Files
memgraph/src/dbms/replication_handler.cpp
andrejtonev 8b9e1fa08b Replication refactor part 6 (#1484)
Single (instance level) connection to a replica (messages from all databases get multiplexed through it)
ReplicationClient split in two: ReplicationClient and ReplicationStorageClient
New ReplicationClient, moved under replication, handles the raw connection, owned by MainRoleData
ReplicationStorageClient handles the storage <-> replica state machine and holds to a stream
Removed epoch and storage from *Clients
rpc::Stream proactively aborts on error and sets itself to a defunct state
Removed HandleRpcFailure, instead we simply log the error and let the FrequentCheck handle re-connection
replica_state is now a synced variable
ReplicaStorageClient state machine bugfixes
Single FrequentCheck that goes through DBMS
Moved ReplicationState under DbmsHandler
Moved some replication startup logic under the DbmsHandler's constructor
Removed InMemoryReplicationClient
CreateReplicationClient has been removed from Storage
Simplified GetRecoverySteps and made safer

---------

Co-authored-by: Gareth Lloyd <gareth.lloyd@memgraph.io>
2023-11-23 11:02:35 +01:00

263 lines
11 KiB
C++

// Copyright 2023 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 "dbms/replication_handler.hpp"
#include "dbms/constants.hpp"
#include "dbms/dbms_handler.hpp"
#include "dbms/inmemory/replication_handlers.hpp"
#include "dbms/inmemory/storage_helper.hpp"
#include "dbms/replication_client.hpp"
#include "replication/state.hpp"
using memgraph::replication::ReplicationClientConfig;
using memgraph::replication::ReplicationState;
using memgraph::replication::RoleMainData;
using memgraph::replication::RoleReplicaData;
namespace memgraph::dbms {
namespace {
std::string RegisterReplicaErrorToString(RegisterReplicaError error) {
switch (error) {
using enum RegisterReplicaError;
case NAME_EXISTS:
return "NAME_EXISTS";
case END_POINT_EXISTS:
return "END_POINT_EXISTS";
case CONNECTION_FAILED:
return "CONNECTION_FAILED";
case COULD_NOT_BE_PERSISTED:
return "COULD_NOT_BE_PERSISTED";
}
}
} // namespace
ReplicationHandler::ReplicationHandler(DbmsHandler &dbms_handler) : dbms_handler_(dbms_handler) {}
bool ReplicationHandler::SetReplicationRoleMain() {
auto const main_handler = [](RoleMainData const &) {
// If we are already MAIN, we don't want to change anything
return false;
};
auto const replica_handler = [this](RoleReplicaData const &) {
// STEP 1) bring down all REPLICA servers
dbms_handler_.ForEach([](Database *db) {
auto *storage = db->storage();
// Remember old epoch + storage timestamp association
storage->PrepareForNewEpoch();
});
// STEP 2) Change to MAIN
// TODO: restore replication servers if false?
if (!dbms_handler_.ReplicationState().SetReplicationRoleMain()) {
// TODO: Handle recovery on failure???
return false;
}
// STEP 3) We are now MAIN, update storage local epoch
const auto &epoch =
std::get<RoleMainData>(std::as_const(dbms_handler_.ReplicationState()).ReplicationData()).epoch_;
dbms_handler_.ForEach([&](Database *db) {
auto *storage = db->storage();
storage->repl_storage_state_.epoch_ = epoch;
});
return true;
};
// TODO: under lock
return std::visit(utils::Overloaded{main_handler, replica_handler},
dbms_handler_.ReplicationState().ReplicationData());
}
bool ReplicationHandler::SetReplicationRoleReplica(const memgraph::replication::ReplicationServerConfig &config) {
// We don't want to restart the server if we're already a REPLICA
if (dbms_handler_.ReplicationState().IsReplica()) {
return false;
}
// TODO StorageState needs to be synched. Could have a dangling reference if someone adds a database as we are
// deleting the replica.
// Remove database specific clients
dbms_handler_.ForEach([&](Database *db) {
auto *storage = db->storage();
storage->repl_storage_state_.replication_clients_.WithLock([](auto &clients) { clients.clear(); });
});
// Remove instance level clients
std::get<RoleMainData>(dbms_handler_.ReplicationState().ReplicationData()).registered_replicas_.clear();
// Creates the server
dbms_handler_.ReplicationState().SetReplicationRoleReplica(config);
// Start
const auto success =
std::visit(utils::Overloaded{[](RoleMainData const &) {
// ASSERT
return false;
},
[this](RoleReplicaData const &data) {
// Register handlers
InMemoryReplicationHandlers::Register(&dbms_handler_, *data.server);
if (!data.server->Start()) {
spdlog::error("Unable to start the replication server.");
return false;
}
return true;
}},
dbms_handler_.ReplicationState().ReplicationData());
// TODO Handle error (restore to main?)
return success;
}
auto ReplicationHandler::RegisterReplica(const memgraph::replication::ReplicationClientConfig &config)
-> memgraph::utils::BasicResult<RegisterReplicaError> {
MG_ASSERT(dbms_handler_.ReplicationState().IsMain(), "Only main instance can register a replica!");
auto instance_client = dbms_handler_.ReplicationState().RegisterReplica(config);
if (instance_client.HasError()) switch (instance_client.GetError()) {
case memgraph::replication::RegisterReplicaError::NOT_MAIN:
MG_ASSERT(false, "Only main instance can register a replica!");
return {};
case memgraph::replication::RegisterReplicaError::NAME_EXISTS:
return memgraph::dbms::RegisterReplicaError::NAME_EXISTS;
case memgraph::replication::RegisterReplicaError::END_POINT_EXISTS:
return memgraph::dbms::RegisterReplicaError::END_POINT_EXISTS;
case memgraph::replication::RegisterReplicaError::COULD_NOT_BE_PERSISTED:
return memgraph::dbms::RegisterReplicaError::COULD_NOT_BE_PERSISTED;
case memgraph::replication::RegisterReplicaError::SUCCESS:
break;
}
if (!allow_mt_repl && dbms_handler_.All().size() > 1) {
spdlog::warn("Multi-tenant replication is currently not supported!");
}
bool all_clients_good = true;
// Add database specific clients (NOTE Currently all databases are connected to each replica)
dbms_handler_.ForEach([&](Database *db) {
auto *storage = db->storage();
if (!allow_mt_repl && storage->id() != kDefaultDB) {
return;
}
// TODO: ATM only IN_MEMORY_TRANSACTIONAL, fix other modes
if (storage->storage_mode_ != storage::StorageMode::IN_MEMORY_TRANSACTIONAL) return;
all_clients_good &=
storage->repl_storage_state_.replication_clients_.WithLock([storage, &instance_client](auto &storage_clients) {
auto client = std::make_unique<storage::ReplicationStorageClient>(*instance_client.GetValue());
client->Start(storage);
// After start the storage <-> replica state should be READY or RECOVERING (if correctly started)
// MAYBE_BEHIND isn't a statement of the current state, this is the default value
// Failed to start due to branching of MAIN and REPLICA
if (client->State() == storage::replication::ReplicaState::MAYBE_BEHIND) {
return false;
}
storage_clients.push_back(std::move(client));
return true;
});
});
// NOTE Currently if any databases fails, we revert back
if (!all_clients_good) {
spdlog::error("Failed to register all databases to the REPLICA \"{}\"", config.name);
UnregisterReplica(config.name);
return RegisterReplicaError::CONNECTION_FAILED;
}
// No client error, start instance level client
StartReplicaClient(dbms_handler_, *instance_client.GetValue());
return {};
}
auto ReplicationHandler::UnregisterReplica(std::string_view name) -> UnregisterReplicaResult {
auto const replica_handler = [](RoleReplicaData const &) -> UnregisterReplicaResult {
return UnregisterReplicaResult::NOT_MAIN;
};
auto const main_handler = [this, name](RoleMainData &mainData) -> UnregisterReplicaResult {
if (!dbms_handler_.ReplicationState().TryPersistUnregisterReplica(name)) {
return UnregisterReplicaResult::COULD_NOT_BE_PERSISTED;
}
// Remove database specific clients
dbms_handler_.ForEach([name](Database *db) {
db->storage()->repl_storage_state_.replication_clients_.WithLock([&name](auto &clients) {
std::erase_if(clients, [name](const auto &client) { return client->Name() == name; });
});
});
// Remove instance level clients
auto const n_unregistered =
std::erase_if(mainData.registered_replicas_, [name](auto const &client) { return client.name_ == name; });
return n_unregistered != 0 ? UnregisterReplicaResult::SUCCESS : UnregisterReplicaResult::CAN_NOT_UNREGISTER;
};
return std::visit(utils::Overloaded{main_handler, replica_handler},
dbms_handler_.ReplicationState().ReplicationData());
}
auto ReplicationHandler::GetRole() const -> memgraph::replication::ReplicationRole {
return dbms_handler_.ReplicationState().GetRole();
}
bool ReplicationHandler::IsMain() const { return dbms_handler_.ReplicationState().IsMain(); }
bool ReplicationHandler::IsReplica() const { return dbms_handler_.ReplicationState().IsReplica(); }
// Per storage
// NOTE Storage will connect to all replicas. Future work might change this
void RestoreReplication(replication::ReplicationState &repl_state, storage::Storage &storage) {
spdlog::info("Restoring replication role.");
/// MAIN
auto const recover_main = [&storage](RoleMainData &mainData) {
// Each individual client has already been restored and started. Here we just go through each database and start its
// client
for (auto &instance_client : mainData.registered_replicas_) {
spdlog::info("Replica {} restoration started for {}.", instance_client.name_, storage.id());
const auto &ret = storage.repl_storage_state_.replication_clients_.WithLock(
[&](auto &storage_clients) -> utils::BasicResult<RegisterReplicaError> {
auto client = std::make_unique<storage::ReplicationStorageClient>(instance_client);
client->Start(&storage);
// After start the storage <-> replica state should be READY or RECOVERING (if correctly started)
// MAYBE_BEHIND isn't a statement of the current state, this is the default value
// Failed to start due to branching of MAIN and REPLICA
if (client->State() == storage::replication::ReplicaState::MAYBE_BEHIND) {
spdlog::warn("Connection failed when registering replica {}. Replica will still be registered.",
instance_client.name_);
}
storage_clients.push_back(std::move(client));
return {};
});
if (ret.HasError()) {
MG_ASSERT(RegisterReplicaError::CONNECTION_FAILED != ret.GetError());
LOG_FATAL("Failure when restoring replica {}: {}.", instance_client.name_,
RegisterReplicaErrorToString(ret.GetError()));
}
spdlog::info("Replica {} restored for {}.", instance_client.name_, storage.id());
}
spdlog::info("Replication role restored to MAIN.");
};
/// REPLICA
auto const recover_replica = [](RoleReplicaData const &data) { /*nothing to do*/ };
std::visit(
utils::Overloaded{
recover_main,
recover_replica,
},
repl_state.ReplicationData());
}
} // namespace memgraph::dbms