Merge pull request #1262 from Legion345/fix/stop-hook-crash

fix(storage): stop ChromaDB from crashing when reopening an existing …

mempalace/backends/chroma.py

@@ -8,6 +8,7 @@ from pathlib import Path
 from typing import Any, Optional
 
 import chromadb
+from chromadb.errors import NotFoundError as _ChromaNotFoundError
 
 from .base import (
     BaseBackend,
@@ -1093,15 +1094,18 @@ class ChromaBackend(BaseBackend):
         ef_kwargs = {"embedding_function": ef} if ef is not None else {}
 
         if create:
-            collection = client.get_or_create_collection(
+            try:
-                collection_name,
+                collection = client.get_collection(collection_name, **ef_kwargs)
-                metadata={
+            except _ChromaNotFoundError:
-                    "hnsw:space": hnsw_space,
+                collection = client.create_collection(
-                    "hnsw:num_threads": 1,
+                    collection_name,
-                    **_HNSW_BLOAT_GUARD,
+                    metadata={
-                },
+                        "hnsw:space": hnsw_space,
-                **ef_kwargs,
+                        "hnsw:num_threads": 1,
-            )
+                        **_HNSW_BLOAT_GUARD,
+                    },
+                    **ef_kwargs,
+                )
         else:
             collection = client.get_collection(collection_name, **ef_kwargs)
         _pin_hnsw_threads(collection)

tests/test_backends.py

@@ -372,6 +372,32 @@ def test_chroma_backend_create_collection_sets_hnsw_bloat_guard(tmp_path):
     assert col.metadata.get("hnsw:sync_threshold") == 50_000
 
 
+def test_get_collection_create_true_is_idempotent(tmp_path):
+    """Calling get_collection(create=True) twice on the same name must not crash.
+
+    ChromaDB 1.5.x's Rust bindings SIGSEGV when get_or_create_collection is
+    called with metadata that differs from the stored collection metadata. The
+    fix splits the call into get_collection -> fallback create_collection so the
+    metadata-comparison codepath in chromadb_rust_bindings is never reached for
+    existing collections. Regression guard for issue #1089.
+    """
+    palace = str(tmp_path / "palace")
+    backend = ChromaBackend()
+    backend.get_collection(palace, collection_name="mempalace_drawers", create=True)
+    col2 = backend.get_collection(palace, collection_name="mempalace_drawers", create=True)
+    assert isinstance(col2, ChromaCollection)
+
+
+def test_get_collection_create_true_preserves_existing_metadata(tmp_path):
+    """Existing collection metadata is not overwritten when reopened with create=True."""
+    palace = str(tmp_path / "palace")
+    backend = ChromaBackend()
+    backend.get_collection(palace, collection_name="mempalace_drawers", create=True)
+    col = backend.get_collection(palace, collection_name="mempalace_drawers", create=True)
+    assert col._collection.metadata["hnsw:space"] == "cosine"
+    assert col._collection.metadata.get("hnsw:batch_size") == 50_000
+
+
 def test_fix_blob_seq_ids_converts_blobs_to_integers(tmp_path):
     """Simulate a ChromaDB 0.6.x database with BLOB seq_ids and verify repair."""
     db_path = tmp_path / "chroma.sqlite3"