chore(tests): fix responses and vector_io tests (#3119)

Some fixes to MCP tests. And a bunch of fixes for Vector providers.

I also enabled a bunch of Vector IO tests to be used with
`LlamaStackLibraryClient`

## Test Plan

Run Responses tests with llama stack library client:
```
pytest -s -v tests/integration/non_ci/responses/ --stack-config=server:starter \
   --text-model openai/gpt-4o \
  --embedding-model=sentence-transformers/all-MiniLM-L6-v2 \
  -k "client_with_models"
```

Do the same with `-k openai_client`

The rest should be taken care of by CI.

llama_stack/core/build.py

@@ -91,7 +91,7 @@ def get_provider_dependencies(
 
 
 def print_pip_install_help(config: BuildConfig):
-    normal_deps, special_deps = get_provider_dependencies(config)
+    normal_deps, special_deps, _ = get_provider_dependencies(config)
 
     cprint(
         f"Please install needed dependencies using the following commands:\n\nuv pip install {' '.join(normal_deps)}",

llama_stack/core/routers/inference.py

@@ -65,7 +65,7 @@ from llama_stack.providers.datatypes import HealthResponse, HealthStatus, Routin
 from llama_stack.providers.utils.inference.inference_store import InferenceStore
 from llama_stack.providers.utils.telemetry.tracing import get_current_span
 
-logger = get_logger(name=__name__, category="core")
+logger = get_logger(name=__name__, category="inference")
 
 
 class InferenceRouter(Inference):
@@ -854,4 +854,5 @@ class InferenceRouter(Inference):
                     model=model.identifier,
                     object="chat.completion",
                 )
+                logger.debug(f"InferenceRouter.completion_response: {final_response}")
                 await self.store.store_chat_completion(final_response, messages)

llama_stack/core/routing_tables/models.py

@@ -63,6 +63,8 @@ class ModelsRoutingTable(CommonRoutingTableImpl, Models):
 
     async def get_provider_impl(self, model_id: str) -> Any:
         model = await lookup_model(self, model_id)
+        if model.provider_id not in self.impls_by_provider_id:
+            raise ValueError(f"Provider {model.provider_id} not found in the routing table")
         return self.impls_by_provider_id[model.provider_id]
 
     async def register_model(

llama_stack/log.py

@@ -32,6 +32,7 @@ CATEGORIES = [
     "tools",
     "client",
     "telemetry",
+    "openai_responses",
 ]
 
 # Initialize category levels with default level

llama_stack/models/llama/llama3/chat_format.py

@@ -236,6 +236,7 @@ class ChatFormat:
                     arguments_json=json.dumps(tool_arguments),
                 )
             )
+            content = ""
 
         return RawMessage(
             role="assistant",

llama_stack/providers/inline/agents/meta_reference/openai_responses.py

@@ -488,8 +488,12 @@ class OpenAIResponsesImpl:
             # Convert collected chunks to complete response
             if chat_response_tool_calls:
                 tool_calls = [chat_response_tool_calls[i] for i in sorted(chat_response_tool_calls.keys())]
+
+                # when there are tool calls, we need to clear the content
+                chat_response_content = []
             else:
                 tool_calls = None
+
             assistant_message = OpenAIAssistantMessageParam(
                 content="".join(chat_response_content),
                 tool_calls=tool_calls,

llama_stack/providers/inline/vector_io/faiss/faiss.py

@@ -33,6 +33,7 @@ from llama_stack.providers.utils.kvstore import kvstore_impl
 from llama_stack.providers.utils.kvstore.api import KVStore
 from llama_stack.providers.utils.memory.openai_vector_store_mixin import OpenAIVectorStoreMixin
 from llama_stack.providers.utils.memory.vector_store import (
+    ChunkForDeletion,
     EmbeddingIndex,
     VectorDBWithIndex,
 )
@@ -128,11 +129,12 @@ class FaissIndex(EmbeddingIndex):
         # Save updated index
         await self._save_index()
 
-    async def delete_chunk(self, chunk_id: str) -> None:
-        if chunk_id not in self.chunk_ids:
+    async def delete_chunks(self, chunks_for_deletion: list[ChunkForDeletion]) -> None:
+        chunk_ids = [c.chunk_id for c in chunks_for_deletion]
+        if not set(chunk_ids).issubset(self.chunk_ids):
             return
 
-        async with self.chunk_id_lock:
+        def remove_chunk(chunk_id: str):
             index = self.chunk_ids.index(chunk_id)
             self.index.remove_ids(np.array([index]))
 
@@ -146,6 +148,10 @@ class FaissIndex(EmbeddingIndex):
             self.chunk_by_index = new_chunk_by_index
             self.chunk_ids.pop(index)
 
+        async with self.chunk_id_lock:
+            for chunk_id in chunk_ids:
+                remove_chunk(chunk_id)
+
         await self._save_index()
 
     async def query_vector(
@@ -297,8 +303,7 @@ class FaissVectorIOAdapter(OpenAIVectorStoreMixin, VectorIO, VectorDBsProtocolPr
 
         return await index.query_chunks(query, params)
 
-    async def delete_chunks(self, store_id: str, chunk_ids: list[str]) -> None:
-        """Delete a chunk from a faiss index"""
+    async def delete_chunks(self, store_id: str, chunks_for_deletion: list[ChunkForDeletion]) -> None:
+        """Delete chunks from a faiss index"""
         faiss_index = self.cache[store_id].index
-        for chunk_id in chunk_ids:
-            await faiss_index.delete_chunk(chunk_id)
+        await faiss_index.delete_chunks(chunks_for_deletion)

llama_stack/providers/inline/vector_io/sqlite_vec/sqlite_vec.py

@@ -31,6 +31,7 @@ from llama_stack.providers.utils.memory.openai_vector_store_mixin import OpenAIV
 from llama_stack.providers.utils.memory.vector_store import (
     RERANKER_TYPE_RRF,
     RERANKER_TYPE_WEIGHTED,
+    ChunkForDeletion,
     EmbeddingIndex,
     VectorDBWithIndex,
 )
@@ -426,34 +427,36 @@ class SQLiteVecIndex(EmbeddingIndex):
 
         return QueryChunksResponse(chunks=chunks, scores=scores)
 
-    async def delete_chunk(self, chunk_id: str) -> None:
+    async def delete_chunks(self, chunks_for_deletion: list[ChunkForDeletion]) -> None:
         """Remove a chunk from the SQLite vector store."""
+        chunk_ids = [c.chunk_id for c in chunks_for_deletion]
 
-        def _delete_chunk():
+        def _delete_chunks():
             connection = _create_sqlite_connection(self.db_path)
             cur = connection.cursor()
             try:
                 cur.execute("BEGIN TRANSACTION")
 
                 # Delete from metadata table
-                cur.execute(f"DELETE FROM {self.metadata_table} WHERE id = ?", (chunk_id,))
+                placeholders = ",".join("?" * len(chunk_ids))
+                cur.execute(f"DELETE FROM {self.metadata_table} WHERE id IN ({placeholders})", chunk_ids)
 
                 # Delete from vector table
-                cur.execute(f"DELETE FROM {self.vector_table} WHERE id = ?", (chunk_id,))
+                cur.execute(f"DELETE FROM {self.vector_table} WHERE id IN ({placeholders})", chunk_ids)
 
                 # Delete from FTS table
-                cur.execute(f"DELETE FROM {self.fts_table} WHERE id = ?", (chunk_id,))
+                cur.execute(f"DELETE FROM {self.fts_table} WHERE id IN ({placeholders})", chunk_ids)
 
                 connection.commit()
             except Exception as e:
                 connection.rollback()
-                logger.error(f"Error deleting chunk {chunk_id}: {e}")
+                logger.error(f"Error deleting chunks: {e}")
                 raise
             finally:
                 cur.close()
                 connection.close()
 
-        await asyncio.to_thread(_delete_chunk)
+        await asyncio.to_thread(_delete_chunks)
 
 
 class SQLiteVecVectorIOAdapter(OpenAIVectorStoreMixin, VectorIO, VectorDBsProtocolPrivate):
@@ -551,12 +554,10 @@ class SQLiteVecVectorIOAdapter(OpenAIVectorStoreMixin, VectorIO, VectorDBsProtoc
             raise VectorStoreNotFoundError(vector_db_id)
         return await index.query_chunks(query, params)
 
-    async def delete_chunks(self, store_id: str, chunk_ids: list[str]) -> None:
-        """Delete a chunk from a sqlite_vec index."""
+    async def delete_chunks(self, store_id: str, chunks_for_deletion: list[ChunkForDeletion]) -> None:
+        """Delete chunks from a sqlite_vec index."""
         index = await self._get_and_cache_vector_db_index(store_id)
         if not index:
             raise VectorStoreNotFoundError(store_id)
 
-        for chunk_id in chunk_ids:
-            # Use the index's delete_chunk method
-            await index.index.delete_chunk(chunk_id)
+        await index.index.delete_chunks(chunks_for_deletion)

llama_stack/providers/registry/vector_io.py

@@ -342,6 +342,7 @@ See [Chroma's documentation](https://docs.trychroma.com/docs/overview/introducti
 """,
             ),
             api_dependencies=[Api.inference],
+            optional_api_dependencies=[Api.files],
         ),
         InlineProviderSpec(
             api=Api.vector_io,
@@ -350,6 +351,7 @@ See [Chroma's documentation](https://docs.trychroma.com/docs/overview/introducti
             module="llama_stack.providers.inline.vector_io.chroma",
             config_class="llama_stack.providers.inline.vector_io.chroma.ChromaVectorIOConfig",
             api_dependencies=[Api.inference],
+            optional_api_dependencies=[Api.files],
             description="""
 [Chroma](https://www.trychroma.com/) is an inline and remote vector
 database provider for Llama Stack. It allows you to store and query vectors directly within a Chroma database.
@@ -464,6 +466,7 @@ See [Weaviate's documentation](https://weaviate.io/developers/weaviate) for more
 """,
             ),
             api_dependencies=[Api.inference],
+            optional_api_dependencies=[Api.files],
         ),
         InlineProviderSpec(
             api=Api.vector_io,
@@ -731,6 +734,7 @@ For more details on TLS configuration, refer to the [TLS setup guide](https://mi
 """,
             ),
             api_dependencies=[Api.inference],
+            optional_api_dependencies=[Api.files],
         ),
         InlineProviderSpec(
             api=Api.vector_io,

llama_stack/providers/remote/inference/fireworks/fireworks.py

@@ -235,6 +235,7 @@ class FireworksInferenceAdapter(ModelRegistryHelper, Inference, NeedsRequestProv
 
         llama_model = self.get_llama_model(request.model)
         if isinstance(request, ChatCompletionRequest):
+            # TODO: tools are never added to the request, so we need to add them here
             if media_present or not llama_model:
                 input_dict["messages"] = [
                     await convert_message_to_openai_dict(m, download=True) for m in request.messages
@@ -378,6 +379,7 @@ class FireworksInferenceAdapter(ModelRegistryHelper, Inference, NeedsRequestProv
         # Fireworks chat completions OpenAI-compatible API does not support
         # tool calls properly.
         llama_model = self.get_llama_model(model_obj.provider_resource_id)
+
         if llama_model:
             return await OpenAIChatCompletionToLlamaStackMixin.openai_chat_completion(
                 self,
@@ -431,4 +433,5 @@ class FireworksInferenceAdapter(ModelRegistryHelper, Inference, NeedsRequestProv
             user=user,
         )
 
+        logger.debug(f"fireworks params: {params}")
         return await self._get_openai_client().chat.completions.create(model=model_obj.provider_resource_id, **params)

llama_stack/providers/remote/vector_io/chroma/chroma.py

@@ -26,6 +26,7 @@ from llama_stack.providers.utils.kvstore import kvstore_impl
 from llama_stack.providers.utils.kvstore.api import KVStore
 from llama_stack.providers.utils.memory.openai_vector_store_mixin import OpenAIVectorStoreMixin
 from llama_stack.providers.utils.memory.vector_store import (
+    ChunkForDeletion,
     EmbeddingIndex,
     VectorDBWithIndex,
 )
@@ -115,8 +116,10 @@ class ChromaIndex(EmbeddingIndex):
     ) -> QueryChunksResponse:
         raise NotImplementedError("Keyword search is not supported in Chroma")
 
-    async def delete_chunk(self, chunk_id: str) -> None:
-        raise NotImplementedError("delete_chunk is not supported in Chroma")
+    async def delete_chunks(self, chunks_for_deletion: list[ChunkForDeletion]) -> None:
+        """Delete a single chunk from the Chroma collection by its ID."""
+        ids = [f"{chunk.document_id}:{chunk.chunk_id}" for chunk in chunks_for_deletion]
+        await maybe_await(self.collection.delete(ids=ids))
 
     async def query_hybrid(
         self,
@@ -144,6 +147,7 @@ class ChromaVectorIOAdapter(OpenAIVectorStoreMixin, VectorIO, VectorDBsProtocolP
         self.cache = {}
         self.kvstore: KVStore | None = None
         self.vector_db_store = None
+        self.files_api = files_api
 
     async def initialize(self) -> None:
         self.kvstore = await kvstore_impl(self.config.kvstore)
@@ -227,5 +231,10 @@ class ChromaVectorIOAdapter(OpenAIVectorStoreMixin, VectorIO, VectorDBsProtocolP
         self.cache[vector_db_id] = index
         return index
 
-    async def delete_chunks(self, store_id: str, chunk_ids: list[str]) -> None:
-        raise NotImplementedError("OpenAI Vector Stores API is not supported in Chroma")
+    async def delete_chunks(self, store_id: str, chunks_for_deletion: list[ChunkForDeletion]) -> None:
+        """Delete chunks from a Chroma vector store."""
+        index = await self._get_and_cache_vector_db_index(store_id)
+        if not index:
+            raise ValueError(f"Vector DB {store_id} not found")
+
+        await index.index.delete_chunks(chunks_for_deletion)

llama_stack/providers/remote/vector_io/milvus/milvus.py

@@ -28,6 +28,7 @@ from llama_stack.providers.utils.kvstore.api import KVStore
 from llama_stack.providers.utils.memory.openai_vector_store_mixin import OpenAIVectorStoreMixin
 from llama_stack.providers.utils.memory.vector_store import (
     RERANKER_TYPE_WEIGHTED,
+    ChunkForDeletion,
     EmbeddingIndex,
     VectorDBWithIndex,
 )
@@ -287,14 +288,17 @@ class MilvusIndex(EmbeddingIndex):
 
         return QueryChunksResponse(chunks=filtered_chunks, scores=filtered_scores)
 
-    async def delete_chunk(self, chunk_id: str) -> None:
+    async def delete_chunks(self, chunks_for_deletion: list[ChunkForDeletion]) -> None:
         """Remove a chunk from the Milvus collection."""
+        chunk_ids = [c.chunk_id for c in chunks_for_deletion]
         try:
+            # Use IN clause with square brackets and single quotes for VARCHAR field
+            chunk_ids_str = ", ".join(f"'{chunk_id}'" for chunk_id in chunk_ids)
             await asyncio.to_thread(
-                self.client.delete, collection_name=self.collection_name, filter=f'chunk_id == "{chunk_id}"'
+                self.client.delete, collection_name=self.collection_name, filter=f"chunk_id in [{chunk_ids_str}]"
             )
         except Exception as e:
-            logger.error(f"Error deleting chunk {chunk_id} from Milvus collection {self.collection_name}: {e}")
+            logger.error(f"Error deleting chunks from Milvus collection {self.collection_name}: {e}")
             raise
 
 
@@ -420,12 +424,10 @@ class MilvusVectorIOAdapter(OpenAIVectorStoreMixin, VectorIO, VectorDBsProtocolP
 
         return await index.query_chunks(query, params)
 
-    async def delete_chunks(self, store_id: str, chunk_ids: list[str]) -> None:
+    async def delete_chunks(self, store_id: str, chunks_for_deletion: list[ChunkForDeletion]) -> None:
         """Delete a chunk from a milvus vector store."""
         index = await self._get_and_cache_vector_db_index(store_id)
         if not index:
             raise VectorStoreNotFoundError(store_id)
 
-        for chunk_id in chunk_ids:
-            # Use the index's delete_chunk method
-            await index.index.delete_chunk(chunk_id)
+        await index.index.delete_chunks(chunks_for_deletion)

llama_stack/providers/remote/vector_io/pgvector/pgvector.py

@@ -27,6 +27,7 @@ from llama_stack.providers.utils.kvstore import kvstore_impl
 from llama_stack.providers.utils.kvstore.api import KVStore
 from llama_stack.providers.utils.memory.openai_vector_store_mixin import OpenAIVectorStoreMixin
 from llama_stack.providers.utils.memory.vector_store import (
+    ChunkForDeletion,
     EmbeddingIndex,
     VectorDBWithIndex,
 )
@@ -163,10 +164,11 @@ class PGVectorIndex(EmbeddingIndex):
         with self.conn.cursor(cursor_factory=psycopg2.extras.DictCursor) as cur:
             cur.execute(f"DROP TABLE IF EXISTS {self.table_name}")
 
-    async def delete_chunk(self, chunk_id: str) -> None:
+    async def delete_chunks(self, chunks_for_deletion: list[ChunkForDeletion]) -> None:
         """Remove a chunk from the PostgreSQL table."""
+        chunk_ids = [c.chunk_id for c in chunks_for_deletion]
         with self.conn.cursor(cursor_factory=psycopg2.extras.DictCursor) as cur:
-            cur.execute(f"DELETE FROM {self.table_name} WHERE id = %s", (chunk_id,))
+            cur.execute(f"DELETE FROM {self.table_name} WHERE id = ANY(%s)", (chunk_ids,))
 
 
 class PGVectorVectorIOAdapter(OpenAIVectorStoreMixin, VectorIO, VectorDBsProtocolPrivate):
@@ -275,12 +277,10 @@ class PGVectorVectorIOAdapter(OpenAIVectorStoreMixin, VectorIO, VectorDBsProtoco
         self.cache[vector_db_id] = VectorDBWithIndex(vector_db, index, self.inference_api)
         return self.cache[vector_db_id]
 
-    async def delete_chunks(self, store_id: str, chunk_ids: list[str]) -> None:
+    async def delete_chunks(self, store_id: str, chunks_for_deletion: list[ChunkForDeletion]) -> None:
         """Delete a chunk from a PostgreSQL vector store."""
         index = await self._get_and_cache_vector_db_index(store_id)
         if not index:
             raise VectorStoreNotFoundError(store_id)
 
-        for chunk_id in chunk_ids:
-            # Use the index's delete_chunk method
-            await index.index.delete_chunk(chunk_id)
+        await index.index.delete_chunks(chunks_for_deletion)

llama_stack/providers/remote/vector_io/qdrant/qdrant.py

@@ -29,6 +29,7 @@ from llama_stack.providers.inline.vector_io.qdrant import QdrantVectorIOConfig a
 from llama_stack.providers.utils.kvstore import KVStore, kvstore_impl
 from llama_stack.providers.utils.memory.openai_vector_store_mixin import OpenAIVectorStoreMixin
 from llama_stack.providers.utils.memory.vector_store import (
+    ChunkForDeletion,
     EmbeddingIndex,
     VectorDBWithIndex,
 )
@@ -88,15 +89,16 @@ class QdrantIndex(EmbeddingIndex):
 
         await self.client.upsert(collection_name=self.collection_name, points=points)
 
-    async def delete_chunk(self, chunk_id: str) -> None:
+    async def delete_chunks(self, chunks_for_deletion: list[ChunkForDeletion]) -> None:
         """Remove a chunk from the Qdrant collection."""
+        chunk_ids = [convert_id(c.chunk_id) for c in chunks_for_deletion]
         try:
             await self.client.delete(
                 collection_name=self.collection_name,
-                points_selector=models.PointIdsList(points=[convert_id(chunk_id)]),
+                points_selector=models.PointIdsList(points=chunk_ids),
             )
         except Exception as e:
-            log.error(f"Error deleting chunk {chunk_id} from Qdrant collection {self.collection_name}: {e}")
+            log.error(f"Error deleting chunks from Qdrant collection {self.collection_name}: {e}")
             raise
 
     async def query_vector(self, embedding: NDArray, k: int, score_threshold: float) -> QueryChunksResponse:
@@ -264,12 +266,14 @@ class QdrantVectorIOAdapter(OpenAIVectorStoreMixin, VectorIO, VectorDBsProtocolP
     ) -> VectorStoreFileObject:
         # Qdrant doesn't allow multiple clients to access the same storage path simultaneously.
         async with self._qdrant_lock:
-            await super().openai_attach_file_to_vector_store(vector_store_id, file_id, attributes, chunking_strategy)
+            return await super().openai_attach_file_to_vector_store(
+                vector_store_id, file_id, attributes, chunking_strategy
+            )
 
-    async def delete_chunks(self, store_id: str, chunk_ids: list[str]) -> None:
+    async def delete_chunks(self, store_id: str, chunks_for_deletion: list[ChunkForDeletion]) -> None:
         """Delete chunks from a Qdrant vector store."""
         index = await self._get_and_cache_vector_db_index(store_id)
         if not index:
             raise ValueError(f"Vector DB {store_id} not found")
-        for chunk_id in chunk_ids:
-            await index.index.delete_chunk(chunk_id)
+
+        await index.index.delete_chunks(chunks_for_deletion)

llama_stack/providers/remote/vector_io/weaviate/weaviate.py

@@ -26,6 +26,7 @@ from llama_stack.providers.utils.memory.openai_vector_store_mixin import (
     OpenAIVectorStoreMixin,
 )
 from llama_stack.providers.utils.memory.vector_store import (
+    ChunkForDeletion,
     EmbeddingIndex,
     VectorDBWithIndex,
 )
@@ -67,6 +68,7 @@ class WeaviateIndex(EmbeddingIndex):
             data_objects.append(
                 wvc.data.DataObject(
                     properties={
+                        "chunk_id": chunk.chunk_id,
                         "chunk_content": chunk.model_dump_json(),
                     },
                     vector=embeddings[i].tolist(),
@@ -79,10 +81,11 @@ class WeaviateIndex(EmbeddingIndex):
         # TODO: make this async friendly
         collection.data.insert_many(data_objects)
 
-    async def delete_chunk(self, chunk_id: str) -> None:
+    async def delete_chunks(self, chunks_for_deletion: list[ChunkForDeletion]) -> None:
         sanitized_collection_name = sanitize_collection_name(self.collection_name, weaviate_format=True)
         collection = self.client.collections.get(sanitized_collection_name)
-        collection.data.delete_many(where=Filter.by_property("id").contains_any([chunk_id]))
+        chunk_ids = [chunk.chunk_id for chunk in chunks_for_deletion]
+        collection.data.delete_many(where=Filter.by_property("chunk_id").contains_any(chunk_ids))
 
     async def query_vector(self, embedding: NDArray, k: int, score_threshold: float) -> QueryChunksResponse:
         sanitized_collection_name = sanitize_collection_name(self.collection_name, weaviate_format=True)
@@ -307,10 +310,10 @@ class WeaviateVectorIOAdapter(
 
         return await index.query_chunks(query, params)
 
-    async def delete_chunks(self, store_id: str, chunk_ids: list[str]) -> None:
+    async def delete_chunks(self, store_id: str, chunks_for_deletion: list[ChunkForDeletion]) -> None:
         sanitized_collection_name = sanitize_collection_name(store_id, weaviate_format=True)
         index = await self._get_and_cache_vector_db_index(sanitized_collection_name)
         if not index:
             raise ValueError(f"Vector DB {sanitized_collection_name} not found")
 
-        await index.delete(chunk_ids)
+        await index.index.delete_chunks(chunks_for_deletion)

llama_stack/providers/utils/memory/openai_vector_store_mixin.py

@@ -6,7 +6,6 @@
 
 import asyncio
 import json
-import logging
 import mimetypes
 import time
 import uuid
@@ -37,10 +36,15 @@ from llama_stack.apis.vector_io import (
     VectorStoreSearchResponse,
     VectorStoreSearchResponsePage,
 )
+from llama_stack.log import get_logger
 from llama_stack.providers.utils.kvstore.api import KVStore
-from llama_stack.providers.utils.memory.vector_store import content_from_data_and_mime_type, make_overlapped_chunks
+from llama_stack.providers.utils.memory.vector_store import (
+    ChunkForDeletion,
+    content_from_data_and_mime_type,
+    make_overlapped_chunks,
+)
 
-logger = logging.getLogger(__name__)
+logger = get_logger(__name__, category="vector_io")
 
 # Constants for OpenAI vector stores
 CHUNK_MULTIPLIER = 5
@@ -154,8 +158,8 @@ class OpenAIVectorStoreMixin(ABC):
         self.openai_vector_stores = await self._load_openai_vector_stores()
 
     @abstractmethod
-    async def delete_chunks(self, store_id: str, chunk_ids: list[str]) -> None:
-        """Delete a chunk from a vector store."""
+    async def delete_chunks(self, store_id: str, chunks_for_deletion: list[ChunkForDeletion]) -> None:
+        """Delete chunks from a vector store."""
         pass
 
     @abstractmethod
@@ -614,7 +618,7 @@ class OpenAIVectorStoreMixin(ABC):
                 )
                 vector_store_file_object.status = "completed"
         except Exception as e:
-            logger.error(f"Error attaching file to vector store: {e}")
+            logger.exception("Error attaching file to vector store")
             vector_store_file_object.status = "failed"
             vector_store_file_object.last_error = VectorStoreFileLastError(
                 code="server_error",
@@ -767,7 +771,21 @@ class OpenAIVectorStoreMixin(ABC):
 
         dict_chunks = await self._load_openai_vector_store_file_contents(vector_store_id, file_id)
         chunks = [Chunk.model_validate(c) for c in dict_chunks]
-        await self.delete_chunks(vector_store_id, [str(c.chunk_id) for c in chunks if c.chunk_id])
+
+        # Create ChunkForDeletion objects with both chunk_id and document_id
+        chunks_for_deletion = []
+        for c in chunks:
+            if c.chunk_id:
+                document_id = c.metadata.get("document_id") or (
+                    c.chunk_metadata.document_id if c.chunk_metadata else None
+                )
+                if document_id:
+                    chunks_for_deletion.append(ChunkForDeletion(chunk_id=str(c.chunk_id), document_id=document_id))
+                else:
+                    logger.warning(f"Chunk {c.chunk_id} has no document_id, skipping deletion")
+
+        if chunks_for_deletion:
+            await self.delete_chunks(vector_store_id, chunks_for_deletion)
 
         store_info = self.openai_vector_stores[vector_store_id].copy()
 

llama_stack/providers/utils/memory/vector_store.py

@@ -16,6 +16,7 @@ from urllib.parse import unquote
 import httpx
 import numpy as np
 from numpy.typing import NDArray
+from pydantic import BaseModel
 
 from llama_stack.apis.common.content_types import (
     URL,
@@ -34,6 +35,18 @@ from llama_stack.providers.utils.vector_io.vector_utils import generate_chunk_id
 
 log = logging.getLogger(__name__)
 
+
+class ChunkForDeletion(BaseModel):
+    """Information needed to delete a chunk from a vector store.
+
+    :param chunk_id: The ID of the chunk to delete
+    :param document_id: The ID of the document this chunk belongs to
+    """
+
+    chunk_id: str
+    document_id: str
+
+
 # Constants for reranker types
 RERANKER_TYPE_RRF = "rrf"
 RERANKER_TYPE_WEIGHTED = "weighted"
@@ -232,7 +245,7 @@ class EmbeddingIndex(ABC):
         raise NotImplementedError()
 
     @abstractmethod
-    async def delete_chunk(self, chunk_id: str):
+    async def delete_chunks(self, chunks_for_deletion: list[ChunkForDeletion]):
         raise NotImplementedError()
 
     @abstractmethod