pymongo_search_utils.drop_vector_search_index waits on the wrong condition. It never succeeds on a collection that holds more than one search index.

Cause

collection.drop_search_index(index_name)
if wait_until_complete:
    wait_for_predicate(
        predicate=lambda: len(list(collection.list_search_indexes())) == 0,
        err=f"Index {index_name} did not drop in {wait_until_complete}!",
        timeout=wait_until_complete,
    )

The predicate means "the collection has zero search indexes". It should mean "index_name is gone".

Effect

Any caller that drops one index from a collection that keeps another index waits the full timeout and then raises, even though the drop succeeded. The error message names the dropped index, so it reports a failure that did not happen.

Found in langchain-mongodb. tests/integration_tests/test_retrievers.py::test_hybrid_retriever_auto_create_index drops text_index on a collection that keeps vector_index. The test waited 60s and failed. Because the drop did take effect, the next run found no text_index, skipped the drop, and passed. The test therefore alternated between passing and failing, and each failure forced a full-suite retry in CI.

Suggested fix

Wait for the named index only:

predicate=lambda: not any(
    ix["name"] == index_name for ix in collection.list_search_indexes()
)

Consider whether the function name is still right. It drops any search index, not only a vector search index.

Related

INTPYTHON-1103 works around this in the langchain-mongodb test suite with a local helper, drop_search_index_and_wait. See https://github.com/langchain-ai/langchain-mongodb/pull/461. That helper should be deleted once this is fixed.