phoenix-oss/llama-stack-mirror
0
0
Fork 1
mirror of https://github.com/meta-llama/llama-stack.git synced 2025-10-22 16:23:08 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

feat(stores)!: use backend storage references instead of configs (#3697)

Browse source 
**This PR changes configurations in a backward incompatible way.**

Run configs today repeat full SQLite/Postgres snippets everywhere a
store is needed, which means duplicated credentials, extra connection
pools, and lots of drift between files. This PR introduces named storage
backends so the stack and providers can share a single catalog and
reference those backends by name.

## Key Changes

- Add `storage.backends` to `StackRunConfig`, register each KV/SQL
backend once at startup, and validate that references point to the right
family.
- Move server stores under `storage.stores` with lightweight references
(backend + namespace/table) instead of full configs.
- Update every provider/config/doc to use the new reference style;
docs/codegen now surface the simplified YAML.

## Migration

Before:
```yaml
metadata_store:
  type: sqlite
  db_path: ~/.llama/distributions/foo/registry.db
inference_store:
  type: postgres
  host: ${env.POSTGRES_HOST}
  port: ${env.POSTGRES_PORT}
  db: ${env.POSTGRES_DB}
  user: ${env.POSTGRES_USER}
  password: ${env.POSTGRES_PASSWORD}
conversations_store:
  type: postgres
  host: ${env.POSTGRES_HOST}
  port: ${env.POSTGRES_PORT}
  db: ${env.POSTGRES_DB}
  user: ${env.POSTGRES_USER}
  password: ${env.POSTGRES_PASSWORD}
```

After:
```yaml
storage:
  backends:
    kv_default:
      type: kv_sqlite
      db_path: ~/.llama/distributions/foo/kvstore.db
    sql_default:
      type: sql_postgres
      host: ${env.POSTGRES_HOST}
      port: ${env.POSTGRES_PORT}
      db: ${env.POSTGRES_DB}
      user: ${env.POSTGRES_USER}
      password: ${env.POSTGRES_PASSWORD}
  stores:
    metadata:
      backend: kv_default
      namespace: registry
    inference:
      backend: sql_default
      table_name: inference_store
      max_write_queue_size: 10000
      num_writers: 4
    conversations:
      backend: sql_default
      table_name: openai_conversations
```

Provider configs follow the same pattern—for example, a Chroma vector
adapter switches from:

```yaml
providers:
  vector_io:
  - provider_id: chromadb
    provider_type: remote::chromadb
    config:
      url: ${env.CHROMADB_URL}
      kvstore:
        type: sqlite
        db_path: ~/.llama/distributions/foo/chroma.db
```

to:

```yaml
providers:
  vector_io:
  - provider_id: chromadb
    provider_type: remote::chromadb
    config:
      url: ${env.CHROMADB_URL}
      persistence:
        backend: kv_default
        namespace: vector_io::chroma_remote
```

Once the backends are declared, everything else just points at them, so
rotating credentials or swapping to Postgres happens in one place and
the stack reuses a single connection pool.
This commit is contained in:
Ashwin Bharambe 2025-10-20 13:20:09 -07:00 • committed by GitHub
parent add64e8e2a
commit 2c43285e22
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
105 changed files with 2290 additions and 1292 deletions
Download patch file Download diff file Expand all files Collapse all files

2
llama_stack/providers/remote/vector_io/chroma/chroma.py
View file

@ -151,7 +151,7 @@ class ChromaVectorIOAdapter(OpenAIVectorStoreMixin, VectorIO, VectorDBsProtocolP
self.vector_db_store = None
async def initialize(self) -> None:
self.kvstore = await kvstore_impl(self.config.kvstore)
self.kvstore = await kvstore_impl(self.config.persistence)
self.vector_db_store = self.kvstore
if isinstance(self.config, RemoteChromaVectorIOConfig):

12
llama_stack/providers/remote/vector_io/chroma/config.py
View file

@ -8,21 +8,21 @@ from typing import Any
from pydantic import BaseModel, Field
from llama_stack.providers.utils.kvstore.config import KVStoreConfig, SqliteKVStoreConfig
from llama_stack.core.storage.datatypes import KVStoreReference
from llama_stack.schema_utils import json_schema_type
@json_schema_type
class ChromaVectorIOConfig(BaseModel):
url: str | None
kvstore: KVStoreConfig = Field(description="Config for KV store backend")
persistence: KVStoreReference = Field(description="Config for KV store backend")
@classmethod
def sample_run_config(cls, __distro_dir__: str, url: str = "${env.CHROMADB_URL}", **kwargs: Any) -> dict[str, Any]:
return {
"url": url,
"kvstore": SqliteKVStoreConfig.sample_run_config(
__distro_dir__=__distro_dir__,
db_name="chroma_remote_registry.db",
),
"persistence": KVStoreReference(
backend="kv_default",
namespace="vector_io::chroma_remote",
).model_dump(exclude_none=True),
}

12
llama_stack/providers/remote/vector_io/milvus/config.py
View file

@ -8,7 +8,7 @@ from typing import Any
from pydantic import BaseModel, ConfigDict, Field
from llama_stack.providers.utils.kvstore.config import KVStoreConfig, SqliteKVStoreConfig
from llama_stack.core.storage.datatypes import KVStoreReference
from llama_stack.schema_utils import json_schema_type
@ -17,7 +17,7 @@ class MilvusVectorIOConfig(BaseModel):
uri: str = Field(description="The URI of the Milvus server")
token: str | None = Field(description="The token of the Milvus server")
consistency_level: str = Field(description="The consistency level of the Milvus server", default="Strong")
kvstore: KVStoreConfig = Field(description="Config for KV store backend")
persistence: KVStoreReference = Field(description="Config for KV store backend")
# This configuration allows additional fields to be passed through to the underlying Milvus client.
# See the [Milvus](https://milvus.io/docs/install-overview.md) documentation for more details about Milvus in general.
@ -28,8 +28,8 @@ class MilvusVectorIOConfig(BaseModel):
return {
"uri": "${env.MILVUS_ENDPOINT}",
"token": "${env.MILVUS_TOKEN}",
"kvstore": SqliteKVStoreConfig.sample_run_config(
__distro_dir__=__distro_dir__,
db_name="milvus_remote_registry.db",
),
"persistence": KVStoreReference(
backend="kv_default",
namespace="vector_io::milvus_remote",
).model_dump(exclude_none=True),
}

2
llama_stack/providers/remote/vector_io/milvus/milvus.py
View file

@ -321,7 +321,7 @@ class MilvusVectorIOAdapter(OpenAIVectorStoreMixin, VectorIO, VectorDBsProtocolP
self.metadata_collection_name = "openai_vector_stores_metadata"
async def initialize(self) -> None:
self.kvstore = await kvstore_impl(self.config.kvstore)
self.kvstore = await kvstore_impl(self.config.persistence)
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)

17
llama_stack/providers/remote/vector_io/pgvector/config.py
View file

@ -8,10 +8,7 @@ from typing import Any
from pydantic import BaseModel, Field
from llama_stack.providers.utils.kvstore.config import (
KVStoreConfig,
SqliteKVStoreConfig,
)
from llama_stack.core.storage.datatypes import KVStoreReference
from llama_stack.schema_utils import json_schema_type
@ -22,7 +19,9 @@ class PGVectorVectorIOConfig(BaseModel):
db: str | None = Field(default="postgres")
user: str | None = Field(default="postgres")
password: str | None = Field(default="mysecretpassword")
kvstore: KVStoreConfig | None = Field(description="Config for KV store backend (SQLite only for now)", default=None)
persistence: KVStoreReference | None = Field(
description="Config for KV store backend (SQLite only for now)", default=None
)
@classmethod
def sample_run_config(
@ -41,8 +40,8 @@ class PGVectorVectorIOConfig(BaseModel):
"db": db,
"user": user,
"password": password,
"kvstore": SqliteKVStoreConfig.sample_run_config(
__distro_dir__=__distro_dir__,
db_name="pgvector_registry.db",
),
"persistence": KVStoreReference(
backend="kv_default",
namespace="vector_io::pgvector",
).model_dump(exclude_none=True),
}

2
llama_stack/providers/remote/vector_io/pgvector/pgvector.py
View file

@ -358,7 +358,7 @@ class PGVectorVectorIOAdapter(OpenAIVectorStoreMixin, VectorIO, VectorDBsProtoco
async def initialize(self) -> None:
log.info(f"Initializing PGVector memory adapter with config: {self.config}")
self.kvstore = await kvstore_impl(self.config.kvstore)
self.kvstore = await kvstore_impl(self.config.persistence)
await self.initialize_openai_vector_stores()
try:

15
llama_stack/providers/remote/vector_io/qdrant/config.py
View file

@ -8,10 +8,7 @@ from typing import Any
from pydantic import BaseModel
from llama_stack.providers.utils.kvstore.config import (
KVStoreConfig,
SqliteKVStoreConfig,
)
from llama_stack.core.storage.datatypes import KVStoreReference
from llama_stack.schema_utils import json_schema_type
@ -27,14 +24,14 @@ class QdrantVectorIOConfig(BaseModel):
prefix: str | None = None
timeout: int | None = None
host: str | None = None
kvstore: KVStoreConfig
persistence: KVStoreReference
@classmethod
def sample_run_config(cls, __distro_dir__: str, **kwargs: Any) -> dict[str, Any]:
return {
"api_key": "${env.QDRANT_API_KEY:=}",
"kvstore": SqliteKVStoreConfig.sample_run_config(
__distro_dir__=__distro_dir__,
db_name="qdrant_registry.db",
),
"persistence": KVStoreReference(
backend="kv_default",
namespace="vector_io::qdrant_remote",
).model_dump(exclude_none=True),
}

4
llama_stack/providers/remote/vector_io/qdrant/qdrant.py
View file

@ -174,9 +174,9 @@ class QdrantVectorIOAdapter(OpenAIVectorStoreMixin, VectorIO, VectorDBsProtocolP
self._qdrant_lock = asyncio.Lock()
async def initialize(self) -> None:
client_config = self.config.model_dump(exclude_none=True, exclude={"kvstore"})
client_config = self.config.model_dump(exclude_none=True, exclude={"persistence"})
self.client = AsyncQdrantClient(**client_config)
self.kvstore = await kvstore_impl(self.config.kvstore)
self.kvstore = await kvstore_impl(self.config.persistence)
start_key = VECTOR_DBS_PREFIX
end_key = f"{VECTOR_DBS_PREFIX}\xff"

17
llama_stack/providers/remote/vector_io/weaviate/config.py
View file

@ -8,10 +8,7 @@ from typing import Any
from pydantic import BaseModel, Field
from llama_stack.providers.utils.kvstore.config import (
KVStoreConfig,
SqliteKVStoreConfig,
)
from llama_stack.core.storage.datatypes import KVStoreReference
from llama_stack.schema_utils import json_schema_type
@ -19,7 +16,9 @@ from llama_stack.schema_utils import json_schema_type
class WeaviateVectorIOConfig(BaseModel):
weaviate_api_key: str | None = Field(description="The API key for the Weaviate instance", default=None)
weaviate_cluster_url: str | None = Field(description="The URL of the Weaviate cluster", default="localhost:8080")
kvstore: KVStoreConfig | None = Field(description="Config for KV store backend (SQLite only for now)", default=None)
persistence: KVStoreReference | None = Field(
description="Config for KV store backend (SQLite only for now)", default=None
)
@classmethod
def sample_run_config(
@ -30,8 +29,8 @@ class WeaviateVectorIOConfig(BaseModel):
return {
"weaviate_api_key": None,
"weaviate_cluster_url": "${env.WEAVIATE_CLUSTER_URL:=localhost:8080}",
"kvstore": SqliteKVStoreConfig.sample_run_config(
__distro_dir__=__distro_dir__,
db_name="weaviate_registry.db",
),
"persistence": KVStoreReference(
backend="kv_default",
namespace="vector_io::weaviate",
).model_dump(exclude_none=True),
}

4
llama_stack/providers/remote/vector_io/weaviate/weaviate.py
View file

@ -320,8 +320,8 @@ class WeaviateVectorIOAdapter(
async def initialize(self) -> None:
"""Set up KV store and load existing vector DBs and OpenAI vector stores."""
# Initialize KV store for metadata if configured
if self.config.kvstore is not None:
self.kvstore = await kvstore_impl(self.config.kvstore)
if self.config.persistence is not None:
self.kvstore = await kvstore_impl(self.config.persistence)
else:
self.kvstore = None
log.info("No kvstore configured, registry will not persist across restarts")