils::pmr::map<memgraph::utils::pmr::string,
std::pair<const memgraph::query::v2::procedure::CypherType *, bool>>
results;
};
struct mgp_func {
using allocator_type = memgraph::utils::Allocator<mgp_func>;
/// @throw std::bad_alloc
/// @throw std::length_error
mgp_func(const char *name, mgp_func_cb cb, memgraph::utils::MemoryResource *memory)
: name(name, memory), cb(cb), args(memory), opt_args(memory) {}
/// @throw std::bad_alloc
/// @throw std::length_error
mgp_func(const char *name, std::function<void(mgp_list *, mgp_func_context *, mgp_func_result *, mgp_memory *)> cb,
memgraph::utils::MemoryResource *memory)
: name(name, memory), cb(cb), args(memory), opt_args(memory) {}
/// @throw std::bad_alloc
/// @throw std::length_error
mgp_func(const mgp_func &other, memgraph::utils::MemoryResource *memory)
: name(other.name, memory), cb(other.cb), args(other.args, memory), opt_args(other.opt_args, memory) {}
mgp_func(mgp_func &&other, memgraph::utils::MemoryResource *memory)
: name(std::move(other.name), memory),
cb(std::move(other.cb)),
args(std::move(other.args), memory),
opt_args(std::move(other.opt_args), memory) {}
mgp_func(const mgp_func &other) = default;
mgp_func(mgp_func &&other) = default;
mgp_func &operator=(const mgp_func &) = delete;
mgp_func &operator=(mgp_func &&) = delete;
~mgp_func() = default;
/// Name of the function.
memgraph::utils::pmr::string name;
/// Entry-point for the function.
std::function<void(mgp_list *, mgp_func_context *, mgp_func_result *, mgp_memory *)> cb;
/// Required, positional arguments as a (name, type) pair.
memgraph::utils::pmr::vector<
std::pair<memgraph::utils::pmr::string, const memgraph::query::v2::procedure::CypherType *>>
args;
/// Optional positional arguments as a (name, type, default_value) tuple.
memgraph::utils::pmr::vector<
std::tuple<memgraph::utils::pmr::string, const memgraph::query::v2::procedure::CypherType *,
memgraph::query::v2::TypedValue>>
opt_args;
};
mgp_error MgpTransAddFixedResult(mgp_trans *trans) noexcept;
struct mgp_module {
using allocator_type = memgraph::utils::Allocator<mgp_module>;
explicit mgp_module(memgraph::utils::MemoryResource *memory)
: procedures(memory), transformations(memory), functions(memory) {}
mgp_module(const mgp_module &other, memgraph::utils::MemoryResource *memory)
: procedures(other.procedures, memory),
transformations(other.transformations, memory),
functions(other.functions, memory) {}
mgp_module(mgp_module &&other, memgraph::utils::MemoryResource *memory)
: procedures(std::move(other.procedures), memory),
transformations(std::move(other.transformations), memory),
functions(std::move(other.functions), memory) {}
mgp_module(const mgp_module &) = default;
mgp_module(mgp_module &&) = default;
mgp_module &operator=(const mgp_module &) = delete;
mgp_module &operator=(mgp_module &&) = delete;
~mgp_module() = default;
memgraph::utils::pmr::map<memgraph::utils::pmr::string, mgp_proc> procedures;
memgraph::utils::pmr::map<memgraph::utils::pmr::string, mgp_trans> transformations;
memgraph::utils::pmr::map<memgraph::utils::pmr::string, mgp_func> functions;
};
namespace memgraph::query::v2::procedure {
/// @throw std::bad_alloc
/// @throw std::length_error
/// @throw anything std::ostream::operator<< may throw.
void PrintProcSignature(const mgp_proc &, std::ostream *);
/// @throw std::bad_alloc
/// @throw std::length_error
/// @throw anything std::ostream::operator<< may throw.
void PrintFuncSignature(const mgp_func &, std::ostream &);
bool IsValidIdentifierName(const char *name);
} // namespace memgraph::query::v2::procedure
struct mgp_message {
explicit mgp_message(const memgraph::integrations::kafka::Message &message) : msg{&message} {}
explicit mgp_message(const memgraph::integrations::pulsar::Message &message) : msg{message} {}
using KafkaMessage = const memgraph::integrations::kafka::Message *;
using PulsarMessage = memgraph::integrations::pulsar::Message;
std::variant<KafkaMessage, PulsarMessage> msg;
};
struct mgp_messages {
using allocator_type = memgraph::utils::Allocator<mgp_messages>;
using storage_type = memgraph::utils::pmr::vector<mgp_message>;
explicit mgp_messages(storage_type &&storage) : messages(std::move(storage)) {}
mgp_messages(const mgp_messages &) = delete;
mgp_messages &operator=(const mgp_messages &) = delete;
mgp_messages(mgp_messages &&) = delete;
mgp_messages &operator=(mgp_messages &&) = delete;
~mgp_messages() = default;
storage_type messages;
};
memgraph::query::v2::TypedValue ToTypedValue(const mgp_value &val, memgraph::utils::MemoryResource *memory);

File diff suppressed because it is too large Load Diff

View File

@@ -1,246 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
/// @file
/// API for loading and registering modules providing custom oC procedures
#pragma once
#include <dlfcn.h>
#include <filesystem>
#include <functional>
#include <optional>
#include <shared_mutex>
#include <string>
#include <string_view>
#include <unordered_map>
#include "query/v2/procedure/cypher_types.hpp"
#include "query/v2/procedure/mg_procedure_impl.hpp"
#include "utils/memory.hpp"
#include "utils/rw_lock.hpp"
class CypherMainVisitorTest;
namespace memgraph::query::v2::procedure {
class Module {
public:
Module() {}
virtual ~Module();
Module(const Module &) = delete;
Module(Module &&) = delete;
Module &operator=(const Module &) = delete;
Module &operator=(Module &&) = delete;
/// Invokes the (optional) shutdown function and closes the module.
virtual bool Close() = 0;
/// Returns registered procedures of this module
virtual const std::map<std::string, mgp_proc, std::less<>> *Procedures() const = 0;
/// Returns registered transformations of this module
virtual const std::map<std::string, mgp_trans, std::less<>> *Transformations() const = 0;
// /// Returns registered functions of this module
virtual const std::map<std::string, mgp_func, std::less<>> *Functions() const = 0;
virtual std::optional<std::filesystem::path> Path() const = 0;
};
/// Proxy for a registered Module, acquires a read lock from ModuleRegistry.
class ModulePtr final {
const Module *module_{nullptr};
std::shared_lock<utils::RWLock> lock_;
public:
ModulePtr() = default;
ModulePtr(std::nullptr_t) {}
ModulePtr(const Module *module, std::shared_lock<utils::RWLock> lock) : module_(module), lock_(std::move(lock)) {}
explicit operator bool() const { return static_cast<bool>(module_); }
const Module &operator*() const { return *module_; }
const Module *operator->() const { return module_; }
};
/// Thread-safe registration of modules from libraries, uses utils::RWLock.
class ModuleRegistry final {
friend CypherMainVisitorTest;
std::map<std::string, std::unique_ptr<Module>, std::less<>> modules_;
mutable utils::RWLock lock_{utils::RWLock::Priority::WRITE};
std::unique_ptr<utils::MemoryResource> shared_{std::make_unique<utils::ResourceWithOutOfMemoryException>()};
bool RegisterModule(std::string_view name, std::unique_ptr<Module> module);
void DoUnloadAllModules();
/// Loads the module if it's in the modules_dir directory
/// @return Whether the module was loaded
bool LoadModuleIfFound(const std::filesystem::path &modules_dir, std::string_view name);
void LoadModulesFromDirectory(const std::filesystem::path &modules_dir);
public:
ModuleRegistry();
/// Set the modules directories that will be used when (re)loading modules.
void SetModulesDirectory(std::vector<std::filesystem::path> modules_dir, const std::filesystem::path &data_directory);
const std::vector<std::filesystem::path> &GetModulesDirectory() const;
/// Atomically load or reload a module with a particular name from the given
/// directory.
///
/// Takes a write lock. If the module exists it is reloaded. Otherwise, the
/// module is loaded from the file whose filename, without the extension,
/// matches the module's name. If multiple such files exist, only one is
/// chosen, in an unspecified manner. If loading of the chosen file fails, no
/// other files are tried.
///
/// Return true if the module was loaded or reloaded successfully, false
/// otherwise.
bool LoadOrReloadModuleFromName(std::string_view name);
/// Atomically unload all modules and then load all possible modules from the
/// set directories.
///
/// Takes a write lock.
void UnloadAndLoadModulesFromDirectories();
/// Find a module with given name or return nullptr.
/// Takes a read lock.
ModulePtr GetModuleNamed(std::string_view name) const;
/// Remove all loaded (non-builtin) modules.
/// Takes a write lock.
void UnloadAllModules();
/// Returns the shared memory allocator used by modules
utils::MemoryResource &GetSharedMemoryResource() noexcept;
bool RegisterMgProcedure(std::string_view name, mgp_proc proc);
const std::filesystem::path &InternalModuleDir() const noexcept;
private:
class SharedLibraryHandle {
public:
SharedLibraryHandle(const std::string &shared_library, int mode) : handle_{dlopen(shared_library.c_str(), mode)} {}
SharedLibraryHandle(const SharedLibraryHandle &) = delete;
SharedLibraryHandle(SharedLibraryHandle &&) = delete;
SharedLibraryHandle operator=(const SharedLibraryHandle &) = delete;
SharedLibraryHandle operator=(SharedLibraryHandle &&) = delete;
~SharedLibraryHandle() {
if (handle_) {
dlclose(handle_);
}
}
private:
void *handle_;
};
#if __has_feature(address_sanitizer)
// This is why we need RTLD_NODELETE and we must not use RTLD_DEEPBIND with
// ASAN: https://github.com/google/sanitizers/issues/89
SharedLibraryHandle libstd_handle{"libstdc++.so.6", RTLD_NOW | RTLD_LOCAL | RTLD_NODELETE};
#else
// The reason behind opening share library during runtime is to avoid issues
// with loading symbols from stdlib. We have encounter issues with locale
// that cause std::cout not being printed and issues when python libraries
// would call stdlib (e.g. pytorch).
// The way that those issues were solved was
// by using RTLD_DEEPBIND. RTLD_DEEPBIND ensures that the lookup for the
// mentioned library will be first performed in the already existing binded
// libraries and then the global namespace.
// RTLD_DEEPBIND => https://linux.die.net/man/3/dlopen
SharedLibraryHandle libstd_handle{"libstdc++.so.6", RTLD_NOW | RTLD_LOCAL | RTLD_DEEPBIND};
#endif
std::vector<std::filesystem::path> modules_dirs_;
std::filesystem::path internal_module_dir_;
};
/// Single, global module registry.
extern ModuleRegistry gModuleRegistry;
/// Return the ModulePtr and `mgp_proc *` of the found procedure after resolving
/// `fully_qualified_procedure_name`. `memory` is used for temporary allocations
/// inside this function. ModulePtr must be kept alive to make sure it won't be
/// unloaded.
std::optional<std::pair<procedure::ModulePtr, const mgp_proc *>> FindProcedure(
const ModuleRegistry &module_registry, std::string_view fully_qualified_procedure_name,
utils::MemoryResource *memory);
/// Return the ModulePtr and `mgp_trans *` of the found transformation after resolving
/// `fully_qualified_transformation_name`. `memory` is used for temporary allocations
/// inside this function. ModulePtr must be kept alive to make sure it won't be
/// unloaded.
std::optional<std::pair<procedure::ModulePtr, const mgp_trans *>> FindTransformation(
const ModuleRegistry &module_registry, std::string_view fully_qualified_transformation_name,
utils::MemoryResource *memory);
/// Return the ModulePtr and `mgp_func *` of the found function after resolving
/// `fully_qualified_function_name` if found. If there is no such function
/// std::nullopt is returned. `memory` is used for temporary allocations
/// inside this function. ModulePtr must be kept alive to make sure it won't be unloaded.
std::optional<std::pair<procedure::ModulePtr, const mgp_func *>> FindFunction(
const ModuleRegistry &module_registry, std::string_view fully_qualified_function_name,
utils::MemoryResource *memory);
template <typename T>
concept IsCallable = utils::SameAsAnyOf<T, mgp_proc, mgp_func>;
template <IsCallable TCall>
void ConstructArguments(const std::vector<TypedValue> &args, const TCall &callable,
const std::string_view fully_qualified_name, mgp_list &args_list, mgp_graph &graph) {
const auto n_args = args.size();
const auto c_args_sz = callable.args.size();
const auto c_opt_args_sz = callable.opt_args.size();
if (n_args < c_args_sz || (n_args - c_args_sz > c_opt_args_sz)) {
if (callable.args.empty() && callable.opt_args.empty()) {
throw QueryRuntimeException("'{}' requires no arguments.", fully_qualified_name);
}
if (callable.opt_args.empty()) {
throw QueryRuntimeException("'{}' requires exactly {} {}.", fully_qualified_name, c_args_sz,
c_args_sz == 1U ? "argument" : "arguments");
}
throw QueryRuntimeException("'{}' requires between {} and {} arguments.", fully_qualified_name, c_args_sz,
c_args_sz + c_opt_args_sz);
}
args_list.elems.reserve(n_args);
auto is_not_optional_arg = [c_args_sz](int i) { return c_args_sz > i; };
for (size_t i = 0; i < n_args; ++i) {
auto arg = args[i];
std::string_view name;
const query::v2::procedure::CypherType *type;
if (is_not_optional_arg(i)) {
name = callable.args[i].first;
type = callable.args[i].second;
} else {
name = std::get<0>(callable.opt_args[i - c_args_sz]);
type = std::get<1>(callable.opt_args[i - c_args_sz]);
}
if (!type->SatisfiesType(arg)) {
throw QueryRuntimeException("'{}' argument named '{}' at position {} must be of type {}.", fully_qualified_name,
name, i, type->GetPresentableName());
}
args_list.elems.emplace_back(std::move(arg), &graph);
}
// Fill missing optional arguments with their default values.
const size_t passed_in_opt_args = n_args - c_args_sz;
for (size_t i = passed_in_opt_args; i < c_opt_args_sz; ++i) {
args_list.elems.emplace_back(std::get<2>(callable.opt_args[i]), &graph);
}
}
} // namespace memgraph::query::v2::procedure

File diff suppressed because it is too large Load Diff

View File

@@ -1,82 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
/// @file
/// Functions and types for loading Query Modules written in Python.
#pragma once
#include "py/py.hpp"
struct mgp_graph;
struct mgp_memory;
struct mgp_module;
struct mgp_value;
namespace memgraph::query::v2::procedure {
struct PyGraph;
/// Convert an `mgp_value` into a Python object, referencing the given `PyGraph`
/// instance and using the same allocator as the graph.
///
/// Values of type `MGP_VALUE_TYPE_VERTEX`, `MGP_VALUE_TYPE_EDGE` and
/// `MGP_VALUE_TYPE_PATH` are returned as `mgp.Vertex`, `mgp.Edge` and
/// `mgp.Path` respectively, and *not* their internal `_mgp`
/// representations. Other value types are converted to equivalent builtin
/// Python objects.
///
/// Return a non-null `py::Object` instance on success. Otherwise, return a null
/// `py::Object` instance and set the appropriate Python exception.
py::Object MgpValueToPyObject(const mgp_value &value, PyGraph *py_graph);
py::Object MgpValueToPyObject(const mgp_value &value, PyObject *py_graph);
/// Convert a Python object into `mgp_value`, constructing it using the given
/// `mgp_memory` allocator.
///
/// If the user-facing 'mgp' module can be imported, this function will handle
/// conversion of 'mgp.Vertex', 'mgp.Edge' and 'mgp.Path' values.
///
/// @throw std::bad_alloc
/// @throw std::overflow_error if attempting to convert a Python integer which
/// too large to fit into int64_t.
/// @throw std::invalid_argument if the given Python object cannot be converted
/// to an mgp_value (e.g. a dictionary whose keys aren't strings or an object
/// of unsupported type).
mgp_value *PyObjectToMgpValue(PyObject *, mgp_memory *);
/// Create the _mgp module for use in embedded Python.
///
/// The function is to be used before Py_Initialize via the following code.
///
/// PyImport_AppendInittab("_mgp", &query::v2::procedure::PyInitMgpModule);
PyObject *PyInitMgpModule();
/// Create an instance of _mgp.Graph class.
PyObject *MakePyGraph(mgp_graph *, mgp_memory *);
/// Import a module with given name in the context of mgp_module.
///
/// This function can only be called when '_mgp' module has been initialized in
/// Python.
///
/// Return nullptr and set appropriate Python exception on failure.
py::Object ImportPyModule(const char *, mgp_module *);
/// Reload already loaded Python module in the context of mgp_module.
///
/// This function can only be called when '_mgp' module has been initialized in
/// Python.
///
/// Return nullptr and set appropriate Python exception on failure.
py::Object ReloadPyModule(PyObject *, mgp_module *);
} // namespace memgraph::query::v2::procedure

View File

@@ -14,6 +14,7 @@
#include <chrono>
#include <iostream>
#include <map>
#include <memory>
#include <optional>
#include <unordered_map>
#include <utility>
@@ -23,18 +24,10 @@
#include "coordinator/hybrid_logical_clock.hpp"
#include "storage/v3/id_types.hpp"
#include "storage/v3/property_value.hpp"
#include "utils/fnv.hpp"
namespace memgraph::msgs {
// TODO(gvolfing) introduce structure for property updates
// TODO(gvolfing) remove tuples from ExpandOne and add specific structs instead
// TODO(gvolfing) discuss with kostas:
// - EdgeId
// - Everything about ExpandOne
// - Edge properties
// - Ctors
using coordinator::Hlc;
using storage::v3::LabelId;
@@ -42,48 +35,46 @@ struct Value;
struct Label {
LabelId id;
friend bool operator==(const Label &lhs, const Label &rhs) { return lhs.id == rhs.id; }
};
// TODO(kostasrim) update this with CompoundKey, same for the rest of the file.
using PrimaryKey = std::vector<Value>;
using VertexId = std::pair<Label, PrimaryKey>;
inline bool operator==(const VertexId &lhs, const VertexId &rhs) {
return (lhs.first == rhs.first) && (lhs.second == rhs.second);
}
using Gid = size_t;
using PropertyId = memgraph::storage::v3::PropertyId;
// struct PropertyUpdate{
// PropertyId id;
// Value value;
// };
using EdgeTypeId = memgraph::storage::v3::EdgeTypeId;
struct EdgeType {
uint64_t id;
EdgeTypeId id;
friend bool operator==(const EdgeType &lhs, const EdgeType &rhs) = default;
};
struct EdgeId {
Gid gid;
friend bool operator==(const EdgeId &lhs, const EdgeId &rhs) { return lhs.gid == rhs.gid; }
friend bool operator<(const EdgeId &lhs, const EdgeId &rhs) { return lhs.gid < rhs.gid; }
};
struct Edge {
VertexId src;
VertexId dst;
std::optional<std::vector<std::pair<PropertyId, Value>>> properties;
std::vector<std::pair<PropertyId, Value>> properties;
EdgeId id;
EdgeType type;
friend bool operator==(const Edge &lhs, const Edge &rhs) { return lhs.id == rhs.id; }
};
struct Vertex {
VertexId id;
std::vector<Label> labels;
};
struct PathPart {
Vertex dst;
Gid edge;
};
struct Path {
Vertex src;
std::vector<PathPart> parts;
friend bool operator==(const Vertex &lhs, const Vertex &rhs) { return lhs.id == rhs.id; }
};
struct Null {};
@@ -96,6 +87,7 @@ struct Value {
explicit Value(const double val) : type(Type::Double), double_v(val) {}
explicit Value(const Vertex val) : type(Type::Vertex), vertex_v(val) {}
explicit Value(const Edge val) : type(Type::Edge), edge_v(val) {}
explicit Value(const std::string &val) : type(Type::String) { new (&string_v) std::string(val); }
explicit Value(const char *val) : type(Type::String) { new (&string_v) std::string(val); }
@@ -134,13 +126,9 @@ struct Value {
case Type::Map:
std::destroy_at(&map_v);
return;
case Type::Vertex:
std::destroy_at(&vertex_v);
return;
case Type::Path:
std::destroy_at(&path_v);
return;
case Type::Edge:
std::destroy_at(&edge_v);
}
@@ -174,9 +162,6 @@ struct Value {
case Type::Edge:
new (&edge_v) Edge(other.edge_v);
return;
case Type::Path:
new (&path_v) Path(other.path_v);
return;
}
}
@@ -208,9 +193,6 @@ struct Value {
case Type::Edge:
new (&edge_v) Edge(std::move(other.edge_v));
break;
case Type::Path:
new (&path_v) Path(std::move(other.path_v));
break;
}
other.DestroyValue();
@@ -250,9 +232,6 @@ struct Value {
case Type::Edge:
new (&edge_v) Edge(other.edge_v);
break;
case Type::Path:
new (&path_v) Path(other.path_v);
break;
}
return *this;
@@ -291,9 +270,6 @@ struct Value {
case Type::Edge:
new (&edge_v) Edge(std::move(other.edge_v));
break;
case Type::Path:
new (&path_v) Path(std::move(other.path_v));
break;
}
other.DestroyValue();
@@ -301,7 +277,7 @@ struct Value {
return *this;
}
enum class Type : uint8_t { Null, Bool, Int64, Double, String, List, Map, Vertex, Edge, Path };
enum class Type : uint8_t { Null, Bool, Int64, Double, String, List, Map, Vertex, Edge };
Type type{Type::Null};
union {
Null null_v;
@@ -313,24 +289,35 @@ struct Value {
std::map<std::string, Value> map_v;
Vertex vertex_v;
Edge edge_v;
Path path_v;
};
};
struct ValuesMap {
std::unordered_map<PropertyId, Value> values_map;
friend bool operator==(const Value &lhs, const Value &rhs) {
if (lhs.type != rhs.type) {
return false;
}
switch (lhs.type) {
case Value::Type::Null:
return true;
case Value::Type::Bool:
return lhs.bool_v == rhs.bool_v;
case Value::Type::Int64:
return lhs.int_v == rhs.int_v;
case Value::Type::Double:
return lhs.double_v == rhs.double_v;
case Value::Type::String:
return lhs.string_v == rhs.string_v;
case Value::Type::List:
return lhs.list_v == rhs.list_v;
case Value::Type::Map:
return lhs.map_v == rhs.map_v;
case Value::Type::Vertex:
return lhs.vertex_v == rhs.vertex_v;
case Value::Type::Edge:
return lhs.edge_v == rhs.edge_v;
}
}
};
struct MappedValues {
std::vector<ValuesMap> values_map;
};
struct ListedValues {
std::vector<std::vector<Value>> properties;
};
using Values = std::variant<ListedValues, MappedValues>;
struct Expression {
std::string expression;
};
@@ -350,17 +337,29 @@ enum class StorageView { OLD = 0, NEW = 1 };
struct ScanVerticesRequest {
Hlc transaction_id;
// This should be optional
VertexId start_id;
std::vector<VertexId> scanned_vertices;
// The empty optional means return all of the properties, while an empty list means do not return any properties
std::optional<std::vector<PropertyId>> props_to_return;
std::optional<std::vector<std::string>> filter_expressions;
// expression that determines if vertex is returned or not
std::vector<std::string> filter_expressions;
// expression whose result is returned for every vertex
std::vector<std::string> vertex_expressions;
std::optional<size_t> batch_limit;
StorageView storage_view;
std::vector<OrderBy> order_bys;
StorageView storage_view{StorageView::NEW};
std::optional<Label> label;
std::optional<std::pair<PropertyId, std::string>> property_expression_pair;
};
struct ScanResultRow {
Value vertex;
Vertex vertex;
// empty() is no properties returned
std::map<PropertyId, Value> props;
// This should be changed to std::map<PropertyId, Value>
std::vector<std::pair<PropertyId, Value>> props;
std::vector<Value> evaluated_vertex_expressions;
};
struct ScanVerticesResponse {
@@ -373,6 +372,7 @@ using VertexOrEdgeIds = std::variant<VertexId, EdgeId>;
struct GetPropertiesRequest {
Hlc transaction_id;
// Shouldn't contain mixed vertex and edge ids
VertexOrEdgeIds vertex_or_edge_ids;
std::vector<PropertyId> property_ids;
std::vector<Expression> expressions;
@@ -384,7 +384,6 @@ struct GetPropertiesRequest {
struct GetPropertiesResponse {
bool success;
Values values;
};
enum class EdgeDirection : uint8_t { OUT = 1, IN = 2, BOTH = 3 };
@@ -392,31 +391,41 @@ enum class EdgeDirection : uint8_t { OUT = 1, IN = 2, BOTH = 3 };
struct ExpandOneRequest {
Hlc transaction_id;
std::vector<VertexId> src_vertices;
// return types that type is in this list
// empty means all the types
std::vector<EdgeType> edge_types;
EdgeDirection direction;
EdgeDirection direction{EdgeDirection::OUT};
// Wether to return multiple edges between the same neighbors
bool only_unique_neighbor_rows = false;
// The empty optional means return all of the properties, while an empty
// list means do not return any properties
// TODO(antaljanosbenjamin): All of the special values should be communicated through a single vertex object
// after schema is implemented
// Special values are accepted:
// * __mg__labels
// The empty optional means return all of the properties, while an empty list means do not return any properties
std::optional<std::vector<PropertyId>> src_vertex_properties;
// TODO(antaljanosbenjamin): All of the special values should be communicated through a single vertex object
// after schema is implemented
// Special values are accepted:
// * __mg__dst_id (Vertex, but without labels)
// * __mg__type (binary)
// The empty optional means return all of the properties, while an empty list means do not return any properties
std::optional<std::vector<PropertyId>> edge_properties;
// QUESTION(antaljanosbenjamin): Maybe also add possibility to expressions evaluated on the source vertex?
// List of expressions evaluated on edges
std::vector<Expression> expressions;
std::vector<std::string> vertex_expressions;
std::vector<std::string> edge_expressions;
std::optional<std::vector<OrderBy>> order_by;
// Limit the edges or the vertices?
std::optional<size_t> limit;
std::optional<Filter> filter;
std::vector<std::string> filters;
};
struct ExpandOneResultRow {
struct EdgeWithAllProperties {
VertexId other_end;
EdgeType type;
Gid gid;
std::map<PropertyId, Value> properties;
};
struct EdgeWithSpecificProperties {
VertexId other_end;
EdgeType type;
Gid gid;
std::vector<Value> properties;
};
// NOTE: This struct could be a single Values with columns something like this:
// src_vertex(Vertex), vertex_prop1(Value), vertex_prop2(Value), edges(list<Value>)
// where edges might be a list of:
@@ -425,23 +434,27 @@ struct ExpandOneResultRow {
// The drawback of this is currently the key of the map is always interpreted as a string in Value, not as an
// integer, which should be in case of mapped properties.
Vertex src_vertex;
std::optional<std::map<PropertyId, Value>> src_vertex_properties;
std::map<PropertyId, Value> src_vertex_properties;
// NOTE: If the desired edges are specified in the request,
// edges_with_specific_properties will have a value and it will
// return the properties as a vector of property values. The order
// of the values returned should be the same as the PropertyIds
// were defined in the request.
std::optional<std::vector<std::tuple<VertexId, Gid, std::map<PropertyId, Value>>>> edges_with_all_properties;
std::optional<std::vector<std::tuple<VertexId, Gid, std::vector<Value>>>> edges_with_specific_properties;
std::vector<EdgeWithAllProperties> in_edges_with_all_properties;
std::vector<EdgeWithSpecificProperties> in_edges_with_specific_properties;
std::vector<EdgeWithAllProperties> out_edges_with_all_properties;
std::vector<EdgeWithSpecificProperties> out_edges_with_specific_properties;
};
struct ExpandOneResponse {
bool success;
std::vector<ExpandOneResultRow> result;
};
struct UpdateVertexProp {
PrimaryKey primary_key;
// This should be a map
std::vector<std::pair<PropertyId, Value>> property_updates;
};
@@ -449,6 +462,7 @@ struct UpdateEdgeProp {
EdgeId edge_id;
VertexId src;
VertexId dst;
// This should be a map
std::vector<std::pair<PropertyId, Value>> property_updates;
};
@@ -458,6 +472,7 @@ struct UpdateEdgeProp {
struct NewVertex {
std::vector<Label> label_ids;
PrimaryKey primary_key;
// This should be a map
std::vector<std::pair<PropertyId, Value>> properties;
};
@@ -493,12 +508,22 @@ struct UpdateVerticesResponse {
/*
* Edges
*/
struct CreateEdgesRequest {
Hlc transaction_id;
std::vector<Edge> edges;
// No need for specifying direction since it has to be in one, and src and dest
// vertices clearly communicate the direction
struct NewExpand {
EdgeId id;
EdgeType type;
VertexId src_vertex;
VertexId dest_vertex;
std::vector<std::pair<PropertyId, Value>> properties;
};
struct CreateEdgesResponse {
struct CreateExpandRequest {
Hlc transaction_id;
std::vector<NewExpand> new_expands;
};
struct CreateExpandResponse {
bool success;
};
@@ -520,12 +545,66 @@ struct UpdateEdgesResponse {
bool success;
};
struct CommitRequest {
Hlc transaction_id;
Hlc commit_timestamp;
};
struct CommitResponse {
bool success;
};
using ReadRequests = std::variant<ExpandOneRequest, GetPropertiesRequest, ScanVerticesRequest>;
using ReadResponses = std::variant<ExpandOneResponse, GetPropertiesResponse, ScanVerticesResponse>;
using WriteRequests = std::variant<CreateVerticesRequest, DeleteVerticesRequest, UpdateVerticesRequest,
CreateEdgesRequest, DeleteEdgesRequest, UpdateEdgesRequest>;
CreateExpandRequest, DeleteEdgesRequest, UpdateEdgesRequest, CommitRequest>;
using WriteResponses = std::variant<CreateVerticesResponse, DeleteVerticesResponse, UpdateVerticesResponse,
CreateEdgesResponse, DeleteEdgesResponse, UpdateEdgesResponse>;
CreateExpandResponse, DeleteEdgesResponse, UpdateEdgesResponse, CommitResponse>;
} // namespace memgraph::msgs
namespace std {
template <>
struct hash<memgraph::msgs::Value>;
template <>
struct hash<memgraph::msgs::VertexId> {
size_t operator()(const memgraph::msgs::VertexId &id) const {
using LabelId = memgraph::storage::v3::LabelId;
using Value = memgraph::msgs::Value;
return memgraph::utils::HashCombine<LabelId, std::vector<Value>, std::hash<LabelId>,
memgraph::utils::FnvCollection<std::vector<Value>, Value>>{}(id.first.id,
id.second);
}
};
template <>
struct hash<memgraph::msgs::Value> {
size_t operator()(const memgraph::msgs::Value &value) const {
using Type = memgraph::msgs::Value::Type;
switch (value.type) {
case Type::Null:
return std::hash<size_t>{}(0U);
case Type::Bool:
return std::hash<bool>{}(value.bool_v);
case Type::Int64:
return std::hash<int64_t>{}(value.int_v);
case Type::Double:
return std::hash<double>{}(value.double_v);
case Type::String:
return std::hash<std::string>{}(value.string_v);
case Type::List:
LOG_FATAL("Add hash for lists");
case Type::Map:
LOG_FATAL("Add hash for maps");
case Type::Vertex:
LOG_FATAL("Add hash for vertices");
case Type::Edge:
LOG_FATAL("Add hash for edges");
}
}
};
} // namespace std

View File

@@ -0,0 +1,757 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#pragma once
#include <chrono>
#include <deque>
#include <iostream>
#include <iterator>
#include <map>
#include <numeric>
#include <optional>
#include <random>
#include <set>
#include <stdexcept>
#include <thread>
#include <unordered_map>
#include <vector>
#include "coordinator/coordinator.hpp"
#include "coordinator/coordinator_client.hpp"
#include "coordinator/coordinator_rsm.hpp"
#include "coordinator/shard_map.hpp"
#include "io/address.hpp"
#include "io/errors.hpp"
#include "io/rsm/raft.hpp"
#include "io/rsm/rsm_client.hpp"
#include "io/rsm/shard_rsm.hpp"
#include "io/simulator/simulator.hpp"
#include "io/simulator/simulator_transport.hpp"
#include "query/v2/accessors.hpp"
#include "query/v2/requests.hpp"
#include "storage/v3/id_types.hpp"
#include "storage/v3/value_conversions.hpp"
#include "utils/result.hpp"
namespace memgraph::msgs {
template <typename TStorageClient>
class RsmStorageClientManager {
public:
using CompoundKey = memgraph::io::rsm::ShardRsmKey;
using Shard = memgraph::coordinator::Shard;
using LabelId = memgraph::storage::v3::LabelId;
RsmStorageClientManager() = default;
RsmStorageClientManager(const RsmStorageClientManager &) = delete;
RsmStorageClientManager(RsmStorageClientManager &&) = delete;
RsmStorageClientManager &operator=(const RsmStorageClientManager &) = delete;
RsmStorageClientManager &operator=(RsmStorageClientManager &&) = delete;
~RsmStorageClientManager() = default;
void AddClient(Shard key, TStorageClient client) { cli_cache_.emplace(std::move(key), std::move(client)); }
bool Exists(const Shard &key) { return cli_cache_.contains(key); }
void PurgeCache() { cli_cache_.clear(); }
TStorageClient &GetClient(const Shard &key) {
auto it = cli_cache_.find(key);
MG_ASSERT(it != cli_cache_.end(), "Non-existing shard client");
return it->second;
}
private:
std::map<Shard, TStorageClient> cli_cache_;
};
template <typename TRequest>
struct ExecutionState {
using CompoundKey = memgraph::io::rsm::ShardRsmKey;
using Shard = memgraph::coordinator::Shard;
enum State : int8_t { INITIALIZING, EXECUTING, COMPLETED };
// label is optional because some operators can create/remove etc, vertices. These kind of requests contain the label
// on the request itself.
std::optional<std::string> label;
// CompoundKey is optional because some operators require to iterate over all the available keys
// of a shard. One example is ScanAll, where we only require the field label.
std::optional<CompoundKey> key;
// Transaction id to be filled by the ShardRequestManager implementation
memgraph::coordinator::Hlc transaction_id;
// Initialized by ShardRequestManager implementation. This vector is filled with the shards that
// the ShardRequestManager impl will send requests to. When a request to a shard exhausts it, meaning that
// it pulled all the requested data from the given Shard, it will be removed from the Vector. When the Vector becomes
// empty, it means that all of the requests have completed succefully.
// TODO(gvolfing)
// Maybe make this into a more complex object to be able to keep track of paginated resutls. E.g. instead of a vector
// of Shards make it into a std::vector<std::pair<Shard, PaginatedResultType>> (probably a struct instead of a pair)
// where PaginatedResultType is an enum signaling the progress on the given request. This way we can easily check if
// a partial response on a shard(if there is one) is finished and we can send off the request for the next batch.
std::vector<Shard> shard_cache;
// 1-1 mapping with `shard_cache`.
// A vector that tracks request metadata for each shard (For example, next_id for a ScanAll on Shard A)
std::vector<TRequest> requests;
State state = INITIALIZING;
};
class ShardRequestManagerInterface {
public:
using VertexAccessor = memgraph::query::v2::accessors::VertexAccessor;
ShardRequestManagerInterface() = default;
ShardRequestManagerInterface(const ShardRequestManagerInterface &) = delete;
ShardRequestManagerInterface(ShardRequestManagerInterface &&) = delete;
ShardRequestManagerInterface &operator=(const ShardRequestManagerInterface &) = delete;
ShardRequestManagerInterface &&operator=(ShardRequestManagerInterface &&) = delete;
virtual ~ShardRequestManagerInterface() = default;
virtual void StartTransaction() = 0;
virtual void Commit() = 0;
virtual std::vector<VertexAccessor> Request(ExecutionState<ScanVerticesRequest> &state,
std::vector<VertexId> &&scanned_vertices) = 0;
virtual std::vector<CreateVerticesResponse> Request(ExecutionState<CreateVerticesRequest> &state,
std::vector<NewVertex> new_vertices) = 0;
virtual std::vector<ExpandOneResultRow> Request(ExecutionState<ExpandOneRequest> &state,
ExpandOneRequest request) = 0;
virtual std::vector<CreateExpandResponse> Request(ExecutionState<CreateExpandRequest> &state,
std::vector<NewExpand> new_edges) = 0;
virtual storage::v3::EdgeTypeId NameToEdgeType(const std::string &name) const = 0;
virtual storage::v3::PropertyId NameToProperty(const std::string &name) const = 0;
virtual storage::v3::LabelId NameToLabel(const std::string &name) const = 0;
virtual const std::string &PropertyToName(memgraph::storage::v3::PropertyId prop) const = 0;
virtual const std::string &LabelToName(memgraph::storage::v3::LabelId label) const = 0;
virtual const std::string &EdgeTypeToName(memgraph::storage::v3::EdgeTypeId type) const = 0;
virtual bool IsPrimaryLabel(LabelId label) const = 0;
virtual bool IsPrimaryProperty(LabelId primary_label, PropertyId property) const = 0;
virtual bool IsPrimaryKey(LabelId primary_label, PropertyId property) const = 0;
};
// TODO(kostasrim)rename this class template
template <typename TTransport>
class ShardRequestManager : public ShardRequestManagerInterface {
public:
using StorageClient =
memgraph::coordinator::RsmClient<TTransport, WriteRequests, WriteResponses, ReadRequests, ReadResponses>;
using CoordinatorWriteRequests = memgraph::coordinator::CoordinatorWriteRequests;
using CoordinatorClient = memgraph::coordinator::CoordinatorClient<TTransport>;
using Address = memgraph::io::Address;
using Shard = memgraph::coordinator::Shard;
using ShardMap = memgraph::coordinator::ShardMap;
using CompoundKey = memgraph::coordinator::PrimaryKey;
using VertexAccessor = memgraph::query::v2::accessors::VertexAccessor;
ShardRequestManager(CoordinatorClient coord, memgraph::io::Io<TTransport> &&io)
: coord_cli_(std::move(coord)), io_(std::move(io)) {}
ShardRequestManager(const ShardRequestManager &) = delete;
ShardRequestManager(ShardRequestManager &&) = delete;
ShardRequestManager &operator=(const ShardRequestManager &) = delete;
ShardRequestManager &operator=(ShardRequestManager &&) = delete;
~ShardRequestManager() override {}
void StartTransaction() override {
memgraph::coordinator::HlcRequest req{.last_shard_map_version = shards_map_.GetHlc()};
CoordinatorWriteRequests write_req = req;
auto write_res = coord_cli_.SendWriteRequest(write_req);
if (write_res.HasError()) {
throw std::runtime_error("HLC request failed");
}
auto coordinator_write_response = write_res.GetValue();
auto hlc_response = std::get<memgraph::coordinator::HlcResponse>(coordinator_write_response);
// Transaction ID to be used later...
transaction_id_ = hlc_response.new_hlc;
if (hlc_response.fresher_shard_map) {
shards_map_ = hlc_response.fresher_shard_map.value();
SetUpNameIdMappers();
}
}
void Commit() override {
memgraph::coordinator::HlcRequest req{.last_shard_map_version = shards_map_.GetHlc()};
CoordinatorWriteRequests write_req = req;
auto write_res = coord_cli_.SendWriteRequest(write_req);
if (write_res.HasError()) {
throw std::runtime_error("HLC request for commit failed");
}
auto coordinator_write_response = write_res.GetValue();
auto hlc_response = std::get<memgraph::coordinator::HlcResponse>(coordinator_write_response);
if (hlc_response.fresher_shard_map) {
shards_map_ = hlc_response.fresher_shard_map.value();
SetUpNameIdMappers();
}
auto commit_timestamp = hlc_response.new_hlc;
msgs::CommitRequest commit_req{.transaction_id = transaction_id_, .commit_timestamp = commit_timestamp};
for (const auto &[label, space] : shards_map_.label_spaces) {
for (const auto &[key, shard] : space.shards) {
auto &storage_client = GetStorageClientForShard(shard);
// TODO(kostasrim) Currently requests return the result directly. Adjust this when the API works MgFuture
// instead.
auto commit_response = storage_client.SendWriteRequest(commit_req);
// RETRY on timeouts?
// Sometimes this produces a timeout. Temporary solution is to use a while(true) as was done in shard_map test
if (commit_response.HasError()) {
throw std::runtime_error("Commit request timed out");
}
WriteResponses write_response_variant = commit_response.GetValue();
auto &response = std::get<CommitResponse>(write_response_variant);
if (!response.success) {
throw std::runtime_error("Commit request did not succeed");
}
}
}
}
storage::v3::EdgeTypeId NameToEdgeType(const std::string &name) const override {
return shards_map_.GetEdgeTypeId(name).value();
}
storage::v3::PropertyId NameToProperty(const std::string &name) const override {
return shards_map_.GetPropertyId(name).value();
}
storage::v3::LabelId NameToLabel(const std::string &name) const override {
return shards_map_.GetLabelId(name).value();
}
const std::string &PropertyToName(memgraph::storage::v3::PropertyId id) const override {
return properties_.IdToName(id.AsUint());
}
const std::string &LabelToName(memgraph::storage::v3::LabelId id) const override {
return labels_.IdToName(id.AsUint());
}
const std::string &EdgeTypeToName(memgraph::storage::v3::EdgeTypeId id) const override {
return edge_types_.IdToName(id.AsUint());
}
bool IsPrimaryLabel(LabelId label) const override { return shards_map_.label_spaces.contains(label); }
bool IsPrimaryProperty(LabelId primary_label, PropertyId property) const override {
const auto schema_it = shards_map_.schemas.find(primary_label);
MG_ASSERT(schema_it != shards_map_.schemas.end(), "Invalid primary label id: {}", primary_label.AsUint());
return std::find_if(schema_it->second.begin(), schema_it->second.end(), [property](const auto &schema_prop) {
return schema_prop.property_id == property;
}) != schema_it->second.end();
}
bool IsPrimaryKey(LabelId primary_label, PropertyId property) const override {
const auto schema_it = shards_map_.schemas.find(primary_label);
MG_ASSERT(schema_it != shards_map_.schemas.end(), "Invalid primary label id: {}", primary_label.AsUint());
return schema_it->second.size() == 1 && schema_it->second[0].property_id == property;
}
// TODO(kostasrim) Simplify return result
std::vector<VertexAccessor> Request(ExecutionState<ScanVerticesRequest> &state,
std::vector<VertexId> &&scanned_vertices) override {
MaybeInitializeExecutionState(state, std::move(scanned_vertices));
std::vector<ScanVerticesResponse> responses;
SendAllRequests(state);
auto all_requests_gathered = [](auto &paginated_rsp_tracker) {
return std::ranges::all_of(paginated_rsp_tracker, [](const auto &state) {
return state.second == PaginatedResponseState::PartiallyFinished;
});
};
std::map<Shard, PaginatedResponseState> paginated_response_tracker;
for (const auto &shard : state.shard_cache) {
paginated_response_tracker.insert(std::make_pair(shard, PaginatedResponseState::Pending));
}
do {
AwaitOnPaginatedRequests(state, responses, paginated_response_tracker);
} while (!all_requests_gathered(paginated_response_tracker));
MaybeCompleteState(state);
// TODO(kostasrim) Before returning start prefetching the batch (this shall be done once we get MgFuture as return
// result of storage_client.SendReadRequest()).
return PostProcess(std::move(responses));
}
std::vector<CreateVerticesResponse> Request(ExecutionState<CreateVerticesRequest> &state,
std::vector<NewVertex> new_vertices) override {
MG_ASSERT(!new_vertices.empty());
MaybeInitializeExecutionState(state, new_vertices);
std::vector<CreateVerticesResponse> responses;
auto &shard_cache_ref = state.shard_cache;
// 1. Send the requests.
SendAllRequests(state, shard_cache_ref);
// 2. Block untill all the futures are exhausted
do {
AwaitOnResponses(state, responses);
} while (!state.shard_cache.empty());
MaybeCompleteState(state);
// TODO(kostasrim) Before returning start prefetching the batch (this shall be done once we get MgFuture as return
// result of storage_client.SendReadRequest()).
return responses;
}
std::vector<CreateExpandResponse> Request(ExecutionState<CreateExpandRequest> &state,
std::vector<NewExpand> new_edges) override {
MG_ASSERT(!new_edges.empty());
MaybeInitializeExecutionState(state, new_edges);
std::vector<CreateExpandResponse> responses;
auto &shard_cache_ref = state.shard_cache;
size_t id{0};
for (auto shard_it = shard_cache_ref.begin(); shard_it != shard_cache_ref.end(); ++id) {
auto &storage_client = GetStorageClientForShard(*shard_it);
WriteRequests req = state.requests[id];
auto write_response_result = storage_client.SendWriteRequest(std::move(req));
if (write_response_result.HasError()) {
throw std::runtime_error("CreateVertices request timedout");
}
WriteResponses response_variant = write_response_result.GetValue();
CreateExpandResponse mapped_response = std::get<CreateExpandResponse>(response_variant);
if (!mapped_response.success) {
throw std::runtime_error("CreateExpand request did not succeed");
}
responses.push_back(mapped_response);
shard_it = shard_cache_ref.erase(shard_it);
}
// We are done with this state
MaybeCompleteState(state);
return responses;
}
std::vector<ExpandOneResultRow> Request(ExecutionState<ExpandOneRequest> &state, ExpandOneRequest request) override {
// TODO(kostasrim)Update to limit the batch size here
// Expansions of the destination must be handled by the caller. For example
// match (u:L1 { prop : 1 })-[:Friend]-(v:L1)
// For each vertex U, the ExpandOne will result in <U, Edges>. The destination vertex and its properties
// must be fetched again with an ExpandOne(Edges.dst)
MaybeInitializeExecutionState(state, std::move(request));
std::vector<ExpandOneResponse> responses;
auto &shard_cache_ref = state.shard_cache;
// 1. Send the requests.
SendAllRequests(state, shard_cache_ref);
// 2. Block untill all the futures are exhausted
do {
AwaitOnResponses(state, responses);
} while (!state.shard_cache.empty());
std::vector<ExpandOneResultRow> result_rows;
const auto total_row_count = std::accumulate(
responses.begin(), responses.end(), 0,
[](const int64_t partial_count, const ExpandOneResponse &resp) { return partial_count + resp.result.size(); });
result_rows.reserve(total_row_count);
for (auto &response : responses) {
result_rows.insert(result_rows.end(), std::make_move_iterator(response.result.begin()),
std::make_move_iterator(response.result.end()));
}
MaybeCompleteState(state);
return result_rows;
}
private:
enum class PaginatedResponseState { Pending, PartiallyFinished };
std::vector<VertexAccessor> PostProcess(std::vector<ScanVerticesResponse> &&responses) const {
std::vector<VertexAccessor> accessors;
for (auto &response : responses) {
for (auto &result_row : response.results) {
accessors.emplace_back(VertexAccessor(std::move(result_row.vertex), std::move(result_row.props), this));
}
}
return accessors;
}
template <typename ExecutionState>
void ThrowIfStateCompleted(ExecutionState &state) const {
if (state.state == ExecutionState::COMPLETED) [[unlikely]] {
throw std::runtime_error("State is completed and must be reset");
}
}
template <typename ExecutionState>
void MaybeCompleteState(ExecutionState &state) const {
if (state.requests.empty()) {
state.state = ExecutionState::COMPLETED;
}
}
template <typename ExecutionState>
bool ShallNotInitializeState(ExecutionState &state) const {
return state.state != ExecutionState::INITIALIZING;
}
void MaybeInitializeExecutionState(ExecutionState<CreateVerticesRequest> &state,
std::vector<NewVertex> new_vertices) {
ThrowIfStateCompleted(state);
if (ShallNotInitializeState(state)) {
return;
}
state.transaction_id = transaction_id_;
std::map<Shard, CreateVerticesRequest> per_shard_request_table;
for (auto &new_vertex : new_vertices) {
MG_ASSERT(!new_vertex.label_ids.empty(), "This is error!");
const auto &shard = shards_map_.GetShardForKey(
new_vertex.label_ids[0].id, storage::conversions::ConvertPropertyVector(new_vertex.primary_key));
if (!per_shard_request_table.contains(shard)) {
CreateVerticesRequest create_v_rqst{.transaction_id = transaction_id_};
per_shard_request_table.insert(std::pair(shard, std::move(create_v_rqst)));
state.shard_cache.push_back(shard);
}
per_shard_request_table[shard].new_vertices.push_back(std::move(new_vertex));
}
for (auto &[shard, rqst] : per_shard_request_table) {
state.requests.push_back(std::move(rqst));
}
state.state = ExecutionState<CreateVerticesRequest>::EXECUTING;
}
void MaybeInitializeExecutionState(ExecutionState<CreateExpandRequest> &state, std::vector<NewExpand> new_expands) {
ThrowIfStateCompleted(state);
if (ShallNotInitializeState(state)) {
return;
}
state.transaction_id = transaction_id_;
std::map<Shard, CreateExpandRequest> per_shard_request_table;
auto ensure_shard_exists_in_table = [&per_shard_request_table,
transaction_id = transaction_id_](const Shard &shard) {
if (!per_shard_request_table.contains(shard)) {
CreateExpandRequest create_expand_request{.transaction_id = transaction_id};
per_shard_request_table.insert({shard, std::move(create_expand_request)});
}
};
for (auto &new_expand : new_expands) {
const auto &shard_src_vertex = shards_map_.GetShardForKey(
new_expand.src_vertex.first.id, storage::conversions::ConvertPropertyVector(new_expand.src_vertex.second));
const auto &shard_dest_vertex = shards_map_.GetShardForKey(
new_expand.dest_vertex.first.id, storage::conversions::ConvertPropertyVector(new_expand.dest_vertex.second));
ensure_shard_exists_in_table(shard_src_vertex);
if (shard_src_vertex != shard_dest_vertex) {
ensure_shard_exists_in_table(shard_dest_vertex);
per_shard_request_table[shard_dest_vertex].new_expands.push_back(new_expand);
}
per_shard_request_table[shard_src_vertex].new_expands.push_back(std::move(new_expand));
}
for (auto &[shard, request] : per_shard_request_table) {
state.shard_cache.push_back(shard);
state.requests.push_back(std::move(request));
}
state.state = ExecutionState<CreateExpandRequest>::EXECUTING;
}
void MaybeInitializeExecutionState(ExecutionState<ScanVerticesRequest> &state,
std::vector<VertexId> &&scanned_vertices) {
ThrowIfStateCompleted(state);
if (ShallNotInitializeState(state)) {
return;
}
if (!scanned_vertices.empty()) {
MG_ASSERT(state.label.has_value(), "Fill out the label member of the state!");
std::map<Shard, ScanVerticesRequest> shards_and_requests;
for (auto &vertex_id : scanned_vertices) {
const auto &shard = shards_map_.GetShardForKey(vertex_id.first.id,
storage::conversions::ConvertPropertyVector(vertex_id.second));
shards_and_requests[shard].scanned_vertices.push_back(std::move(vertex_id));
}
for (auto &[shard, request] : shards_and_requests) {
state.shard_cache.push_back(shard);
request.transaction_id = transaction_id_;
state.requests.push_back(std::move(request));
}
} else {
std::vector<coordinator::Shards> multi_shards;
state.transaction_id = transaction_id_;
if (!state.label) {
multi_shards = shards_map_.GetAllShards();
} else {
const auto label_id = shards_map_.GetLabelId(*state.label);
MG_ASSERT(label_id);
MG_ASSERT(IsPrimaryLabel(*label_id));
multi_shards = {shards_map_.GetShardsForLabel(*state.label)};
}
for (auto &shards : multi_shards) {
for (auto &[key, shard] : shards) {
MG_ASSERT(!shard.empty());
state.shard_cache.push_back(std::move(shard));
ScanVerticesRequest rqst;
rqst.transaction_id = transaction_id_;
rqst.start_id.second = storage::conversions::ConvertValueVector(key);
state.requests.push_back(std::move(rqst));
}
}
}
state.state = ExecutionState<ScanVerticesRequest>::EXECUTING;
}
void MaybeInitializeExecutionState(ExecutionState<ExpandOneRequest> &state, ExpandOneRequest request) {
ThrowIfStateCompleted(state);
if (ShallNotInitializeState(state)) {
return;
}
state.transaction_id = transaction_id_;
std::map<Shard, ExpandOneRequest> per_shard_request_table;
auto top_level_rqst_template = request;
top_level_rqst_template.transaction_id = transaction_id_;
top_level_rqst_template.src_vertices.clear();
state.requests.clear();
for (auto &vertex : request.src_vertices) {
const auto &shard =
shards_map_.GetShardForKey(vertex.first.id, storage::conversions::ConvertPropertyVector(vertex.second));
if (!per_shard_request_table.contains(shard)) {
per_shard_request_table.insert(std::pair(shard, top_level_rqst_template));
}
per_shard_request_table[shard].src_vertices.push_back(std::move(vertex));
}
for (auto &[shard, rqst] : per_shard_request_table) {
state.shard_cache.push_back(shard);
state.requests.push_back(std::move(rqst));
}
state.state = ExecutionState<ExpandOneRequest>::EXECUTING;
}
StorageClient &GetStorageClientForShard(Shard shard) {
if (!storage_cli_manager_.Exists(shard)) {
AddStorageClientToManager(shard);
}
return storage_cli_manager_.GetClient(shard);
}
StorageClient &GetStorageClientForShard(const std::string &label, const CompoundKey &key) {
auto shard = shards_map_.GetShardForKey(label, key);
return GetStorageClientForShard(std::move(shard));
}
void AddStorageClientToManager(Shard target_shard) {
MG_ASSERT(!target_shard.empty());
auto leader_addr = target_shard.front();
std::vector<Address> addresses;
addresses.reserve(target_shard.size());
for (auto &address : target_shard) {
addresses.push_back(std::move(address.address));
}
auto cli = StorageClient(io_, std::move(leader_addr.address), std::move(addresses));
storage_cli_manager_.AddClient(target_shard, std::move(cli));
}
void SendAllRequests(ExecutionState<ScanVerticesRequest> &state) {
int64_t shard_idx = 0;
for (const auto &request : state.requests) {
const auto &current_shard = state.shard_cache[shard_idx];
auto &storage_client = GetStorageClientForShard(current_shard);
ReadRequests req = request;
storage_client.SendAsyncReadRequest(request);
++shard_idx;
}
}
void SendAllRequests(ExecutionState<CreateVerticesRequest> &state,
std::vector<memgraph::coordinator::Shard> &shard_cache_ref) {
size_t id = 0;
for (auto shard_it = shard_cache_ref.begin(); shard_it != shard_cache_ref.end(); ++shard_it) {
// This is fine because all new_vertices of each request end up on the same shard
const auto labels = state.requests[id].new_vertices[0].label_ids;
auto req_deep_copy = state.requests[id];
for (auto &new_vertex : req_deep_copy.new_vertices) {
new_vertex.label_ids.erase(new_vertex.label_ids.begin());
}
auto &storage_client = GetStorageClientForShard(*shard_it);
WriteRequests req = req_deep_copy;
storage_client.SendAsyncWriteRequest(req);
++id;
}
}
void SendAllRequests(ExecutionState<ExpandOneRequest> &state,
std::vector<memgraph::coordinator::Shard> &shard_cache_ref) {
size_t id = 0;
for (auto shard_it = shard_cache_ref.begin(); shard_it != shard_cache_ref.end(); ++shard_it) {
auto &storage_client = GetStorageClientForShard(*shard_it);
ReadRequests req = state.requests[id];
storage_client.SendAsyncReadRequest(req);
++id;
}
}
void AwaitOnResponses(ExecutionState<CreateVerticesRequest> &state, std::vector<CreateVerticesResponse> &responses) {
auto &shard_cache_ref = state.shard_cache;
int64_t request_idx = 0;
for (auto shard_it = shard_cache_ref.begin(); shard_it != shard_cache_ref.end();) {
// This is fine because all new_vertices of each request end up on the same shard
const auto labels = state.requests[request_idx].new_vertices[0].label_ids;
auto &storage_client = GetStorageClientForShard(*shard_it);
auto poll_result = storage_client.AwaitAsyncWriteRequest();
if (!poll_result) {
++shard_it;
++request_idx;
continue;
}
if (poll_result->HasError()) {
throw std::runtime_error("CreateVertices request timed out");
}
WriteResponses response_variant = poll_result->GetValue();
auto response = std::get<CreateVerticesResponse>(response_variant);
if (!response.success) {
throw std::runtime_error("CreateVertices request did not succeed");
}
responses.push_back(response);
shard_it = shard_cache_ref.erase(shard_it);
// Needed to maintain the 1-1 mapping between the ShardCache and the requests.
auto it = state.requests.begin() + request_idx;
state.requests.erase(it);
}
}
void AwaitOnResponses(ExecutionState<ExpandOneRequest> &state, std::vector<ExpandOneResponse> &responses) {
auto &shard_cache_ref = state.shard_cache;
int64_t request_idx = 0;
for (auto shard_it = shard_cache_ref.begin(); shard_it != shard_cache_ref.end();) {
auto &storage_client = GetStorageClientForShard(*shard_it);
auto poll_result = storage_client.PollAsyncReadRequest();
if (!poll_result) {
++shard_it;
++request_idx;
continue;
}
if (poll_result->HasError()) {
throw std::runtime_error("ExpandOne request timed out");
}
ReadResponses response_variant = poll_result->GetValue();
auto response = std::get<ExpandOneResponse>(response_variant);
// -NOTE-
// Currently a boolean flag for signaling the overall success of the
// ExpandOne request does not exist. But it should, so here we assume
// that it is already in place.
if (!response.success) {
throw std::runtime_error("ExpandOne request did not succeed");
}
responses.push_back(std::move(response));
shard_it = shard_cache_ref.erase(shard_it);
// Needed to maintain the 1-1 mapping between the ShardCache and the requests.
auto it = state.requests.begin() + request_idx;
state.requests.erase(it);
}
}
void AwaitOnPaginatedRequests(ExecutionState<ScanVerticesRequest> &state,
std::vector<ScanVerticesResponse> &responses,
std::map<Shard, PaginatedResponseState> &paginated_response_tracker) {
auto &shard_cache_ref = state.shard_cache;
// Find the first request that is not holding a paginated response.
int64_t request_idx = 0;
for (auto shard_it = shard_cache_ref.begin(); shard_it != shard_cache_ref.end();) {
if (paginated_response_tracker.at(*shard_it) != PaginatedResponseState::Pending) {
++shard_it;
++request_idx;
continue;
}
auto &storage_client = GetStorageClientForShard(*shard_it);
auto await_result = storage_client.AwaitAsyncReadRequest();
if (!await_result) {
// Redirection has occured.
++shard_it;
++request_idx;
continue;
}
if (await_result->HasError()) {
throw std::runtime_error("ScanAll request timed out");
}
ReadResponses read_response_variant = await_result->GetValue();
auto response = std::get<ScanVerticesResponse>(read_response_variant);
if (!response.success) {
throw std::runtime_error("ScanAll request did not succeed");
}
if (!response.next_start_id) {
paginated_response_tracker.erase((*shard_it));
shard_cache_ref.erase(shard_it);
// Needed to maintain the 1-1 mapping between the ShardCache and the requests.
auto it = state.requests.begin() + request_idx;
state.requests.erase(it);
} else {
state.requests[request_idx].start_id.second = response.next_start_id->second;
paginated_response_tracker[*shard_it] = PaginatedResponseState::PartiallyFinished;
}
responses.push_back(std::move(response));
}
}
void SetUpNameIdMappers() {
std::unordered_map<uint64_t, std::string> id_to_name;
for (const auto &[name, id] : shards_map_.labels) {
id_to_name.emplace(id.AsUint(), name);
}
labels_.StoreMapping(std::move(id_to_name));
id_to_name.clear();
for (const auto &[name, id] : shards_map_.properties) {
id_to_name.emplace(id.AsUint(), name);
}
properties_.StoreMapping(std::move(id_to_name));
id_to_name.clear();
for (const auto &[name, id] : shards_map_.edge_types) {
id_to_name.emplace(id.AsUint(), name);
}
edge_types_.StoreMapping(std::move(id_to_name));
}
ShardMap shards_map_;
storage::v3::NameIdMapper properties_;
storage::v3::NameIdMapper edge_types_;
storage::v3::NameIdMapper labels_;
CoordinatorClient coord_cli_;
RsmStorageClientManager<StorageClient> storage_cli_manager_;
memgraph::io::Io<TTransport> io_;
memgraph::coordinator::Hlc transaction_id_;
// TODO(kostasrim) Add batch prefetching
};
} // namespace memgraph::msgs

View File

@@ -1,45 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#include "query/v2/stream/common.hpp"
#include <json/json.hpp>
namespace memgraph::query::v2::stream {
namespace {
const std::string kBatchIntervalKey{"batch_interval"};
const std::string kBatchSizeKey{"batch_size"};
const std::string kTransformationName{"transformation_name"};
} // namespace
void to_json(nlohmann::json &data, CommonStreamInfo &&common_info) {
data[kBatchIntervalKey] = common_info.batch_interval.count();
data[kBatchSizeKey] = common_info.batch_size;
data[kTransformationName] = common_info.transformation_name;
}
void from_json(const nlohmann::json &data, CommonStreamInfo &common_info) {
if (const auto batch_interval = data.at(kBatchIntervalKey); !batch_interval.is_null()) {
using BatchInterval = decltype(common_info.batch_interval);
common_info.batch_interval = BatchInterval{batch_interval.get<typename BatchInterval::rep>()};
} else {
common_info.batch_interval = kDefaultBatchInterval;
}
if (const auto batch_size = data.at(kBatchSizeKey); !batch_size.is_null()) {
common_info.batch_size = batch_size.get<decltype(common_info.batch_size)>();
} else {
common_info.batch_size = kDefaultBatchSize;
}
data.at(kTransformationName).get_to(common_info.transformation_name);
}
} // namespace memgraph::query::v2::stream

View File

@@ -1,87 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#pragma once
#include <chrono>
#include <cstdint>
#include <functional>
#include <optional>
#include <string>
#include <json/json.hpp>
#include "query/v2/procedure/mg_procedure_impl.hpp"
namespace memgraph::query::v2::stream {
inline constexpr std::chrono::milliseconds kDefaultBatchInterval{100};
inline constexpr int64_t kDefaultBatchSize{1000};
template <typename TMessage>
using ConsumerFunction = std::function<void(const std::vector<TMessage> &)>;
struct CommonStreamInfo {
std::chrono::milliseconds batch_interval;
int64_t batch_size;
std::string transformation_name;
};
template <typename T>
concept ConvertableToJson = requires(T value, nlohmann::json data) {
{ to_json(data, std::move(value)) } -> std::same_as<void>;
{ from_json(data, value) } -> std::same_as<void>;
};
template <typename T>
concept ConvertableToMgpMessage = requires(T value) {
mgp_message{value};
};
template <typename TStream>
concept Stream = requires(TStream stream) {
typename TStream::StreamInfo;
typename TStream::Message;
TStream{std::string{""}, typename TStream::StreamInfo{}, ConsumerFunction<typename TStream::Message>{}};
{ stream.Start() } -> std::same_as<void>;
{ stream.StartWithLimit(uint64_t{}, std::optional<std::chrono::milliseconds>{}) } -> std::same_as<void>;
{ stream.Stop() } -> std::same_as<void>;
{ stream.IsRunning() } -> std::same_as<bool>;
{
stream.Check(std::optional<std::chrono::milliseconds>{}, std::optional<uint64_t>{},
ConsumerFunction<typename TStream::Message>{})
} -> std::same_as<void>;
requires std::same_as<std::decay_t<decltype(std::declval<typename TStream::StreamInfo>().common_info)>,
CommonStreamInfo>;
requires ConvertableToMgpMessage<typename TStream::Message>;
requires ConvertableToJson<typename TStream::StreamInfo>;
};
enum class StreamSourceType : uint8_t { KAFKA, PULSAR };
constexpr std::string_view StreamSourceTypeToString(StreamSourceType type) {
switch (type) {
case StreamSourceType::KAFKA:
return "kafka";
case StreamSourceType::PULSAR:
return "pulsar";
}
}
template <Stream T>
StreamSourceType StreamType(const T & /*stream*/);
const std::string kCommonInfoKey = "common_info";
void to_json(nlohmann::json &data, CommonStreamInfo &&info);
void from_json(const nlohmann::json &data, CommonStreamInfo &common_info);
} // namespace memgraph::query::v2::stream

View File

@@ -1,137 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#include "query/v2/stream/sources.hpp"
#include <json/json.hpp>
#include "integrations/constants.hpp"
namespace memgraph::query::v2::stream {
KafkaStream::KafkaStream(std::string stream_name, StreamInfo stream_info,
ConsumerFunction<integrations::kafka::Message> consumer_function) {
integrations::kafka::ConsumerInfo consumer_info{
.consumer_name = std::move(stream_name),
.topics = std::move(stream_info.topics),
.consumer_group = std::move(stream_info.consumer_group),
.bootstrap_servers = std::move(stream_info.bootstrap_servers),
.batch_interval = stream_info.common_info.batch_interval,
.batch_size = stream_info.common_info.batch_size,
.public_configs = std::move(stream_info.configs),
.private_configs = std::move(stream_info.credentials),
};
consumer_.emplace(std::move(consumer_info), std::move(consumer_function));
};
KafkaStream::StreamInfo KafkaStream::Info(std::string transformation_name) const {
const auto &info = consumer_->Info();
return {{.batch_interval = info.batch_interval,
.batch_size = info.batch_size,
.transformation_name = std::move(transformation_name)},
.topics = info.topics,
.consumer_group = info.consumer_group,
.bootstrap_servers = info.bootstrap_servers,
.configs = info.public_configs,
.credentials = info.private_configs};
}
void KafkaStream::Start() { consumer_->Start(); }
void KafkaStream::StartWithLimit(uint64_t batch_limit, std::optional<std::chrono::milliseconds> timeout) const {
consumer_->StartWithLimit(batch_limit, timeout);
}
void KafkaStream::Stop() { consumer_->Stop(); }
bool KafkaStream::IsRunning() const { return consumer_->IsRunning(); }
void KafkaStream::Check(std::optional<std::chrono::milliseconds> timeout, std::optional<uint64_t> batch_limit,
const ConsumerFunction<integrations::kafka::Message> &consumer_function) const {
consumer_->Check(timeout, batch_limit, consumer_function);
}
utils::BasicResult<std::string> KafkaStream::SetStreamOffset(const int64_t offset) {
return consumer_->SetConsumerOffsets(offset);
}
namespace {
const std::string kTopicsKey{"topics"};
const std::string kConsumerGroupKey{"consumer_group"};
const std::string kBoostrapServers{"bootstrap_servers"};
const std::string kConfigs{"configs"};
const std::string kCredentials{"credentials"};
const std::unordered_map<std::string, std::string> kDefaultConfigsMap;
} // namespace
void to_json(nlohmann::json &data, KafkaStream::StreamInfo &&info) {
data[kCommonInfoKey] = std::move(info.common_info);
data[kTopicsKey] = std::move(info.topics);
data[kConsumerGroupKey] = info.consumer_group;
data[kBoostrapServers] = std::move(info.bootstrap_servers);
data[kConfigs] = std::move(info.configs);
data[kCredentials] = std::move(info.credentials);
}
void from_json(const nlohmann::json &data, KafkaStream::StreamInfo &info) {
data.at(kCommonInfoKey).get_to(info.common_info);
data.at(kTopicsKey).get_to(info.topics);
data.at(kConsumerGroupKey).get_to(info.consumer_group);
data.at(kBoostrapServers).get_to(info.bootstrap_servers);
// These values might not be present in the persisted JSON object
info.configs = data.value(kConfigs, kDefaultConfigsMap);
info.credentials = data.value(kCredentials, kDefaultConfigsMap);
}
PulsarStream::PulsarStream(std::string stream_name, StreamInfo stream_info,
ConsumerFunction<integrations::pulsar::Message> consumer_function) {
integrations::pulsar::ConsumerInfo consumer_info{.batch_size = stream_info.common_info.batch_size,
.batch_interval = stream_info.common_info.batch_interval,
.topics = std::move(stream_info.topics),
.consumer_name = std::move(stream_name),
.service_url = std::move(stream_info.service_url)};
consumer_.emplace(std::move(consumer_info), std::move(consumer_function));
};
PulsarStream::StreamInfo PulsarStream::Info(std::string transformation_name) const {
const auto &info = consumer_->Info();
return {{.batch_interval = info.batch_interval,
.batch_size = info.batch_size,
.transformation_name = std::move(transformation_name)},
.topics = info.topics,
.service_url = info.service_url};
}
void PulsarStream::Start() { consumer_->Start(); }
void PulsarStream::StartWithLimit(uint64_t batch_limit, std::optional<std::chrono::milliseconds> timeout) const {
consumer_->StartWithLimit(batch_limit, timeout);
}
void PulsarStream::Stop() { consumer_->Stop(); }
bool PulsarStream::IsRunning() const { return consumer_->IsRunning(); }
void PulsarStream::Check(std::optional<std::chrono::milliseconds> timeout, std::optional<uint64_t> batch_limit,
const ConsumerFunction<Message> &consumer_function) const {
consumer_->Check(timeout, batch_limit, consumer_function);
}
namespace {
const std::string kServiceUrl{"service_url"};
} // namespace
void to_json(nlohmann::json &data, PulsarStream::StreamInfo &&info) {
data[kCommonInfoKey] = std::move(info.common_info);
data[kTopicsKey] = std::move(info.topics);
data[kServiceUrl] = std::move(info.service_url);
}
void from_json(const nlohmann::json &data, PulsarStream::StreamInfo &info) {
data.at(kCommonInfoKey).get_to(info.common_info);
data.at(kTopicsKey).get_to(info.topics);
data.at(kServiceUrl).get_to(info.service_url);
}
} // namespace memgraph::query::v2::stream

View File

@@ -1,95 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#pragma once
#include "query/v2/stream/common.hpp"
#include "integrations/kafka/consumer.hpp"
#include "integrations/pulsar/consumer.hpp"
namespace memgraph::query::v2::stream {
struct KafkaStream {
struct StreamInfo {
CommonStreamInfo common_info;
std::vector<std::string> topics;
std::string consumer_group;
std::string bootstrap_servers;
std::unordered_map<std::string, std::string> configs;
std::unordered_map<std::string, std::string> credentials;
};
using Message = integrations::kafka::Message;
KafkaStream(std::string stream_name, StreamInfo stream_info,
ConsumerFunction<integrations::kafka::Message> consumer_function);
StreamInfo Info(std::string transformation_name) const;
void Start();
void StartWithLimit(uint64_t batch_limit, std::optional<std::chrono::milliseconds> timeout) const;
void Stop();
bool IsRunning() const;
void Check(std::optional<std::chrono::milliseconds> timeout, std::optional<uint64_t> batch_limit,
const ConsumerFunction<Message> &consumer_function) const;
utils::BasicResult<std::string> SetStreamOffset(int64_t offset);
private:
using Consumer = integrations::kafka::Consumer;
std::optional<Consumer> consumer_;
};
void to_json(nlohmann::json &data, KafkaStream::StreamInfo &&info);
void from_json(const nlohmann::json &data, KafkaStream::StreamInfo &info);
template <>
inline StreamSourceType StreamType(const KafkaStream & /*stream*/) {
return StreamSourceType::KAFKA;
}
struct PulsarStream {
struct StreamInfo {
CommonStreamInfo common_info;
std::vector<std::string> topics;
std::string service_url;
};
using Message = integrations::pulsar::Message;
PulsarStream(std::string stream_name, StreamInfo stream_info, ConsumerFunction<Message> consumer_function);
StreamInfo Info(std::string transformation_name) const;
void Start();
void StartWithLimit(uint64_t batch_limit, std::optional<std::chrono::milliseconds> timeout) const;
void Stop();
bool IsRunning() const;
void Check(std::optional<std::chrono::milliseconds> timeout, std::optional<uint64_t> batch_limit,
const ConsumerFunction<Message> &consumer_function) const;
private:
using Consumer = integrations::pulsar::Consumer;
std::optional<Consumer> consumer_;
};
void to_json(nlohmann::json &data, PulsarStream::StreamInfo &&info);
void from_json(const nlohmann::json &data, PulsarStream::StreamInfo &info);
template <>
inline StreamSourceType StreamType(const PulsarStream & /*stream*/) {
return StreamSourceType::PULSAR;
}
} // namespace memgraph::query::v2::stream

View File

@@ -1,773 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#include "query/v2/stream/streams.hpp"
#include <shared_mutex>
#include <string_view>
#include <utility>
#include <spdlog/spdlog.h>
#include <json/json.hpp>
#include "integrations/constants.hpp"
#include "mg_procedure.h"
#include "query/v2/bindings/typed_value.hpp"
#include "query/v2/db_accessor.hpp"
#include "query/v2/discard_value_stream.hpp"
#include "query/v2/exceptions.hpp"
#include "query/v2/interpreter.hpp"
#include "query/v2/procedure/mg_procedure_helpers.hpp"
#include "query/v2/procedure/mg_procedure_impl.hpp"
#include "query/v2/procedure/module.hpp"
#include "query/v2/stream/sources.hpp"
#include "storage/v3/conversions.hpp"
#include "utils/event_counter.hpp"
#include "utils/logging.hpp"
#include "utils/memory.hpp"
#include "utils/on_scope_exit.hpp"
#include "utils/pmr/string.hpp"
#include "utils/variant_helpers.hpp"
namespace EventCounter {
extern const Event MessagesConsumed;
} // namespace EventCounter
namespace memgraph::query::v2::stream {
namespace {
inline constexpr auto kExpectedTransformationResultSize = 2;
inline constexpr auto kCheckStreamResultSize = 2;
const utils::pmr::string query_param_name{"query", utils::NewDeleteResource()};
const utils::pmr::string params_param_name{"parameters", utils::NewDeleteResource()};
const std::map<std::string, storage::v3::PropertyValue> empty_parameters{};
auto GetStream(auto &map, const std::string &stream_name) {
if (auto it = map.find(stream_name); it != map.end()) {
return it;
}
throw StreamsException("Couldn't find stream '{}'", stream_name);
}
std::pair<TypedValue /*query*/, TypedValue /*parameters*/> ExtractTransformationResult(
const utils::pmr::map<utils::pmr::string, TypedValue> &values, const std::string_view transformation_name,
const std::string_view stream_name) {
if (values.size() != kExpectedTransformationResultSize) {
throw StreamsException(
"Transformation '{}' in stream '{}' did not yield all fields (query, parameters) as required.",
transformation_name, stream_name);
}
auto get_value = [&](const utils::pmr::string &field_name) mutable -> const TypedValue & {
auto it = values.find(field_name);
if (it == values.end()) {
throw StreamsException{"Transformation '{}' in stream '{}' did not yield a record with '{}' field.",
transformation_name, stream_name, field_name};
};
return it->second;
};
const auto &query_value = get_value(query_param_name);
MG_ASSERT(query_value.IsString());
const auto &params_value = get_value(params_param_name);
MG_ASSERT(params_value.IsNull() || params_value.IsMap());
return {query_value, params_value};
}
template <typename TMessage>
void CallCustomTransformation(const std::string &transformation_name, const std::vector<TMessage> &messages,
mgp_result &result, storage::v3::Shard::Accessor &storage_accessor,
utils::MemoryResource &memory_resource, const std::string &stream_name) {
DbAccessor db_accessor{&storage_accessor};
{
auto maybe_transformation =
procedure::FindTransformation(procedure::gModuleRegistry, transformation_name, utils::NewDeleteResource());
if (!maybe_transformation) {
throw StreamsException("Couldn't find transformation {} for stream '{}'", transformation_name, stream_name);
};
const auto &trans = *maybe_transformation->second;
mgp_messages mgp_messages{mgp_messages::storage_type{&memory_resource}};
std::transform(messages.begin(), messages.end(), std::back_inserter(mgp_messages.messages),
[](const TMessage &message) { return mgp_message{message}; });
mgp_graph graph{&db_accessor, storage::v3::View::OLD, nullptr};
mgp_memory memory{&memory_resource};
result.rows.clear();
result.error_msg.reset();
result.signature = &trans.results;
MG_ASSERT(result.signature->size() == kExpectedTransformationResultSize);
MG_ASSERT(result.signature->contains(query_param_name));
MG_ASSERT(result.signature->contains(params_param_name));
spdlog::trace("Calling transformation in stream '{}'", stream_name);
trans.cb(&mgp_messages, &graph, &result, &memory);
}
if (result.error_msg.has_value()) {
throw StreamsException(result.error_msg->c_str());
}
}
template <Stream TStream>
StreamStatus<TStream> CreateStatus(std::string stream_name, std::string transformation_name,
std::optional<std::string> owner, const TStream &stream) {
return {.name = std::move(stream_name),
.type = StreamType(stream),
.is_running = stream.IsRunning(),
.info = stream.Info(std::move(transformation_name)),
.owner = std::move(owner)};
}
// nlohmann::json doesn't support string_view access yet
const std::string kStreamName{"name"};
const std::string kIsRunningKey{"is_running"};
const std::string kOwner{"owner"};
const std::string kType{"type"};
} // namespace
template <Stream TStream>
void to_json(nlohmann::json &data, StreamStatus<TStream> &&status) {
data[kStreamName] = std::move(status.name);
data[kType] = status.type;
data[kIsRunningKey] = status.is_running;
if (status.owner.has_value()) {
data[kOwner] = std::move(*status.owner);
} else {
data[kOwner] = nullptr;
}
to_json(data, std::move(status.info));
}
template <Stream TStream>
void from_json(const nlohmann::json &data, StreamStatus<TStream> &status) {
data.at(kStreamName).get_to(status.name);
data.at(kIsRunningKey).get_to(status.is_running);
if (const auto &owner = data.at(kOwner); !owner.is_null()) {
status.owner = owner.get<typename decltype(status.owner)::value_type>();
} else {
status.owner = {};
}
from_json(data, status.info);
}
Streams::Streams(InterpreterContext *interpreter_context, std::filesystem::path directory)
: interpreter_context_(interpreter_context), storage_(std::move(directory)) {
RegisterProcedures();
}
void Streams::RegisterProcedures() {
RegisterKafkaProcedures();
RegisterPulsarProcedures();
}
void Streams::RegisterKafkaProcedures() {
{
static constexpr std::string_view proc_name = "kafka_set_stream_offset";
auto set_stream_offset = [this](mgp_list *args, mgp_graph * /*graph*/, mgp_result *result,
mgp_memory * /*memory*/) {
auto *arg_stream_name = procedure::Call<mgp_value *>(mgp_list_at, args, 0);
const auto *stream_name = procedure::Call<const char *>(mgp_value_get_string, arg_stream_name);
auto *arg_offset = procedure::Call<mgp_value *>(mgp_list_at, args, 1);
const auto offset = procedure::Call<int64_t>(mgp_value_get_int, arg_offset);
auto lock_ptr = streams_.Lock();
auto it = GetStream(*lock_ptr, std::string(stream_name));
std::visit(utils::Overloaded{[&](StreamData<KafkaStream> &kafka_stream) {
auto stream_source_ptr = kafka_stream.stream_source->Lock();
const auto error = stream_source_ptr->SetStreamOffset(offset);
if (error.HasError()) {
MG_ASSERT(mgp_result_set_error_msg(result, error.GetError().c_str()) ==
mgp_error::MGP_ERROR_NO_ERROR,
"Unable to set procedure error message of procedure: {}", proc_name);
}
},
[](auto && /*other*/) {
throw QueryRuntimeException("'{}' can be only used for Kafka stream sources",
proc_name);
}},
it->second);
};
mgp_proc proc(proc_name, set_stream_offset, utils::NewDeleteResource());
MG_ASSERT(mgp_proc_add_arg(&proc, "stream_name", procedure::Call<mgp_type *>(mgp_type_string)) ==
mgp_error::MGP_ERROR_NO_ERROR);
MG_ASSERT(mgp_proc_add_arg(&proc, "offset", procedure::Call<mgp_type *>(mgp_type_int)) ==
mgp_error::MGP_ERROR_NO_ERROR);
procedure::gModuleRegistry.RegisterMgProcedure(proc_name, std::move(proc));
}
{
static constexpr std::string_view proc_name = "kafka_stream_info";
static constexpr std::string_view consumer_group_result_name = "consumer_group";
static constexpr std::string_view topics_result_name = "topics";
static constexpr std::string_view bootstrap_servers_result_name = "bootstrap_servers";
static constexpr std::string_view configs_result_name = "configs";
static constexpr std::string_view credentials_result_name = "credentials";
auto get_stream_info = [this](mgp_list *args, mgp_graph * /*graph*/, mgp_result *result, mgp_memory *memory) {
auto *arg_stream_name = procedure::Call<mgp_value *>(mgp_list_at, args, 0);
const auto *stream_name = procedure::Call<const char *>(mgp_value_get_string, arg_stream_name);
auto lock_ptr = streams_.Lock();
auto it = GetStream(*lock_ptr, std::string(stream_name));
std::visit(
utils::Overloaded{
[&](StreamData<KafkaStream> &kafka_stream) {
auto stream_source_ptr = kafka_stream.stream_source->Lock();
const auto info = stream_source_ptr->Info(kafka_stream.transformation_name);
mgp_result_record *record{nullptr};
if (!procedure::TryOrSetError([&] { return mgp_result_new_record(result, &record); }, result)) {
return;
}
const auto consumer_group_value =
procedure::GetStringValueOrSetError(info.consumer_group.c_str(), memory, result);
if (!consumer_group_value) {
return;
}
procedure::MgpUniquePtr<mgp_list> topic_names{nullptr, mgp_list_destroy};
if (!procedure::TryOrSetError(
[&] {
return procedure::CreateMgpObject(topic_names, mgp_list_make_empty, info.topics.size(),
memory);
},
result)) {
return;
}
for (const auto &topic : info.topics) {
auto topic_value = procedure::GetStringValueOrSetError(topic.c_str(), memory, result);
if (!topic_value) {
return;
}
topic_names->elems.push_back(std::move(*topic_value));
}
procedure::MgpUniquePtr<mgp_value> topics_value{nullptr, mgp_value_destroy};
if (!procedure::TryOrSetError(
[&] {
return procedure::CreateMgpObject(topics_value, mgp_value_make_list, topic_names.get());
},
result)) {
return;
}
static_cast<void>(topic_names.release());
const auto bootstrap_servers_value =
procedure::GetStringValueOrSetError(info.bootstrap_servers.c_str(), memory, result);
if (!bootstrap_servers_value) {
return;
}
const auto convert_config_map =
[result, memory](const std::unordered_map<std::string, std::string> &configs_to_convert)
-> procedure::MgpUniquePtr<mgp_value> {
procedure::MgpUniquePtr<mgp_value> configs_value{nullptr, mgp_value_destroy};
procedure::MgpUniquePtr<mgp_map> configs{nullptr, mgp_map_destroy};
if (!procedure::TryOrSetError(
[&] { return procedure::CreateMgpObject(configs, mgp_map_make_empty, memory); }, result)) {
return configs_value;
}
for (const auto &[key, value] : configs_to_convert) {
auto value_value = procedure::GetStringValueOrSetError(value.c_str(), memory, result);
if (!value_value) {
return configs_value;
}
configs->items.emplace(key, std::move(*value_value));
}
if (!procedure::TryOrSetError(
[&] { return procedure::CreateMgpObject(configs_value, mgp_value_make_map, configs.get()); },
result)) {
return configs_value;
}
static_cast<void>(configs.release());
return configs_value;
};
const auto configs_value = convert_config_map(info.configs);
if (configs_value == nullptr) {
return;
}
using CredentialsType = decltype(KafkaStream::StreamInfo::credentials);
CredentialsType reducted_credentials;
std::transform(info.credentials.begin(), info.credentials.end(),
std::inserter(reducted_credentials, reducted_credentials.end()),
[](const auto &pair) -> CredentialsType::value_type {
return {pair.first, integrations::kReducted};
});
const auto credentials_value = convert_config_map(reducted_credentials);
if (credentials_value == nullptr) {
return;
}
if (!procedure::InsertResultOrSetError(result, record, consumer_group_result_name.data(),
consumer_group_value.get())) {
return;
}
if (!procedure::InsertResultOrSetError(result, record, topics_result_name.data(), topics_value.get())) {
return;
}
if (!procedure::InsertResultOrSetError(result, record, bootstrap_servers_result_name.data(),
bootstrap_servers_value.get())) {
return;
}
if (!procedure::InsertResultOrSetError(result, record, configs_result_name.data(),
configs_value.get())) {
return;
}
if (!procedure::InsertResultOrSetError(result, record, credentials_result_name.data(),
credentials_value.get())) {
return;
}
},
[](auto && /*other*/) {
throw QueryRuntimeException("'{}' can be only used for Kafka stream sources", proc_name);
}},
it->second);
};
mgp_proc proc(proc_name, get_stream_info, utils::NewDeleteResource());
MG_ASSERT(mgp_proc_add_arg(&proc, "stream_name", procedure::Call<mgp_type *>(mgp_type_string)) ==
mgp_error::MGP_ERROR_NO_ERROR);
MG_ASSERT(mgp_proc_add_result(&proc, consumer_group_result_name.data(),
procedure::Call<mgp_type *>(mgp_type_string)) == mgp_error::MGP_ERROR_NO_ERROR);
MG_ASSERT(
mgp_proc_add_result(&proc, topics_result_name.data(),
procedure::Call<mgp_type *>(mgp_type_list, procedure::Call<mgp_type *>(mgp_type_string))) ==
mgp_error::MGP_ERROR_NO_ERROR);
MG_ASSERT(mgp_proc_add_result(&proc, bootstrap_servers_result_name.data(),
procedure::Call<mgp_type *>(mgp_type_string)) == mgp_error::MGP_ERROR_NO_ERROR);
MG_ASSERT(mgp_proc_add_result(&proc, configs_result_name.data(), procedure::Call<mgp_type *>(mgp_type_map)) ==
mgp_error::MGP_ERROR_NO_ERROR);
MG_ASSERT(mgp_proc_add_result(&proc, credentials_result_name.data(), procedure::Call<mgp_type *>(mgp_type_map)) ==
mgp_error::MGP_ERROR_NO_ERROR);
procedure::gModuleRegistry.RegisterMgProcedure(proc_name, std::move(proc));
}
}
void Streams::RegisterPulsarProcedures() {
{
static constexpr std::string_view proc_name = "pulsar_stream_info";
static constexpr std::string_view service_url_result_name = "service_url";
static constexpr std::string_view topics_result_name = "topics";
auto get_stream_info = [this](mgp_list *args, mgp_graph * /*graph*/, mgp_result *result, mgp_memory *memory) {
auto *arg_stream_name = procedure::Call<mgp_value *>(mgp_list_at, args, 0);
const auto *stream_name = procedure::Call<const char *>(mgp_value_get_string, arg_stream_name);
auto lock_ptr = streams_.Lock();
auto it = GetStream(*lock_ptr, std::string(stream_name));
std::visit(
utils::Overloaded{
[&](StreamData<PulsarStream> &pulsar_stream) {
auto stream_source_ptr = pulsar_stream.stream_source->Lock();
const auto info = stream_source_ptr->Info(pulsar_stream.transformation_name);
mgp_result_record *record{nullptr};
if (!procedure::TryOrSetError([&] { return mgp_result_new_record(result, &record); }, result)) {
return;
}
auto service_url_value = procedure::GetStringValueOrSetError(info.service_url.c_str(), memory, result);
if (!service_url_value) {
return;
}
procedure::MgpUniquePtr<mgp_list> topic_names{nullptr, mgp_list_destroy};
if (!procedure::TryOrSetError(
[&] {
return procedure::CreateMgpObject(topic_names, mgp_list_make_empty, info.topics.size(),
memory);
},
result)) {
return;
}
for (const auto &topic : info.topics) {
auto topic_value = procedure::GetStringValueOrSetError(topic.c_str(), memory, result);
if (!topic_value) {
return;
}
topic_names->elems.push_back(std::move(*topic_value));
}
procedure::MgpUniquePtr<mgp_value> topics_value{nullptr, mgp_value_destroy};
if (!procedure::TryOrSetError(
[&] {
return procedure::CreateMgpObject(topics_value, mgp_value_make_list, topic_names.release());
},
result)) {
return;
}
if (!procedure::InsertResultOrSetError(result, record, topics_result_name.data(), topics_value.get())) {
return;
}
if (!procedure::InsertResultOrSetError(result, record, service_url_result_name.data(),
service_url_value.get())) {
return;
}
},
[](auto && /*other*/) {
throw QueryRuntimeException("'{}' can be only used for Pulsar stream sources", proc_name);
}},
it->second);
};
mgp_proc proc(proc_name, get_stream_info, utils::NewDeleteResource());
MG_ASSERT(mgp_proc_add_arg(&proc, "stream_name", procedure::Call<mgp_type *>(mgp_type_string)) ==
mgp_error::MGP_ERROR_NO_ERROR);
MG_ASSERT(mgp_proc_add_result(&proc, service_url_result_name.data(),
procedure::Call<mgp_type *>(mgp_type_string)) == mgp_error::MGP_ERROR_NO_ERROR);
MG_ASSERT(
mgp_proc_add_result(&proc, topics_result_name.data(),
procedure::Call<mgp_type *>(mgp_type_list, procedure::Call<mgp_type *>(mgp_type_string))) ==
mgp_error::MGP_ERROR_NO_ERROR);
procedure::gModuleRegistry.RegisterMgProcedure(proc_name, std::move(proc));
}
}
template <Stream TStream>
void Streams::Create(const std::string &stream_name, typename TStream::StreamInfo info,
std::optional<std::string> owner) {
auto locked_streams = streams_.Lock();
auto it = CreateConsumer<TStream>(*locked_streams, stream_name, std::move(info), std::move(owner));
try {
std::visit(
[&](const auto &stream_data) {
const auto stream_source_ptr = stream_data.stream_source->ReadLock();
Persist(CreateStatus(stream_name, stream_data.transformation_name, stream_data.owner, *stream_source_ptr));
},
it->second);
} catch (...) {
locked_streams->erase(it);
throw;
}
}
template void Streams::Create<KafkaStream>(const std::string &stream_name, KafkaStream::StreamInfo info,
std::optional<std::string> owner);
template void Streams::Create<PulsarStream>(const std::string &stream_name, PulsarStream::StreamInfo info,
std::optional<std::string> owner);
template <Stream TStream>
Streams::StreamsMap::iterator Streams::CreateConsumer(StreamsMap &map, const std::string &stream_name,
typename TStream::StreamInfo stream_info,
std::optional<std::string> owner) {
if (map.contains(stream_name)) {
throw StreamsException{"Stream already exists with name '{}'", stream_name};
}
auto *memory_resource = utils::NewDeleteResource();
auto consumer_function = [interpreter_context = interpreter_context_, memory_resource, stream_name,
transformation_name = stream_info.common_info.transformation_name, owner = owner,
interpreter = std::make_shared<Interpreter>(interpreter_context_),
result = mgp_result{nullptr, memory_resource},
total_retries = interpreter_context_->config.stream_transaction_conflict_retries,
retry_interval = interpreter_context_->config.stream_transaction_retry_interval](
const std::vector<typename TStream::Message> &messages) mutable {
auto accessor = interpreter_context->db->Access();
EventCounter::IncrementCounter(EventCounter::MessagesConsumed, messages.size());
CallCustomTransformation(transformation_name, messages, result, accessor, *memory_resource, stream_name);
DiscardValueResultStream stream;
spdlog::trace("Start transaction in stream '{}'", stream_name);
utils::OnScopeExit cleanup{[&interpreter, &result]() {
result.rows.clear();
interpreter->Abort();
}};
const static std::map<std::string, storage::v3::PropertyValue> empty_parameters{};
uint32_t i = 0;
while (true) {
try {
interpreter->BeginTransaction();
for (auto &row : result.rows) {
spdlog::trace("Processing row in stream '{}'", stream_name);
auto [query_value, params_value] = ExtractTransformationResult(row.values, transformation_name, stream_name);
storage::v3::PropertyValue params_prop = storage::v3::TypedToPropertyValue(params_value);
std::string query{query_value.ValueString()};
spdlog::trace("Executing query '{}' in stream '{}'", query, stream_name);
auto prepare_result =
interpreter->Prepare(query, params_prop.IsNull() ? empty_parameters : params_prop.ValueMap(), nullptr);
if (!interpreter_context->auth_checker->IsUserAuthorized(owner, prepare_result.privileges)) {
throw StreamsException{
"Couldn't execute query '{}' for stream '{}' because the owner is not authorized to execute the "
"query!",
query, stream_name};
}
interpreter->PullAll(&stream);
}
spdlog::trace("Commit transaction in stream '{}'", stream_name);
interpreter->CommitTransaction();
result.rows.clear();
break;
} catch (const query::v2::TransactionSerializationException &e) {
interpreter->Abort();
if (i == total_retries) {
throw;
}
++i;
std::this_thread::sleep_for(retry_interval);
}
}
};
auto insert_result = map.try_emplace(
stream_name, StreamData<TStream>{std::move(stream_info.common_info.transformation_name), std::move(owner),
std::make_unique<SynchronizedStreamSource<TStream>>(
stream_name, std::move(stream_info), std::move(consumer_function))});
MG_ASSERT(insert_result.second, "Unexpected error during storing consumer '{}'", stream_name);
return insert_result.first;
}
void Streams::RestoreStreams() {
spdlog::info("Loading streams...");
auto locked_streams_map = streams_.Lock();
MG_ASSERT(locked_streams_map->empty(), "Cannot restore streams when some streams already exist!");
for (const auto &[stream_name, stream_data] : storage_) {
const auto get_failed_message = [&stream_name = stream_name](const std::string_view message,
const std::string_view nested_message) {
return fmt::format("Failed to load stream '{}', because: {} caused by {}", stream_name, message, nested_message);
};
const auto create_consumer = [&, &stream_name = stream_name, this]<typename T>(StreamStatus<T> status,
auto &&stream_json_data) {
try {
stream_json_data.get_to(status);
} catch (const nlohmann::json::type_error &exception) {
spdlog::warn(get_failed_message("invalid type conversion", exception.what()));
return;
} catch (const nlohmann::json::out_of_range &exception) {
spdlog::warn(get_failed_message("non existing field", exception.what()));
return;
}
MG_ASSERT(status.name == stream_name, "Expected stream name is '{}', but got '{}'", status.name, stream_name);
try {
auto it = CreateConsumer<T>(*locked_streams_map, stream_name, std::move(status.info), std::move(status.owner));
if (status.is_running) {
std::visit(
[&](const auto &stream_data) {
auto stream_source_ptr = stream_data.stream_source->Lock();
stream_source_ptr->Start();
},
it->second);
}
spdlog::info("Stream '{}' is loaded", stream_name);
} catch (const utils::BasicException &exception) {
spdlog::warn(get_failed_message("unexpected error", exception.what()));
}
};
auto stream_json_data = nlohmann::json::parse(stream_data);
if (const auto it = stream_json_data.find(kType); it != stream_json_data.end()) {
const auto stream_type = static_cast<StreamSourceType>(*it);
switch (stream_type) {
case StreamSourceType::KAFKA:
create_consumer(StreamStatus<KafkaStream>{}, std::move(stream_json_data));
break;
case StreamSourceType::PULSAR:
create_consumer(StreamStatus<PulsarStream>{}, std::move(stream_json_data));
break;
}
} else {
spdlog::warn(
"Unable to load stream '{}', because it does not contain the type of the stream. Most probably the stream "
"was saved before Memgraph 2.1. Please recreate the stream manually to make it work. For more information "
"please check https://memgraph.com/docs/memgraph/changelog#v210---nov-22-2021 .",
stream_json_data.value(kStreamName, "<invalid format>"));
}
}
}
void Streams::Drop(const std::string &stream_name) {
auto locked_streams = streams_.Lock();
auto it = GetStream(*locked_streams, stream_name);
// streams_ is write locked, which means there is no access to it outside of this function, thus only the Test
// function can be executing with the consumer, nothing else.
// By acquiring the write lock here for the consumer, we make sure there is
// no running Test function for this consumer, therefore it can be erased.
std::visit([&](const auto &stream_data) { stream_data.stream_source->Lock(); }, it->second);
locked_streams->erase(it);
if (!storage_.Delete(stream_name)) {
throw StreamsException("Couldn't delete stream '{}' from persistent store!", stream_name);
}
// TODO(antaljanosbenjamin) Release the transformation
}
void Streams::Start(const std::string &stream_name) {
auto locked_streams = streams_.Lock();
auto it = GetStream(*locked_streams, stream_name);
std::visit(
[&, this](const auto &stream_data) {
auto stream_source_ptr = stream_data.stream_source->Lock();
stream_source_ptr->Start();
Persist(CreateStatus(stream_name, stream_data.transformation_name, stream_data.owner, *stream_source_ptr));
},
it->second);
}
void Streams::StartWithLimit(const std::string &stream_name, uint64_t batch_limit,
std::optional<std::chrono::milliseconds> timeout) const {
std::optional locked_streams{streams_.ReadLock()};
auto it = GetStream(**locked_streams, stream_name);
std::visit(
[&](const auto &stream_data) {
const auto locked_stream_source = stream_data.stream_source->ReadLock();
locked_streams.reset();
locked_stream_source->StartWithLimit(batch_limit, timeout);
},
it->second);
}
void Streams::Stop(const std::string &stream_name) {
auto locked_streams = streams_.Lock();
auto it = GetStream(*locked_streams, stream_name);
std::visit(
[&, this](const auto &stream_data) {
auto stream_source_ptr = stream_data.stream_source->Lock();
stream_source_ptr->Stop();
Persist(CreateStatus(stream_name, stream_data.transformation_name, stream_data.owner, *stream_source_ptr));
},
it->second);
}
void Streams::StartAll() {
for (auto locked_streams = streams_.Lock(); auto &[stream_name, stream_data] : *locked_streams) {
std::visit(
[&stream_name = stream_name, this](const auto &stream_data) {
auto locked_stream_source = stream_data.stream_source->Lock();
if (!locked_stream_source->IsRunning()) {
locked_stream_source->Start();
Persist(
CreateStatus(stream_name, stream_data.transformation_name, stream_data.owner, *locked_stream_source));
}
},
stream_data);
}
}
void Streams::StopAll() {
for (auto locked_streams = streams_.Lock(); auto &[stream_name, stream_data] : *locked_streams) {
std::visit(
[&stream_name = stream_name, this](const auto &stream_data) {
auto locked_stream_source = stream_data.stream_source->Lock();
if (locked_stream_source->IsRunning()) {
locked_stream_source->Stop();
Persist(
CreateStatus(stream_name, stream_data.transformation_name, stream_data.owner, *locked_stream_source));
}
},
stream_data);
}
}
std::vector<StreamStatus<>> Streams::GetStreamInfo() const {
std::vector<StreamStatus<>> result;
{
for (auto locked_streams = streams_.ReadLock(); const auto &[stream_name, stream_data] : *locked_streams) {
std::visit(
[&, &stream_name = stream_name](const auto &stream_data) {
auto locked_stream_source = stream_data.stream_source->ReadLock();
auto info = locked_stream_source->Info(stream_data.transformation_name);
result.emplace_back(StreamStatus<>{stream_name, StreamType(*locked_stream_source),
locked_stream_source->IsRunning(), std::move(info.common_info),
stream_data.owner});
},
stream_data);
}
}
return result;
}
TransformationResult Streams::Check(const std::string &stream_name, std::optional<std::chrono::milliseconds> timeout,
std::optional<uint64_t> batch_limit) const {
std::optional locked_streams{streams_.ReadLock()};
auto it = GetStream(**locked_streams, stream_name);
return std::visit(
[&](const auto &stream_data) {
// This depends on the fact that Drop will first acquire a write lock to the consumer, and erase it only after
// that
const auto locked_stream_source = stream_data.stream_source->ReadLock();
const auto transformation_name = stream_data.transformation_name;
locked_streams.reset();
auto *memory_resource = utils::NewDeleteResource();
mgp_result result{nullptr, memory_resource};
TransformationResult test_result;
auto consumer_function = [interpreter_context = interpreter_context_, memory_resource, &stream_name,
&transformation_name = transformation_name, &result,
&test_result]<typename T>(const std::vector<T> &messages) mutable {
auto accessor = interpreter_context->db->Access();
CallCustomTransformation(transformation_name, messages, result, accessor, *memory_resource, stream_name);
auto result_row = std::vector<TypedValue>();
result_row.reserve(kCheckStreamResultSize);
auto queries_and_parameters = std::vector<TypedValue>(result.rows.size());
std::transform(
result.rows.cbegin(), result.rows.cend(), queries_and_parameters.begin(), [&](const auto &row) {
auto [query, parameters] = ExtractTransformationResult(row.values, transformation_name, stream_name);
return std::map<std::string, TypedValue>{{"query", std::move(query)},
{"parameters", std::move(parameters)}};
});
result_row.emplace_back(std::move(queries_and_parameters));
auto messages_list = std::vector<TypedValue>(messages.size());
std::transform(messages.cbegin(), messages.cend(), messages_list.begin(), [](const auto &message) {
return std::string_view(message.Payload().data(), message.Payload().size());
});
result_row.emplace_back(std::move(messages_list));
test_result.emplace_back(std::move(result_row));
};
locked_stream_source->Check(timeout, batch_limit, consumer_function);
return test_result;
},
it->second);
}
} // namespace memgraph::query::v2::stream

View File

@@ -1,206 +0,0 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#pragma once
#include <concepts>
#include <functional>
#include <map>
#include <optional>
#include <type_traits>
#include <unordered_map>
#include <json/json.hpp>
#include "integrations/kafka/consumer.hpp"
#include "kvstore/kvstore.hpp"
#include "query/v2/bindings/typed_value.hpp"
#include "query/v2/stream/common.hpp"
#include "query/v2/stream/sources.hpp"
#include "storage/v3/property_value.hpp"
#include "utils/event_counter.hpp"
#include "utils/exceptions.hpp"
#include "utils/rw_lock.hpp"
#include "utils/synchronized.hpp"
class StreamsTest;
namespace memgraph::query::v2 {
struct InterpreterContext;
namespace stream {
class StreamsException : public utils::BasicException {
public:
using BasicException::BasicException;
};
template <typename T>
struct StreamInfo;
template <>
struct StreamInfo<void> {
using Type = CommonStreamInfo;
};
template <Stream TStream>
struct StreamInfo<TStream> {
using Type = typename TStream::StreamInfo;
};
template <typename T>
using StreamInfoType = typename StreamInfo<T>::Type;
template <typename T = void>
struct StreamStatus {
std::string name;
StreamSourceType type;
bool is_running;
StreamInfoType<T> info;
std::optional<std::string> owner;
};
using TransformationResult = std::vector<std::vector<TypedValue>>;
/// Manages Kafka consumers.
///
/// This class is responsible for all query supported actions to happen.
class Streams final {
friend StreamsTest;
public:
/// Initializes the streams.
///
/// @param interpreter_context context to use to run the result of transformations
/// @param directory a directory path to store the persisted streams metadata
Streams(InterpreterContext *interpreter_context, std::filesystem::path directory);
/// Restores the streams from the persisted metadata.
/// The restoration is done in a best effort manner, therefore no exception is thrown on failure, but the error is
/// logged. If a stream was running previously, then after restoration it will be started.
/// This function should only be called when there are no existing streams.
void RestoreStreams();
/// Creates a new import stream.
/// The create implies connecting to the server to get metadata necessary to initialize the stream. This
/// method assures there is no other stream with the same name.
///
/// @param stream_name the name of the stream which can be used to uniquely identify the stream
/// @param stream_info the necessary informations needed to create the Kafka consumer and transform the messages
///
/// @throws StreamsException if the stream with the same name exists or if the creation of Kafka consumer fails
template <Stream TStream>
void Create(const std::string &stream_name, typename TStream::StreamInfo info, std::optional<std::string> owner);
/// Deletes an existing stream and all the data that was persisted.
///
/// @param stream_name name of the stream that needs to be deleted.
///
/// @throws StreamsException if the stream doesn't exist or if the persisted metadata can't be deleted.
void Drop(const std::string &stream_name);
/// Start consuming from a stream.
///
/// @param stream_name name of the stream that needs to be started
///
/// @throws StreamsException if the stream doesn't exist or if the metadata cannot be persisted
/// @throws ConsumerRunningException if the consumer is already running
void Start(const std::string &stream_name);
/// Start consuming from a stream.
///
/// @param stream_name name of the stream that needs to be started
/// @param batch_limit number of batches we want to consume before stopping
/// @param timeout the maximum duration during which the command should run.
///
/// @throws StreamsException if the stream doesn't exist
/// @throws ConsumerRunningException if the consumer is already running
void StartWithLimit(const std::string &stream_name, uint64_t batch_limit,
std::optional<std::chrono::milliseconds> timeout) const;
/// Stop consuming from a stream.
///
/// @param stream_name name of the stream that needs to be stopped
///
/// @throws StreamsException if the stream doesn't exist or if the metadata cannot be persisted
/// @throws ConsumerStoppedException if the consumer is already stopped
void Stop(const std::string &stream_name);
/// Start consuming from all streams that are stopped.
///
/// @throws StreamsException if the metadata cannot be persisted
void StartAll();
/// Stop consuming from all streams that are running.
///
/// @throws StreamsException if the metadata cannot be persisted
void StopAll();
/// Return current status for all streams.
/// It might happend that the is_running field is out of date if the one of the streams stops during the invocation of
/// this function because of an error.
std::vector<StreamStatus<>> GetStreamInfo() const;
/// Do a dry-run consume from a stream.
///
/// @param stream_name name of the stream we want to test
/// @param batch_limit number of batches we want to test before stopping
/// @param timeout the maximum duration during which the command should run.
///
/// @returns A vector of vectors of TypedValue. Each subvector contains two elements, the query string and the
/// nullable parameters map.
///
/// @throws StreamsException if the stream doesn't exist
/// @throws ConsumerRunningException if the consumer is alredy running
/// @throws ConsumerCheckFailedException if the transformation function throws any std::exception during processing
TransformationResult Check(const std::string &stream_name,
std::optional<std::chrono::milliseconds> timeout = std::nullopt,
std::optional<uint64_t> batch_limit = std::nullopt) const;
private:
template <Stream TStream>
using SynchronizedStreamSource = utils::Synchronized<TStream, utils::WritePrioritizedRWLock>;
template <Stream TStream>
struct StreamData {
std::string transformation_name;
std::optional<std::string> owner;
std::unique_ptr<SynchronizedStreamSource<TStream>> stream_source;
};
using StreamDataVariant = std::variant<StreamData<KafkaStream>, StreamData<PulsarStream>>;
using StreamsMap = std::unordered_map<std::string, StreamDataVariant>;
using SynchronizedStreamsMap = utils::Synchronized<StreamsMap, utils::WritePrioritizedRWLock>;
template <Stream TStream>
StreamsMap::iterator CreateConsumer(StreamsMap &map, const std::string &stream_name,
typename TStream::StreamInfo stream_info, std::optional<std::string> owner);
template <Stream TStream>
void Persist(StreamStatus<TStream> &&status) {
const std::string stream_name = status.name;
if (!storage_.Put(stream_name, nlohmann::json(std::move(status)).dump())) {
throw StreamsException{"Couldn't persist steam data for stream '{}'", stream_name};
}
}
void RegisterProcedures();
void RegisterKafkaProcedures();
void RegisterPulsarProcedures();
InterpreterContext *interpreter_context_;
kvstore::KVStore storage_;
SynchronizedStreamsMap streams_;
};
} // namespace stream
} // namespace memgraph::query::v2

View File

@@ -183,7 +183,7 @@ std::shared_ptr<Trigger::TriggerPlan> Trigger::GetPlan(DbAccessor *db_accessor,
[](auto &identifier) { return &identifier.first; });
auto logical_plan = MakeLogicalPlan(std::move(ast_storage), utils::Downcast<CypherQuery>(parsed_statements_.query),
parsed_statements_.parameters, db_accessor, predefined_identifiers);
parsed_statements_.parameters, nullptr, predefined_identifiers);
trigger_plan_ = std::make_shared<TriggerPlan>(std::move(logical_plan), std::move(identifiers));
}
@@ -210,8 +210,8 @@ void Trigger::Execute(DbAccessor *dba, utils::MonotonicBufferResource *execution
ctx.symbol_table = plan.symbol_table();
ctx.evaluation_context.timestamp = QueryTimestamp();
ctx.evaluation_context.parameters = parsed_statements_.parameters;
ctx.evaluation_context.properties = NamesToProperties(plan.ast_storage().properties_, dba);
ctx.evaluation_context.labels = NamesToLabels(plan.ast_storage().labels_, dba);
ctx.evaluation_context.properties = NamesToProperties(plan.ast_storage().properties_, nullptr);
ctx.evaluation_context.labels = NamesToLabels(plan.ast_storage().labels_, nullptr);
ctx.timer = utils::AsyncTimer(max_execution_time_sec);
ctx.is_shutting_down = is_shutting_down;
ctx.is_profile_query = false;

View File

@@ -77,7 +77,7 @@ struct SetObjectProperty {
std::map<std::string, TypedValue> ToMap(DbAccessor *dba) const {
return {{ObjectString<TAccessor>(), TypedValue{object}},
{"key", TypedValue{dba->PropertyToName(key)}},
{"key", TypedValue{1}}, // TODO Fix trigger
{"old", old_value},
{"new", new_value}};
}
@@ -97,7 +97,7 @@ struct RemovedObjectProperty {
std::map<std::string, TypedValue> ToMap(DbAccessor *dba) const {
return {{ObjectString<TAccessor>(), TypedValue{object}},
{"key", TypedValue{dba->PropertyToName(key)}},
{"key", TypedValue{1}}, // TODO Fix triggers
{"old", old_value}};
}

View File

@@ -1,11 +1,12 @@
define_add_lcp(add_storage_ast lcp_storage_v3_cpp_files generated_lcp_storage_v3_files)
add_storage_ast(bindings/ast/ast.lcp)
add_custom_target(generate_lcp_ast_storage_v3 DEPENDS ${generated_lcp_storage_v3_files})
set(storage_v3_src_files
commit_log.cpp
constraints.cpp
${lcp_storage_v3_cpp_files}
temporal.cpp
durability/durability.cpp
durability/serialization.cpp
durability/snapshot.cpp
durability/wal.cpp
edge_accessor.cpp
indices.cpp
key_store.cpp
@@ -16,29 +17,20 @@ set(storage_v3_src_files
schema_validator.cpp
shard.cpp
storage.cpp
shard_rsm.cpp)
# #### Replication #####
define_add_lcp(add_lcp_storage lcp_storage_cpp_files generated_lcp_storage_files)
add_lcp_storage(replication/rpc.lcp SLK_SERIALIZE)
add_custom_target(generate_lcp_storage_v3 DEPENDS ${generated_lcp_storage_files})
set(storage_v3_src_files
${storage_v3_src_files}
replication/replication_client.cpp
replication/replication_server.cpp
replication/serialization.cpp
replication/slk.cpp
${lcp_storage_cpp_files})
shard_rsm.cpp
bindings/typed_value.cpp
expr.cpp
request_helper.cpp
storage.cpp)
# ######################
find_package(gflags REQUIRED)
find_package(Threads REQUIRED)
add_library(mg-storage-v3 STATIC ${storage_v3_src_files})
add_dependencies(mg-storage-v3 generate_lcp_ast_storage_v3)
target_link_libraries(mg-storage-v3 Threads::Threads mg-utils gflags)
target_include_directories(mg-storage-v3 PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/bindings)
add_dependencies(mg-storage-v3 generate_lcp_storage)
target_link_libraries(mg-storage-v3 mg-rpc mg-slk mg-expr)
target_link_libraries(mg-storage-v3 mg-slk mg-expr mg-io)

File diff suppressed because it is too large Load Diff

View File

@@ -11,10 +11,6 @@
#pragma once
#include <functional>
#include <memory>
#include "storage/v3/bindings/bindings.hpp"
namespace memgraph::query::v2::procedure {
class CypherType;
using CypherTypePtr = std::unique_ptr<CypherType, std::function<void(CypherType *)>>;
} // namespace memgraph::query::v2::procedure
#include "expr/ast/ast_visitor.hpp"

View File

@@ -0,0 +1,19 @@
// Copyright 2022 Memgraph Ltd.
//
// Use of this software is governed by the Business Source License
// included in the file licenses/BSL.txt; by using this file, you agree to be bound by the terms of the Business Source
// License, and you may not use this file except in compliance with the Business Source License.
//
// As of the Change Date specified in that file, in accordance with
// the Business Source License, use of this software will be governed
// by the Apache License, Version 2.0, included in the file
// licenses/APL.txt.
#pragma once
#ifdef MG_AST_INCLUDE_PATH
#error You are probably trying to include some files of mg-expr from both the storage and query engines! You will have a rough time kid!
#endif
#define MG_AST_INCLUDE_PATH "storage/v3/bindings/ast/ast.hpp" // NOLINT(cppcoreguidelines-macro-usage)
#define MG_INJECTED_NAMESPACE_NAME memgraph::storage::v3 // NOLINT(cppcoreguidelines-macro-usage)

View File

@@ -10,10 +10,11 @@
// licenses/APL.txt.
#pragma once
#include <cstdint>
namespace memgraph::storage::v3::replication {
enum class ReplicationMode : std::uint8_t { SYNC, ASYNC };
#include "storage/v3/bindings/bindings.hpp"
enum class ReplicaState : std::uint8_t { READY, REPLICATING, RECOVERY, INVALID };
} // namespace memgraph::storage::v3::replication
#include "expr/ast/cypher_main_visitor.hpp"
namespace memgraph::storage::v3 {
using CypherMainVisitor = memgraph::expr::CypherMainVisitor;
} // namespace memgraph::storage::v3

Some files were not shown because too many files have changed in this diff Show More