[Redis] How to filter by key instead of metadata/field?

[ai-learner-00]

When using redis-py directly, we can use the limit_ids method, which will use the INKEYS redis command. For very large filters (all the chunk ids of multiple documents), INKEYS is more efficient than filtering based on fields. Is there a way to do it in Langchain when creating a vector store retriever based on Redis?

search_query = f'(*)=>[KNN {top_k} @embedding_vector $vector AS distance]'
query_vector = generate_vector(1000)
query_params = {
    "vector": query_vector.tobytes()
}
query = Query(search_query).return_fields("chunk_id", "model_name").sort_by("distance").dialect(2).limit_ids(
    *[f"chunk:{chunk.chunk_id}" for chunk in chunks]
)

test-max-filter-size.zip
(in my tests, I went up to 500 000 chunk ids and it times out when using a filter over INKEYS)