python.sqlalchemy.pgvector_optimization

Performance High

Detects pgvector queries without proper index usage.

Why It Matters

Unoptimized vector queries:

• Full table scans — No index means O(n) searches
• Slow queries — Vector similarity is CPU-intensive
• Timeout failures — Large datasets cause timeouts

Example

# ❌ Before (no index, full scan)
# Missing index on embedding column
results = session.query(Document).order_by(
    Document.embedding.l2_distance(query_vector)
).limit(10).all()

# ✅ After (with proper index)
# Create index in migrations:
# CREATE INDEX ON documents USING ivfflat (embedding vector_l2_ops) WITH (lists = 100);

# And ensure probes are set
session.execute(text("SET ivfflat.probes = 10"))
results = session.query(Document).order_by(
    Document.embedding.l2_distance(query_vector)
).limit(10).all()

What Unfault Detects

• Vector queries without indexes
• Missing ivfflat.probes setting
• Unbounded vector searches

Related Rules

• python.n_plus_one_queries