ANN_SEARCH + WHERE clause causing full table scan in OceanBase 4.

We are building a multi-tenant enterprise RAG system and just hit a gnarly performance issue with OceanBase 4.3.x (MySQL mode). The setup is pretty straightforward: a large table that holds 1536-dimensional embeddings (FLOAT array) with a native VSAG vector index, plus a tenant_id integer column with a standard B-tree index and a status column. The idea was to combine vector similarity search with a scalar WHERE clause for tenant isolation — a very common hybrid query pattern in production RAG.

Here’s the exact query we’re stuck with:

SELECT doc_id, chunk_text
FROM enterprise_knowledge_base
WHERE tenant_id = 10025 AND status = 'ACTIVE'
ORDER BY APPROX_DISTANCE(embedding_data, [0.015, -0.023, 0.112, 0.045])
APPROX_TOP 5;

We expected OceanBase to pre‑filter on tenant_id and status first, then run the ANN search only on the matching rows. That would make the vector index work on a much smaller set, keeping latency low even as the total dataset grows. But when we tested it against a realistic dataset (a few million rows across