query_router Benchmarks

The query router integrates all engines and routes queries based on parsed AST type.

Relational Operations

Operation	Time
SELECT * (100 rows)	17 us
SELECT WHERE	17 us
INSERT	290 us
UPDATE	6.5 ms

Graph Operations

Operation	Time
NODE CREATE	2.3 us
EDGE CREATE	3.5 us
NEIGHBORS	1.8 us
PATH (1 -> 10)	85 us
FIND NODE	1.2 us

Vector Operations

Operation	Time
EMBED STORE (128d)	28 us
EMBED GET	1.5 us
SIMILAR LIMIT 5 (100 vectors)	10 ms
SIMILAR LIMIT 10 (100 vectors)	10 ms

Mixed Workload

Configuration	Time	Queries/s
5 mixed queries (SELECT, NEIGHBORS, SIMILAR, INSERT, NODE)	11 ms	455/s

Insert Throughput

Batch Size	Time	Rows/s
100	29 ms	3.4K/s
500	145 ms	3.4K/s
1,000	290 ms	3.4K/s

Analysis

Parse overhead: Parser adds ~200ns-2us per query (negligible vs execution)
Routing overhead: AST-based routing is O(1) pattern matching
Relational: SELECT is fast (17us); UPDATE scans all rows (6.5ms for 100 rows)
Graph: Node/edge creation ~2-3us; path finding scales with path length
Vector: Similarity search dominates mixed workloads (~10ms for 100 vectors)
Bottleneck identification: SIMILAR queries are the slowest operation; use HNSW index for large vector stores

Query Routing Flow

Query String
    │
    ▼
┌─────────┐
│ Parser  │  ~500ns
└────┬────┘
     │
     ▼
┌─────────┐
│   AST   │
└────┬────┘
     │
     ▼
┌─────────────┐
│   Router    │  O(1) dispatch
└──────┬──────┘
       │
       ├──► RelationalEngine
       ├──► GraphEngine
       ├──► VectorEngine
       ├──► Vault
       ├──► Cache
       └──► BlobStore

Performance Recommendations

Query Type	Optimization
High SELECT volume	Create hash indexes on filter columns
Large vector search	Build HNSW index
Graph traversals	Use NEIGHBORS with LIMIT
Batch inserts	Use batch_insert() API
Mixed workloads	Profile to identify bottlenecks

Keyboard shortcuts

Neumann