Skip to content

KNOWN_ISSUES.md

Latest commit

 

History

History
105 lines (84 loc) · 6.13 KB

KNOWN_ISSUES.md

File metadata and controls

105 lines (84 loc) · 6.13 KB

Known Issues

Last Updated: May 2, 2026

For fixed issues and release history, see CHANGELOG.md.

Active Issues

1. Shortest Path on Dense Graphs

Status: Performance limitation
Error: MEMORY_LIMIT_EXCEEDED or query timeout
Cause: Recursive CTE-based shortest path explores all paths. Dense graphs cause exponential explosion.
Workaround: Use bounded path length: shortestPath((a)-[:FOLLOWS*1..5]->(b))

2. Cypher Writes Are Embedded-Only

Status: By design (v0.6.7+) Cause: ClickHouse server's lightweight UPDATE/DELETE path differs from chdb in maturity and concurrency semantics. ClickGraph's write_guard rejects writes against any executor that isn't EmbeddedChdb (server / remote / sql_only modes) and against any node/edge backed by a source: URI in the schema YAML. Workaround: For server / remote workloads, perform writes via the ClickHouse SQL interface directly, then query through ClickGraph as normal.

3. Unsupported Write Combinations

Status: Pending Phase 5 (MERGE) and Phase 5+ (write extensions) Affected: MERGE, CREATE … RETURN, relationship CREATE, DELETE r for an edge alias, SET a += {…} map-merge, REMOVE a:Label. Each case is rejected with an explicit error in the write pipeline rather than silently falling through. Workaround: Issue a separate MATCH … RETURN after a CREATE to read back. For relationship CREATE today, use Connection::create_edges() (direct API). For edge deletion, Connection::delete_edges().

4. Best-Effort Write Counters

Status: chdb limitation Cause: chdb's lightweight write path doesn't return affected-row counts, so ClickGraph attributes one to the relevant counter per write op (Insert, Update, Delete). Workaround: Issue a MATCH … RETURN count(*) before/after the write for an exact count.

Recently Fixed (February 2026)

Aggregations on Empty Results ✅ (Dec 21, 2025)

Fix: Commit 734d65f - Unified aggregation logic

  • ClickHouse aggregations now properly return 1 row with default values (count=0, etc.)
  • Made extract_select_items() aggregation-aware
  • Unified WITH and RETURN aggregation code paths Tests: test_aggregation_empty_result, test_count_empty_result expecting 1 row with count=0

Neo4j Desktop / NeoDash WebSocket Connection ✅ (Feb 2)

Fix: PR #64 (commit 6755d22) - Full WebSocket Bolt transport

  • Added websocket.rs with WebSocketBoltAdapter implementing AsyncRead/AsyncWrite
  • Server detects HTTP GET/POST requests on Bolt port → WebSocket handshake
  • Neo4j Desktop and NeoDash now connect successfully via Bolt WebSocket Location: src/server/bolt_protocol/websocket.rs + integration in src/server/mod.rs

Pattern Comprehensions ✅ (Feb 13)

Fix: Commit f144108 - Full implementation with CTE+JOIN

  • Added target_label/target_property extraction from pattern AST
  • Fixed aggregation type detection (collect() → GroupArray)
  • Generated INNER JOIN to target node table
  • Fixed relationship name matching and JOIN conditions Tests: tests/integration/test_pattern_comprehensions.py passing

Empty Plans with Column References ✅ (Feb 16)

Fix: Commits e5ca181 + b3697e2

  • Empty plans now use FROM system.one WHERE false for valid SQL
  • RETURN-only queries (e.g., RETURN 1) properly handled
  • Column references in Empty plans replaced with typed defaults Location: src/render_plan/plan_builder.rs lines 2301-2400

labels(n) on Untyped Nodes ✅ (Feb 17)

Fix: PR #104 - Branch-specific label extraction

  • Extract single label from each UNION branch's GraphNode
  • Temporarily override plan_ctx during projection tagging
  • Tightened VLP detection with is_cte_reference() check Tests: tests/integration/test_labels_untyped_nodes.py with 7 test cases

Out of Scope (by design)

ClickGraph is primarily an analytical query engine. Limited write support exists in embedded mode (v0.6.7+); the rest stays out of scope:

  • ✅ Cypher writes (CREATE, SET, DELETE, REMOVE) in embedded mode only, against ClickGraph-managed (non-source:) tables — see Active Issues #2 / #3 for caveats
  • ❌ Cypher writes in server / remote / sql_only modes
  • ❌ MERGE clause (planned for v0.7.x)
  • ❌ Schema DDL (CREATE INDEX, CREATE CONSTRAINT)
  • ❌ Transaction management (BEGIN, COMMIT, ROLLBACK)
  • ❌ Stored procedures (APOC/GDS) — only built-in db.* procedures

Historical Fixes (Pre-February 2026)

Issue Fix PR
Unlabeled nodes + labeled rel invalid UNION branches (#6) Phase 0 Case 5 in TypeInference path-direction-fix
Property filtering on unlabeled nodes invalid branches (#7) Phase 2 property-based candidate filtering path-direction-fix
Relationship property access fails with CTE structure (#8) pattern_union CTE exposes direct columns path-direction-fix
Query-level UNION fails plan context merge (#9) Fixed branch pruning for TypeInference placeholders path-direction-fix
count(n) on untyped nodes returns wrong value Aggregation placed above UNION, not inside branches path-direction-fix
FOLLOWS self-join returns empty from_node/to_node aliases for same-table JOINs path-direction-fix
JSON column order alphabetical instead of query order serde_json preserve_order feature path-direction-fix
UNION __label__ injection not projection-guided returns_whole_entity() checks Projection items path-direction-fix
UNWIND crash with collect(DISTINCT) Fixed infinite WITH iteration + DISTINCT handling #91
Cross-session ID leakage between tenants IdMapper scoped by schema + tenant #85
Query cache ignores tenant_id Cache key includes tenant_id + view_parameters main
PackStream arrays/objects not encoded Recursive PackStream encoding #83
UNION column mismatch (literal + aggregate) Extracted helper, fixed branch construction #84
Browser click-to-expand failures (5 schema types) CTE naming, JOIN fixes, VLP rendering #70–#82
Browser EXPLAIN probe noise EXPLAIN handler returns empty SUCCESS #85
Session commands not working in browser ConnectionState::Streaming fix #85