Add E2E test for optimizing merge on supernodes

src/query/plan/rewrite/index_lookup.hpp

@@ -750,7 +750,7 @@ class IndexLookupRewriter final : public HierarchicalLogicalOperatorVisitor {
     auto maybe_label = FindBestLabelIndex(labels);
     if (!maybe_label) return nullptr;
     const auto &label = *maybe_label;
-    if (max_vertex_count && db_->VerticesCount(GetLabel(label)) > *max_vertex_count) {
+    if (max_vertex_count && *max_vertex_count >= 1 && db_->VerticesCount(GetLabel(label)) > *max_vertex_count) {
       // Don't create an indexed lookup, since we have more labeled vertices
       // than the allowed count.
       return nullptr;

tests/e2e/analyze_graph/optimize_indexes.py

@@ -13,6 +13,7 @@ import sys
 
 import pytest
 from common import connect, execute_and_fetch_all
+from gqlalchemy import Memgraph
 
 # E2E tests for checking query semantic
 # ------------------------------------
@@ -278,5 +279,58 @@ def test_same_avg_group_size_diff_distribution(connect):
     execute_and_fetch_all(cursor, "DROP INDEX ON :Label(id2);")
 
 
+def test_given_supernode_when_expanding_then_expand_other_way_around():
+    memgraph = Memgraph()
+
+    memgraph.execute("FOREACH (i in range(1, 20000) | CREATE (:Node {id: i}));")
+    memgraph.execute("CREATE (:SuperNode {id: 1});")
+    memgraph.execute("CREATE INDEX ON :SuperNode(id);")
+    memgraph.execute("CREATE INDEX ON :SuperNode;")
+    memgraph.execute("CREATE INDEX ON :Node(id);")
+    memgraph.execute("CREATE INDEX ON :Node;")
+    memgraph.execute("match (n:Node) match (s:SuperNode {id: 1}) merge (n)<-[:HAS_REL_TO]-(s);")
+
+    result_without_analysis = list(
+        memgraph.execute_and_fetch("explain match (n:Node) match (s:SuperNode {id: 1}) merge (n)<-[:HAS_REL_TO]-(s);")
+    )
+    result_without_analysis = [x["QUERY PLAN"] for x in result_without_analysis]
+    expected_explain_before_analysis = [
+        f" * EmptyResult",
+        f" * Merge",
+        f" |\\ On Match",
+        f" | * Expand (s)-[anon3:HAS_REL_TO]->(n)",
+        f" | * Once",
+        f" |\\ On Create",
+        f" | * CreateExpand (n)<-[anon3:HAS_REL_TO]-(s)",
+        f" | * Once",
+        f" * ScanAllByLabel (n :Node)",
+        f" * ScanAllByLabelPropertyValue (s :SuperNode {{id}})",
+        f" * Once",
+    ]
+
+    memgraph.execute("analyze graph;")
+
+    result_with_analysis = list(
+        memgraph.execute_and_fetch("explain match (n:Node) match (s:SuperNode {id: 1}) merge (n)<-[:HAS_REL_TO]-(s);")
+    )
+    result_with_analysis = [x["QUERY PLAN"] for x in result_with_analysis]
+    expected_explain_after_analysis = [
+        f" * EmptyResult",
+        f" * Merge",
+        f" |\\ On Match",
+        f" | * Expand (n)<-[anon3:HAS_REL_TO]-(s)",
+        f" | * Once",
+        f" |\\ On Create",
+        f" | * CreateExpand (n)<-[anon3:HAS_REL_TO]-(s)",
+        f" | * Once",
+        f" * ScanAllByLabel (n :Node)",
+        f" * ScanAllByLabelPropertyValue (s :SuperNode {{id}})",
+        f" * Once",
+    ]
+
+    assert expected_explain_before_analysis == result_without_analysis
+    assert expected_explain_after_analysis == result_with_analysis
+
+
 if __name__ == "__main__":
     sys.exit(pytest.main([__file__, "-rA"]))

tests/e2e/configuration/default_config.py

@@ -175,8 +175,8 @@ startup_config_dict = {
     "query_cost_planner": ("true", "true", "Use the cost-estimating query planner."),
     "query_plan_cache_ttl": ("60", "60", "Time to live for cached query plans, in seconds."),
     "query_vertex_count_to_expand_existing": (
-        "10",
-        "10",
+        "-1",
+        "-1",
         "Maximum count of indexed vertices which provoke indexed lookup and then expand to existing, instead of a regular expand. Default is 10, to turn off use -1.",
     ),
     "query_max_plans": ("1000", "1000", "Maximum number of generated plans for a query."),

tests/setup.sh

@@ -16,6 +16,7 @@ PIP_DEPS=(
    "pyyaml==5.4.1"
    "six==1.15.0"
    "networkx==2.4"
+   "gqlalchemy==1.3.3"
 )
 cd "$DIR"