Historical NornicDB Test Results¶

Last Updated: November 2025 | Version 0.1.0

Historical test archive: These measurements describe the v0.1.0 test environment and are retained for provenance. They are not the current performance headline. See the current Northwind comparison for direct Neo4j measurements and the BEIR SciFact report for current retrieval-quality evidence.

Executive Summary¶

In this v0.1.0 test run, NornicDB passed the recorded Neo4j-compatibility checks and showed the performance and resource measurements below.

Key Findings¶

Test Environment¶

• Host: macOS (Apple Silicon)
• Docker: Docker Compose with ARM64 images
• Application Version: 4.1.0
• NornicDB Version: 0.1.0
• Test Date: 2025-11-26

1. Batch Operations Test¶

1.1 Small Batch (10 nodes)¶

Time: ~32ms
Throughput: 312 nodes/sec
Status: ✅ PASSED

1.2 Medium Batch (50 nodes)¶

Time: ~26ms
Throughput: 1,923 nodes/sec
Status: ✅ PASSED

1.3 Large Batch (200 nodes)¶

Time: ~246ms
Throughput: 813 nodes/sec
Status: ✅ PASSED

1.4 Stress Test (500 nodes)¶

Time: ~1.1s
Throughput: 454 nodes/sec
Status: ✅ PASSED

Batch Operations Summary¶

2. Memory Footprint Analysis¶

2.1 Container Memory¶

NornicDB: 46.09 MiB / 15.6 GiB (0.29%)
App Server: 79.85 MiB / 15.6 GiB (0.50%)

2.2 Process Memory (inside container)¶

VmPeak:  1,234,928 kB (allocated)
VmSize:  1,234,928 kB (virtual)
VmRSS:      29,368 kB (actual - 29 MiB)
VmData:     58,568 kB (data segment)

2.3 Memory Stability Test¶

Conclusion: Memory usage is stable with no leaks. Go GC efficiently reclaims memory.

2.4 Memory per 1000 Nodes¶

Estimated: ~33-50 MiB per 1000 nodes

3. Query Performance¶

3.1 Count Queries¶

MATCH (n) RETURN count(n)

• Nodes: 909
• Time: ~34ms
• Status: ✅ PASSED

3.2 Filter Queries¶

MATCH (n:LargeTest) WHERE n.idx > 1400 RETURN count(n)

• Result: 100 nodes
• Time: ~32ms
• Status: ✅ PASSED

3.3 Aggregation Queries¶

MATCH (n:LargeTest) RETURN n.batch, count(n) ORDER BY n.batch

• Time: ~41ms
• Status: ✅ PASSED

4. Cypher Function Tests¶

4.1 labels() Function¶

MATCH (c:Concept) RETURN c.name, labels(c) LIMIT 1

• Result: ["Graph Database", ["Concept", "Node"]]
• Status: ✅ PASSED

4.2 id() Function¶

MATCH (c:Concept) RETURN id(c), c.name LIMIT 1

• Result: ["node-38", "LLM Memory"]
• Status: ✅ PASSED

4.3 count() Aggregation¶

MATCH (n) RETURN count(n) as total

• Result: 138
• Status: ✅ PASSED

4.4 WHERE CONTAINS¶

MATCH (n) WHERE n.name CONTAINS "Vector" RETURN n.name

• Result: ["Vector Search"]
• Status: ✅ PASSED

5. E2E Integration Tests¶

5.1 Node Creation¶

5.2 Relationship Creation¶

CREATE (a:Agent:Node {name: "Claudette"})-[:USES]->(t:Tool:Node {name: "NornicDB"})

• Status: ✅ PASSED

5.3 File Indexing¶

• Folder: /app/docs
• Files Indexed: 15
• Chunks Created: 109
• Edges Created: 1,418
• Status: ✅ PASSED

6. Docker Image Comparison¶

7. Database Final State¶

{
  "database": {
    "edges": 1418,
    "nodes": 909
  },
  "server": {
    "active": 1,
    "errors": 0,
    "requests": 55,
    "uptime_seconds": 177.3
  },
  "status": "running"
}

8. Known Limitations¶

8.1 Not Yet Implemented¶

1. MATCH...SET returns empty result in some cases (needs refinement)
2. Complex pattern expressions (p = (a)-[*]->(b)) need full path variable support
3. Some WITH clause edge cases

8.2 Implemented Features (Previously Missing)¶

1. ✅ Relationship traversal - Full support for MATCH (a)-[r:TYPE]->(b) patterns
2. ✅ type(r) function - Returns relationship type correctly
3. ✅ Variable-length paths - [*1..3] patterns supported
4. ✅ shortestPath() - BFS-based shortest path algorithm
5. ✅ allShortestPaths() - Returns all shortest paths
6. ✅ Direction support - Outgoing (->), incoming (<-), both (-)
7. ✅ Relationship type filters - [:KNOWS|FOLLOWS] multiple types
8. ✅ degree(), inDegree(), outDegree() - Graph degree functions
9. ✅ Schema management - Phase 2: CREATE CONSTRAINT/INDEX (89% coverage)
10. ✅ Vector search procedure - Phase 3: Cosine/Euclidean/Dot similarity
11. ✅ Full-text search procedure - Phase 3: BM25-like scoring (IDF+ variant)
12. ✅ Graph traversal procedure - Phase 3: apoc.path.subgraphNodes with BFS
13. ✅ GPU acceleration - Metal, CUDA, OpenCL, Vulkan backends (100% test coverage)
14. ✅ Persistent storage (BadgerDB) - ACID transactions, crash recovery, secondary indexes

8.3 Workarounds¶

• For complex path expressions, use explicit relationship patterns instead

9. Bug Fixes Applied During Testing¶

9.1 Stack Overflow Fix¶

• Issue: Infinite recursion in evaluateExpressionWithContext when property values contained + in strings (like file content)
• Solution: Added hasConcatOperator() helper to properly detect concatenation operators outside of quotes
• Status: ✅ FIXED

9.2 Keyword in Label Names Fix¶

• Issue: Labels containing Cypher keywords (e.g., RemoveReturn, Return, Where) were incorrectly parsed, causing queries like MATCH (n:RemoveReturn) RETURN n to fail with 0 results
• Root Cause: strings.Index(upper, "RETURN") found "RETURN" inside label names like "RemoveReturn"
• Solution: Implemented findKeywordIndex() helper that:
• Uses word boundary detection (not just substring matching)
• Excludes : as a left boundary since it precedes labels in Cypher
• Properly handles keywords embedded in label names, variable names, and property values
• Inspiration: Neo4j's ANTLR lexer uses longest-match-wins and separate token rules for keywords vs identifiers
• Status: ✅ FIXED

10. Conclusion¶

This historical v0.1.0 run recorded the following results:

1. 70x less memory usage
2. 27x smaller Docker image
3. Instant startup (< 1 second)
4. Zero dependencies (pure Go binary)
5. Stable under load (no memory leaks)
6. Full Bolt protocol compatibility

Implemented for Neo4j Compatibility¶

Test Summary¶

Recommended Next Steps¶

1. Add GPU acceleration for vector search (Metal/CUDA) ✅ COMPLETE (Metal, CUDA, OpenCL, Vulkan)
2. Implement HNSW index for faster similarity search ✅ COMPLETE (Vector indexes with similarity search)
3. Add persistent storage backend (BadgerDB) ✅ COMPLETE (Full Engine interface with ACID)
4. Implement composite indexes for multi-property queries ✅ COMPLETE (Full/prefix lookup, filter support)
5. Implement remaining APOC procedures as needed
6. Add transaction support (BEGIN/COMMIT/ROLLBACK)

Appendix: Test Commands¶

Start NornicDB¶

docker compose -f docker-compose.arm64.yml up -d nornicdb

Health Check¶

curl http://localhost:7474/health

Create Nodes¶

curl -X POST http://localhost:7474/db/nornic/tx/commit \
  -H "Content-Type: application/json" \
  -d '{"statements":[{"statement":"CREATE (n:Test {name: \"test\"}) RETURN n"}]}'

Query Nodes¶

curl -X POST http://localhost:7474/db/nornic/tx/commit \
  -H "Content-Type: application/json" \
  -d '{"statements":[{"statement":"MATCH (n) RETURN count(n)"}]}'

Check Memory¶

docker stats --no-stream nornicdb