phoenix-oss/llama-stack-mirror
0
0
Fork 1
mirror of https://github.com/meta-llama/llama-stack.git synced 2025-07-13 08:36:09 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

chore: Adding unit tests for OpenAI vector stores and migrating SQLite-vec registry to kvstore (#2665)

Browse source 
# What does this PR do?

This PR refactors and the VectorIO backend logic for `sqlite-vec` and
adds unit tests and fixtures to make it easy to test both `sqlite-vec`
and `milvus`.

Key changes:
- `sqlite-vec` migrated to `kvstore` registry
- added in-memory cache for sqlite-vec to be consistent with `milvus`
- default fixtures moved to `conftest.py` 
- removed redundant tests from sqlite`-vec`
- made `test_vector_io_openai_vector_stores.py` more easily extensible 


## Test Plan
Unit tests added testing inline providers.

---------

Signed-off-by: Francisco Javier Arceo <farceo@redhat.com>
This commit is contained in:
Francisco Arceo 2025-07-10 14:22:13 -04:00 • committed by GitHub
parent b18f4d1ccf
commit 6a6b66ae4f
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
12 changed files with 422 additions and 424 deletions
Download patch file Download diff file Expand all files Collapse all files

2
llama_stack/providers/inline/vector_io/milvus/config.py
View file

@ -18,7 +18,7 @@ from llama_stack.schema_utils import json_schema_type
@json_schema_type
class MilvusVectorIOConfig(BaseModel):
db_path: str
kvstore: KVStoreConfig
kvstore: KVStoreConfig = Field(description="Config for KV store backend (SQLite only for now)")
consistency_level: str = Field(description="The consistency level of the Milvus server", default="Strong")
@classmethod

14
llama_stack/providers/inline/vector_io/sqlite_vec/config.py
View file

@ -6,14 +6,24 @@
from typing import Any
from pydantic import BaseModel
from pydantic import BaseModel, Field
from llama_stack.providers.utils.kvstore.config import (
KVStoreConfig,
SqliteKVStoreConfig,
)
class SQLiteVectorIOConfig(BaseModel):
db_path: str
db_path: str = Field(description="Path to the SQLite database file")
kvstore: KVStoreConfig = Field(description="Config for KV store backend (SQLite only for now)")
@classmethod
def sample_run_config(cls, __distro_dir__: str) -> dict[str, Any]:
return {
"db_path": "${env.SQLITE_STORE_DIR:=" + __distro_dir__ + "}/" + "sqlite_vec.db",
"kvstore": SqliteKVStoreConfig.sample_run_config(
__distro_dir__=__distro_dir__,
db_name="sqlite_vec_registry.db",
),
}

313
llama_stack/providers/inline/vector_io/sqlite_vec/sqlite_vec.py
View file

@ -24,6 +24,8 @@ from llama_stack.apis.vector_io import (
VectorIO,
)
from llama_stack.providers.datatypes import VectorDBsProtocolPrivate
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 (
RERANKER_TYPE_RRF,
@ -40,6 +42,13 @@ KEYWORD_SEARCH = "keyword"
HYBRID_SEARCH = "hybrid"
SEARCH_MODES = {VECTOR_SEARCH, KEYWORD_SEARCH, HYBRID_SEARCH}
VERSION = "v3"
VECTOR_DBS_PREFIX = f"vector_dbs:sqlite_vec:{VERSION}::"
VECTOR_INDEX_PREFIX = f"vector_index:sqlite_vec:{VERSION}::"
OPENAI_VECTOR_STORES_PREFIX = f"openai_vector_stores:sqlite_vec:{VERSION}::"
OPENAI_VECTOR_STORES_FILES_PREFIX = f"openai_vector_stores_files:sqlite_vec:{VERSION}::"
OPENAI_VECTOR_STORES_FILES_CONTENTS_PREFIX = f"openai_vector_stores_files_contents:sqlite_vec:{VERSION}::"
def serialize_vector(vector: list[float]) -> bytes:
"""Serialize a list of floats into a compact binary representation."""
@ -117,13 +126,14 @@ class SQLiteVecIndex(EmbeddingIndex):
- An FTS5 table (fts_chunks_{bank_id}) for full-text keyword search.
"""
def __init__(self, dimension: int, db_path: str, bank_id: str):
def __init__(self, dimension: int, db_path: str, bank_id: str, kvstore: KVStore | None = None):
self.dimension = dimension
self.db_path = db_path
self.bank_id = bank_id
self.metadata_table = f"chunks_{bank_id}".replace("-", "_")
self.vector_table = f"vec_chunks_{bank_id}".replace("-", "_")
self.fts_table = f"fts_chunks_{bank_id}".replace("-", "_")
self.kvstore = kvstore
@classmethod
async def create(cls, dimension: int, db_path: str, bank_id: str):
@ -425,27 +435,116 @@ class SQLiteVecVectorIOAdapter(OpenAIVectorStoreMixin, VectorIO, VectorDBsProtoc
self.files_api = files_api
self.cache: dict[str, VectorDBWithIndex] = {}
self.openai_vector_stores: dict[str, dict[str, Any]] = {}
self.kvstore: KVStore | None = None
async def initialize(self) -> None:
def _setup_connection():
# Open a connection to the SQLite database (the file is specified in the config).
self.kvstore = await kvstore_impl(self.config.kvstore)
start_key = VECTOR_DBS_PREFIX
end_key = f"{VECTOR_DBS_PREFIX}\xff"
stored_vector_dbs = await self.kvstore.values_in_range(start_key, end_key)
for db_json in stored_vector_dbs:
vector_db = VectorDB.model_validate_json(db_json)
index = await SQLiteVecIndex.create(
vector_db.embedding_dimension,
self.config.db_path,
vector_db.identifier,
)
self.cache[vector_db.identifier] = VectorDBWithIndex(vector_db, index, self.inference_api)
# load any existing OpenAI vector stores
self.openai_vector_stores = await self._load_openai_vector_stores()
async def shutdown(self) -> None:
# nothing to do since we don't maintain a persistent connection
pass
async def list_vector_dbs(self) -> list[VectorDB]:
return [v.vector_db for v in self.cache.values()]
async def register_vector_db(self, vector_db: VectorDB) -> None:
index = await SQLiteVecIndex.create(
vector_db.embedding_dimension,
self.config.db_path,
vector_db.identifier,
)
self.cache[vector_db.identifier] = VectorDBWithIndex(vector_db, index, self.inference_api)
async def _get_and_cache_vector_db_index(self, vector_db_id: str) -> VectorDBWithIndex | None:
if vector_db_id in self.cache:
return self.cache[vector_db_id]
if self.vector_db_store is None:
raise ValueError(f"Vector DB {vector_db_id} not found")
vector_db = self.vector_db_store.get_vector_db(vector_db_id)
if not vector_db:
raise ValueError(f"Vector DB {vector_db_id} not found")
index = VectorDBWithIndex(
vector_db=vector_db,
index=SQLiteVecIndex(
dimension=vector_db.embedding_dimension,
db_path=self.config.db_path,
bank_id=vector_db.identifier,
kvstore=self.kvstore,
),
inference_api=self.inference_api,
)
self.cache[vector_db_id] = index
return index
async def unregister_vector_db(self, vector_db_id: str) -> None:
if vector_db_id not in self.cache:
logger.warning(f"Vector DB {vector_db_id} not found")
return
await self.cache[vector_db_id].index.delete()
del self.cache[vector_db_id]
# OpenAI Vector Store Mixin abstract method implementations
async def _save_openai_vector_store(self, store_id: str, store_info: dict[str, Any]) -> None:
"""Save vector store metadata to SQLite database."""
assert self.kvstore is not None
key = f"{OPENAI_VECTOR_STORES_PREFIX}{store_id}"
await self.kvstore.set(key=key, value=json.dumps(store_info))
self.openai_vector_stores[store_id] = store_info
async def _load_openai_vector_stores(self) -> dict[str, dict[str, Any]]:
"""Load all vector store metadata from SQLite database."""
assert self.kvstore is not None
start_key = OPENAI_VECTOR_STORES_PREFIX
end_key = f"{OPENAI_VECTOR_STORES_PREFIX}\xff"
stored_openai_stores = await self.kvstore.values_in_range(start_key, end_key)
stores = {}
for store_data in stored_openai_stores:
store_info = json.loads(store_data)
stores[store_info["id"]] = store_info
return stores
async def _update_openai_vector_store(self, store_id: str, store_info: dict[str, Any]) -> None:
"""Update vector store metadata in SQLite database."""
assert self.kvstore is not None
key = f"{OPENAI_VECTOR_STORES_PREFIX}{store_id}"
await self.kvstore.set(key=key, value=json.dumps(store_info))
self.openai_vector_stores[store_id] = store_info
async def _delete_openai_vector_store_from_storage(self, store_id: str) -> None:
"""Delete vector store metadata from SQLite database."""
assert self.kvstore is not None
key = f"{OPENAI_VECTOR_STORES_PREFIX}{store_id}"
await self.kvstore.delete(key)
if store_id in self.openai_vector_stores:
del self.openai_vector_stores[store_id]
async def _save_openai_vector_store_file(
self, store_id: str, file_id: str, file_info: dict[str, Any], file_contents: list[dict[str, Any]]
) -> None:
"""Save vector store file metadata to SQLite database."""
def _create_or_store():
connection = _create_sqlite_connection(self.config.db_path)
cur = connection.cursor()
try:
# Create a table to persist vector DB registrations.
cur.execute("""
CREATE TABLE IF NOT EXISTS vector_dbs (
id TEXT PRIMARY KEY,
metadata TEXT
);
""")
# Create a table to persist OpenAI vector stores.
cur.execute("""
CREATE TABLE IF NOT EXISTS openai_vector_stores (
id TEXT PRIMARY KEY,
metadata TEXT
);
""")
# Create a table to persist OpenAI vector store files.
cur.execute("""
CREATE TABLE IF NOT EXISTS openai_vector_store_files (
@ -464,168 +563,6 @@ class SQLiteVecVectorIOAdapter(OpenAIVectorStoreMixin, VectorIO, VectorDBsProtoc
);
""")
connection.commit()
# Load any existing vector DB registrations.
cur.execute("SELECT metadata FROM vector_dbs")
vector_db_rows = cur.fetchall()
return vector_db_rows
finally:
cur.close()
connection.close()
vector_db_rows = await asyncio.to_thread(_setup_connection)
# Load existing vector DBs
for row in vector_db_rows:
vector_db_data = row[0]
vector_db = VectorDB.model_validate_json(vector_db_data)
index = await SQLiteVecIndex.create(
vector_db.embedding_dimension,
self.config.db_path,
vector_db.identifier,
)
self.cache[vector_db.identifier] = VectorDBWithIndex(vector_db, index, self.inference_api)
# Load existing OpenAI vector stores using the mixin method
self.openai_vector_stores = await self._load_openai_vector_stores()
async def shutdown(self) -> None:
# nothing to do since we don't maintain a persistent connection
pass
async def register_vector_db(self, vector_db: VectorDB) -> None:
def _register_db():
connection = _create_sqlite_connection(self.config.db_path)
cur = connection.cursor()
try:
cur.execute(
"INSERT OR REPLACE INTO vector_dbs (id, metadata) VALUES (?, ?)",
(vector_db.identifier, vector_db.model_dump_json()),
)
connection.commit()
finally:
cur.close()
connection.close()
await asyncio.to_thread(_register_db)
index = await SQLiteVecIndex.create(
vector_db.embedding_dimension,
self.config.db_path,
vector_db.identifier,
)
self.cache[vector_db.identifier] = VectorDBWithIndex(vector_db, index, self.inference_api)
async def list_vector_dbs(self) -> list[VectorDB]:
return [v.vector_db for v in self.cache.values()]
async def unregister_vector_db(self, vector_db_id: str) -> None:
if vector_db_id not in self.cache:
logger.warning(f"Vector DB {vector_db_id} not found")
return
await self.cache[vector_db_id].index.delete()
del self.cache[vector_db_id]
def _delete_vector_db_from_registry():
connection = _create_sqlite_connection(self.config.db_path)
cur = connection.cursor()
try:
cur.execute("DELETE FROM vector_dbs WHERE id = ?", (vector_db_id,))
connection.commit()
finally:
cur.close()
connection.close()
await asyncio.to_thread(_delete_vector_db_from_registry)
# OpenAI Vector Store Mixin abstract method implementations
async def _save_openai_vector_store(self, store_id: str, store_info: dict[str, Any]) -> None:
"""Save vector store metadata to SQLite database."""
def _store():
connection = _create_sqlite_connection(self.config.db_path)
cur = connection.cursor()
try:
cur.execute(
"INSERT OR REPLACE INTO openai_vector_stores (id, metadata) VALUES (?, ?)",
(store_id, json.dumps(store_info)),
)
connection.commit()
except Exception as e:
logger.error(f"Error saving openai vector store {store_id}: {e}")
raise
finally:
cur.close()
connection.close()
try:
await asyncio.to_thread(_store)
except Exception as e:
logger.error(f"Error saving openai vector store {store_id}: {e}")
raise
async def _load_openai_vector_stores(self) -> dict[str, dict[str, Any]]:
"""Load all vector store metadata from SQLite database."""
def _load():
connection = _create_sqlite_connection(self.config.db_path)
cur = connection.cursor()
try:
cur.execute("SELECT metadata FROM openai_vector_stores")
rows = cur.fetchall()
return rows
finally:
cur.close()
connection.close()
rows = await asyncio.to_thread(_load)
stores = {}
for row in rows:
store_data = row[0]
store_info = json.loads(store_data)
stores[store_info["id"]] = store_info
return stores
async def _update_openai_vector_store(self, store_id: str, store_info: dict[str, Any]) -> None:
"""Update vector store metadata in SQLite database."""
def _update():
connection = _create_sqlite_connection(self.config.db_path)
cur = connection.cursor()
try:
cur.execute(
"UPDATE openai_vector_stores SET metadata = ? WHERE id = ?",
(json.dumps(store_info), store_id),
)
connection.commit()
finally:
cur.close()
connection.close()
await asyncio.to_thread(_update)
async def _delete_openai_vector_store_from_storage(self, store_id: str) -> None:
"""Delete vector store metadata from SQLite database."""
def _delete():
connection = _create_sqlite_connection(self.config.db_path)
cur = connection.cursor()
try:
cur.execute("DELETE FROM openai_vector_stores WHERE id = ?", (store_id,))
connection.commit()
finally:
cur.close()
connection.close()
await asyncio.to_thread(_delete)
async def _save_openai_vector_store_file(
self, store_id: str, file_id: str, file_info: dict[str, Any], file_contents: list[dict[str, Any]]
) -> None:
"""Save vector store file metadata to SQLite database."""
def _store():
connection = _create_sqlite_connection(self.config.db_path)
cur = connection.cursor()
try:
cur.execute(
"INSERT OR REPLACE INTO openai_vector_store_files (store_id, file_id, metadata) VALUES (?, ?, ?)",
(store_id, file_id, json.dumps(file_info)),
@ -643,7 +580,7 @@ class SQLiteVecVectorIOAdapter(OpenAIVectorStoreMixin, VectorIO, VectorDBsProtoc
connection.close()
try:
await asyncio.to_thread(_store)
await asyncio.to_thread(_create_or_store)
except Exception as e:
logger.error(f"Error saving openai vector store file {store_id} {file_id}: {e}")
raise
@ -722,6 +659,10 @@ class SQLiteVecVectorIOAdapter(OpenAIVectorStoreMixin, VectorIO, VectorDBsProtoc
cur.execute(
"DELETE FROM openai_vector_store_files WHERE store_id = ? AND file_id = ?", (store_id, file_id)
)
cur.execute(
"DELETE FROM openai_vector_store_files_contents WHERE store_id = ? AND file_id = ?",
(store_id, file_id),
)
connection.commit()
finally:
cur.close()
@ -730,15 +671,17 @@ class SQLiteVecVectorIOAdapter(OpenAIVectorStoreMixin, VectorIO, VectorDBsProtoc
await asyncio.to_thread(_delete)
async def insert_chunks(self, vector_db_id: str, chunks: list[Chunk], ttl_seconds: int | None = None) -> None:
if vector_db_id not in self.cache:
raise ValueError(f"Vector DB {vector_db_id} not found. Found: {list(self.cache.keys())}")
index = await self._get_and_cache_vector_db_index(vector_db_id)
if not index:
raise ValueError(f"Vector DB {vector_db_id} not found")
# The VectorDBWithIndex helper is expected to compute embeddings via the inference_api
# and then call our index's add_chunks.
await self.cache[vector_db_id].insert_chunks(chunks)
await index.insert_chunks(chunks)
async def query_chunks(
self, vector_db_id: str, query: Any, params: dict[str, Any] | None = None
) -> QueryChunksResponse:
if vector_db_id not in self.cache:
index = await self._get_and_cache_vector_db_index(vector_db_id)
if not index:
raise ValueError(f"Vector DB {vector_db_id} not found")
return await self.cache[vector_db_id].query_chunks(query, params)
return await index.query_chunks(query, params)