[Bug]: HNSW - Cannot return the results in a contigious 2D array. Probably ef or M is too small

[tazarov]

What happened?

Testing a tangential issue, I've run into a reproducible HNSW index error:

...
  File "/Users/tazarov/experiments/chroma/chroma-taz-21/chromadb/segment/impl/vector/local_hnsw.py", line 157, in query_vectors
    result_labels, distances = self._index.knn_query(
                               ^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: Cannot return the results in a contigious 2D array. Probably ef or M is too small

The code to reproduce this:

import gc
import uuid

import chromadb
import numpy as np

np.random.seed(42)


def main():
    data = np.random.uniform(-1, 1, (1000, 500, 1536))

    client = chromadb.PersistentClient("contiguous2d")
    # client = chromadb.HttpClient()
    collection = client.get_or_create_collection("test_collection")
    for i in range(data.shape[0]):
        print("Iteration: ", str(i))
        gc.collect()
        ids = [f"{uuid.uuid4()}" for i in range(data[i].shape[0])]
        collection.add(ids=ids, embeddings=data[i])
        collection.query(query_embeddings=[data[i][np.random.choice(data[i].shape[0])].tolist()], n_results=10)
        collection.delete(ids=ids)
        gc.collect()


if __name__ == "__main__":
    main()

Note: The issue occurs anywhere between 20 to 50th iterations.

The issue can be reproduced with both PersistentClient and HttpClient with server in Docker container.

Versions

Chroma 0.5.3/latest main, hnswlib 0.7.3/0.7.5

Tested HW configs:

• Python 3.11, MacOS (M3 CPU)
• Python 3.12, Ubuntu 24.04 (4 core/16GB, Intel CPU)

Relevant log output

Traceback (most recent call last):
  File "/home/ubuntu/hnsw_contiguous.py", line 22, in <module>
    main()
  File "/home/ubuntu/hnsw_contiguous.py", line 17, in main
    collection.query(query_embeddings=[data[i][np.random.choice(data[i].shape[0])].tolist()], n_results=10)
  File "/home/ubuntu/venv/lib/python3.12/site-packages/chromadb/api/models/Collection.py", line 195, in query
    query_results = self._client._query(
                    ^^^^^^^^^^^^^^^^^^^^
  File "/home/ubuntu/venv/lib/python3.12/site-packages/chromadb/telemetry/opentelemetry/__init__.py", line 146, in wrapper
    return f(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^
  File "/home/ubuntu/venv/lib/python3.12/site-packages/chromadb/rate_limiting/__init__.py", line 47, in wrapper
    return f(self, *args, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/ubuntu/venv/lib/python3.12/site-packages/chromadb/api/segment.py", line 738, in _query
    results = vector_reader.query_vectors(query)
              ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/ubuntu/venv/lib/python3.12/site-packages/chromadb/telemetry/opentelemetry/__init__.py", line 146, in wrapper
    return f(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^
  File "/home/ubuntu/venv/lib/python3.12/site-packages/chromadb/segment/impl/vector/local_persistent_hnsw.py", line 372, in query_vectors
    hnsw_results = super().query_vectors(hnsw_query)
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/home/ubuntu/venv/lib/python3.12/site-packages/chromadb/telemetry/opentelemetry/__init__.py", line 146, in wrapper
    return f(*args, **kwargs)
           ^^^^^^^^^^^^^^^^^^
  File "/home/ubuntu/venv/lib/python3.12/site-packages/chromadb/segment/impl/vector/local_hnsw.py", line 156, in query_vectors
    result_labels, distances = self._index.knn_query(
                               ^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: Cannot return the results in a contigious 2D array. Probably ef or M is too small