src/common/update.rs (1)

611-658: ⚡ Quick win

Apply the same pre-WAL dimension check to do_update_vectors.

do_upsert_points now rejects bad dimensions before queueing, but /points/vectors still converts and forwards straight to update(). If dimension errors for VectorOperations::UpdateVectors are still raised only during WAL apply, wait=false will keep the old “acknowledged but dropped” behavior on this endpoint. Please verify that path, or reuse the same early validation here and add a regression test.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@src/common/update.rs` around lines 611 - 658, do_update_vectors currently
converts point vectors via convert_point_vectors and forwards them into
UpdateVectorsOp and update() without the pre-WAL dimension validation used by
do_upsert_points; add the same early dimension check after convert_point_vectors
and before constructing CollectionUpdateOperations::VectorOperation so dimension
mismatches are rejected synchronously (i.e., call the shared validation helper
used by do_upsert_points or implement the same validation logic against the
collection schema), return the appropriate StorageError on mismatch, and add a
regression test exercising /points/vectors with wait=false to ensure
bad-dimension updates are rejected before queuing.

tests/openapi/test_vector_dimension_validation.py (1)

34-105: ⚡ Quick win

Add a named-vector regression case.

The Rust change added separate validation paths for named vectors, but this module only exercises default-vector payloads. One named-vector failure case here would keep that new branch from drifting untested.

🤖 Prompt for AI Agents

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

In `@tests/openapi/test_vector_dimension_validation.py` around lines 34 - 105, Add
a new test similar to test_sync_upsert_wrong_dimension_rejected but exercising
the named-vector payload branch: create
test_named_vector_wrong_dimension_rejected(collection_name) that sends a PUT to
'/collections/{collection_name}/points' with query_params {'wait':'true'} and a
body where the point contains a named vector (e.g. "vector": {"name": "myvec",
"values": [0.1,0.2,0.3]} into a 4-dim collection) and assert response.ok is
False, response.status_code == 400, and that "dimension" appears in
response.json()["status"]["error"].lower(); this ensures the named-vector
validation path is covered alongside existing tests (see
test_sync_upsert_wrong_dimension_rejected and
test_async_batch_upsert_wrong_dimension_rejected for payload structure and
assertions).