LCORE-1427: Normalize Tool and Inline RAG default behavior

docs/devel_doc/openapi.json

@@ -18084,13 +18084,13 @@
                         },
                         "type": "array",
                         "title": "Tool RAG IDs",
-                        "description": "RAG IDs made available to the LLM as a file_search tool. Use 'okp' to include the OKP vector store. When omitted, all registered BYOK vector stores are used (backward compatibility)."
+                        "description": "RAG IDs made available to the LLM as a file_search tool. Use 'okp' to include the OKP vector store. When omitted, tool RAG is disabled."
                     }
                 },
                 "additionalProperties": false,
                 "type": "object",
                 "title": "RagConfiguration",
-                "description": "RAG strategy configuration.\n\nControls which RAG sources are used for inline and tool-based retrieval.\n\nEach strategy lists RAG IDs to include. The special ID ``\"okp\"`` defined in constants,\nactivates the OKP provider; all other IDs refer to entries in ``byok_rag``.\n\nBackward compatibility:\n    - ``inline`` defaults to ``[]`` (no inline RAG).\n    - ``tool`` defaults to ``[]`` (no tool RAG).\n\nIf no RAG strategy is defined (inline and tool are empty),\nthe RAG tool will register all stores available to llama-stack."
+                "description": "RAG strategy configuration.\n\nControls which RAG sources are used for inline and tool-based retrieval.\n\nEach strategy lists RAG IDs to include. The special ID ``\"okp\"`` defined in constants,\nactivates the OKP provider; all other IDs refer to entries in ``byok_rag``.\n\nBoth ``inline`` and ``tool`` default to ``[]`` (disabled).\nEach must be explicitly configured to activate its respective RAG strategy."
             },
             "ReadinessResponse": {
                 "properties": {

docs/devel_doc/openapi.md

@@ -8155,18 +8155,14 @@ Controls which RAG sources are used for inline and tool-based retrieval.
 Each strategy lists RAG IDs to include. The special ID ``"okp"`` defined in constants,
 activates the OKP provider; all other IDs refer to entries in ``byok_rag``.
 
-Backward compatibility:
-    - ``inline`` defaults to ``[]`` (no inline RAG).
-    - ``tool`` defaults to ``[]`` (no tool RAG).
-
-If no RAG strategy is defined (inline and tool are empty),
-the RAG tool will register all stores available to llama-stack.
+Both ``inline`` and ``tool`` default to ``[]`` (disabled).
+Each must be explicitly configured to activate its respective RAG strategy.
 
 
 | Field | Type | Description |
 |-------|------|-------------|
 | inline | array | RAG IDs whose sources are injected as context before the LLM call. Use 'okp' to enable OKP inline RAG. Empty by default (no inline RAG). |
-| tool | array | RAG IDs made available to the LLM as a file_search tool. Use 'okp' to include the OKP vector store. When omitted, all registered BYOK vector stores are used (backward compatibility). |
+| tool | array | RAG IDs made available to the LLM as a file_search tool. Use 'okp' to include the OKP vector store. When omitted, tool RAG is disabled. |
 
 
 ## ReadinessResponse

docs/models/responses_succ.md

@@ -1840,18 +1840,14 @@ Controls which RAG sources are used for inline and tool-based retrieval.
 Each strategy lists RAG IDs to include. The special ID ``"okp"`` defined in constants,
 activates the OKP provider; all other IDs refer to entries in ``byok_rag``.
 
-Backward compatibility:
-    - ``inline`` defaults to ``[]`` (no inline RAG).
-    - ``tool`` defaults to ``[]`` (no tool RAG).
-
-If no RAG strategy is defined (inline and tool are empty),
-the RAG tool will register all stores available to llama-stack.
+Both ``inline`` and ``tool`` default to ``[]`` (disabled).
+Each must be explicitly configured to activate its respective RAG strategy.
 
 
 | Field | Type | Description |
 |-------|------|-------------|
 | inline | array | RAG IDs whose sources are injected as context before the LLM call. Use 'okp' to enable OKP inline RAG. Empty by default (no inline RAG). |
-| tool | array | RAG IDs made available to the LLM as a file_search tool. Use 'okp' to include the OKP vector store. When omitted, all registered BYOK vector stores are used (backward compatibility). |
+| tool | array | RAG IDs made available to the LLM as a file_search tool. Use 'okp' to include the OKP vector store. When omitted, tool RAG is disabled. |
 
 
 ## ReadinessResponse

docs/user_doc/byok_guide.md

@@ -261,7 +261,7 @@ rag:
     - okp             # include OKP context inline
 
   # Tool RAG: the LLM can call file_search to retrieve context on demand
-  # If omitted, tool RAG is disabled. If both tool and inline are omitted, all registered stores are used as fallback
+  # If omitted, tool RAG is disabled
   tool:
     - my-docs         # expose this BYOK store as the file_search tool
     - okp             # expose OKP as the file_search tool

docs/user_doc/config.md

@@ -606,18 +606,14 @@ Controls which RAG sources are used for inline and tool-based retrieval.
 Each strategy lists RAG IDs to include. The special ID ``"okp"`` defined in constants,
 activates the OKP provider; all other IDs refer to entries in ``byok_rag``.
 
-Backward compatibility:
-    - ``inline`` defaults to ``[]`` (no inline RAG).
-    - ``tool`` defaults to ``[]`` (no tool RAG).
-
-If no RAG strategy is defined (inline and tool are empty),
-the RAG tool will register all stores available to llama-stack.
+Both ``inline`` and ``tool`` default to ``[]`` (disabled).
+Each must be explicitly configured to activate its respective RAG strategy.
 
 
 | Field  | Type  | Description                                                                                                                                                                            |
 |--------|-------|----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
 | inline | array | RAG IDs whose sources are injected as context before the LLM call. Use 'okp' to enable OKP inline RAG. Empty by default (no inline RAG).                                               |
-| tool   | array | RAG IDs made available to the LLM as a file_search tool. Use 'okp' to include the OKP vector store. When omitted, all registered BYOK vector stores are used (backward compatibility). |
+| tool   | array | RAG IDs made available to the LLM as a file_search tool. Use 'okp' to include the OKP vector store. When omitted, tool RAG is disabled.                                               |
 
 
 ## RerankerConfiguration

examples/lightspeed-stack-byok-okp-rag.yaml

@@ -60,7 +60,7 @@ rag:
     - okp
   # Tool RAG: LLM can call file_search on demand to retrieve context
   # List rag_ids from byok_rag, or 'okp' to include OKP
-  # Omit to use all registered BYOK stores (backward compatibility)
+  # Omit to disable tool RAG
   tool:
     - ocp-docs
     - knowledge-base

src/models/config.py

@@ -2258,12 +2258,8 @@ class RagConfiguration(ConfigurationBase):
     Each strategy lists RAG IDs to include. The special ID ``"okp"`` defined in constants,
     activates the OKP provider; all other IDs refer to entries in ``byok_rag``.
 
-    Backward compatibility:
-        - ``inline`` defaults to ``[]`` (no inline RAG).
-        - ``tool`` defaults to ``[]`` (no tool RAG).
-
-    If no RAG strategy is defined (inline and tool are empty),
-    the RAG tool will register all stores available to llama-stack.
+    Both ``inline`` and ``tool`` default to ``[]`` (disabled).
+    Each must be explicitly configured to activate its respective RAG strategy.
     """
 
     inline: list[str] = Field(
@@ -2278,7 +2274,7 @@ class RagConfiguration(ConfigurationBase):
         title="Tool RAG IDs",
         description="RAG IDs made available to the LLM as a file_search tool. "
         f"Use '{constants.OKP_RAG_ID}' to include the OKP vector store. "
-        "When omitted, all registered BYOK vector stores are used (backward compatibility).",
+        "When omitted, tool RAG is disabled.",
     )
 
 

src/utils/responses.py

@@ -81,7 +81,6 @@
 from llama_stack_client import APIConnectionError, APIStatusError, AsyncLlamaStackClient
 
 import constants
-from client import AsyncLlamaStackClientHolder
 from configuration import configuration
 from constants import DEFAULT_RAG_TOOL
 from log import get_logger
@@ -127,44 +126,6 @@
 logger = get_logger(__name__)
 
 
-async def get_vector_store_ids(
-    client: AsyncLlamaStackClient,
-    vector_store_ids: Optional[list[str]] = None,
-) -> list[str]:
-    """Get vector store IDs for querying.
-
-    If vector_store_ids are provided, returns them. Otherwise fetches all
-    available vector stores from Llama Stack.
-
-    Args:
-        client: The AsyncLlamaStackClient to use for fetching stores
-        vector_store_ids: Optional list of vector store IDs. If provided,
-            returns this list. If None, fetches all available vector stores.
-
-    Returns:
-        List of vector store IDs to query
-
-    Raises:
-        HTTPException: With ServiceUnavailableResponse if connection fails,
-            or InternalServerErrorResponse if API returns an error status
-    """
-    if vector_store_ids is not None:
-        return vector_store_ids
-
-    try:
-        vector_stores = await client.vector_stores.list()
-        return [vector_store.id for vector_store in vector_stores.data]
-    except APIConnectionError as e:
-        error_response = ServiceUnavailableResponse(
-            backend_name="Llama Stack",
-            cause=str(e),
-        )
-        raise HTTPException(**error_response.model_dump()) from e
-    except APIStatusError as e:
-        error_response = InternalServerErrorResponse.generic()
-        raise HTTPException(**error_response.model_dump()) from e
-
-
 async def get_topic_summary(  # pylint: disable=too-many-nested-blocks
     question: str, client: AsyncLlamaStackClient, model_id: str
 ) -> str:
@@ -229,7 +190,6 @@ async def maybe_get_topic_summary(
 
 
 async def prepare_tools(  # pylint: disable=too-many-arguments,too-many-positional-arguments
-    client: AsyncLlamaStackClient,
     vector_store_ids: Optional[list[str]],
     no_tools: Optional[bool],
     token: str,
@@ -239,9 +199,8 @@ async def prepare_tools(  # pylint: disable=too-many-arguments,too-many-position
     """Prepare tools for Responses API including RAG and MCP tools.
 
     Args:
-        client: The Llama Stack client instance
         vector_store_ids: The list of vector store IDs to use for RAG tools
-            or None if all vector stores should be used
+            or None to fall back to rag.tool configuration
         no_tools: Whether to skip tool preparation
         token: Authentication token for MCP tools
         mcp_headers: Per-request headers for MCP servers
@@ -259,22 +218,16 @@ async def prepare_tools(  # pylint: disable=too-many-arguments,too-many-position
     # Vector store ID resolution priority:
     #   1. Per-request IDs: highest prio; customer-facing rag_ids are translated to vector_db_ids.
     #   2. rag.tool config IDs: used when no per-request IDs provided, and rag.tool is configured.
-    #      If rag.inline is configured, but not rag.tool, tool RAG is disabled.
-    #   3. All registered vector DBs: fallback when neither rag.tool nor rag.inline are configured.
-    #      IDs fetched from llama-stack are already internal and need no translation.
     byok_rags = configuration.configuration.byok_rag
 
     is_tool_rag_enabled = len(configuration.configuration.rag.tool) > 0
-    is_inline_rag_enabled = len(configuration.configuration.rag.inline) > 0
 
     if vector_store_ids is not None:
         effective_ids = resolve_vector_store_ids(vector_store_ids, byok_rags)
     elif is_tool_rag_enabled:
         effective_ids = resolve_vector_store_ids(
             configuration.configuration.rag.tool, byok_rags
         )
-    elif not is_inline_rag_enabled:
-        effective_ids = await get_vector_store_ids(client, None)
 
     # Add RAG tools if vector stores are available
     rag_tools = get_rag_tools(effective_ids)
@@ -375,7 +328,6 @@ async def prepare_responses_params(  # pylint: disable=too-many-arguments,too-ma
 
     # Prepare tools for responses API
     tools = await prepare_tools(
-        client,
         query_request.vector_store_ids,
         query_request.no_tools,
         token,
@@ -1754,9 +1706,7 @@ async def _resolve_client_tools(
 
     # Optionally merge server-configured tools (RAG, MCP) with client tools
     if merge_server_tools:
-        client = AsyncLlamaStackClientHolder().get_client()
         server_tools = await prepare_tools(
-            client=client,
             vector_store_ids=vector_store_ids,
             no_tools=False,
             token=token,
@@ -1784,9 +1734,7 @@ async def _resolve_server_tools(
     Returns:
         List of server-configured tools, or None if none are configured.
     """
-    client = AsyncLlamaStackClientHolder().get_client()
     return await prepare_tools(
-        client=client,
         vector_store_ids=None,  # allow all vector stores configured
         no_tools=False,
         token=token,
@@ -1830,9 +1778,7 @@ async def resolve_tool_choice(
 
     if tools is None:
         # Register all tools configured in LCORE configuration
-        client = AsyncLlamaStackClientHolder().get_client()
         prepared_tools = await prepare_tools(
-            client=client,
             vector_store_ids=None,  # allow all vector stores configured
             no_tools=False,
             token=token,

tests/integration/container_lifecycle/test_container_lifecycle.py

@@ -8,6 +8,7 @@
 import time
 import urllib.error
 import urllib.request
+import warnings
 from collections.abc import Generator
 
 import pytest
@@ -18,13 +19,17 @@
 CONTAINER_START_TIMEOUT = 300  # 5 minutes for container start
 CONTAINER_STOP_TIMEOUT = 15
 CONTAINER_CLEANUP_TIMEOUT = 10
+IMAGE_CLEANUP_TIMEOUT = 30
+DANGLING_IMAGES_CLEANUP_TIMEOUT = 300  # 5 minutes for dangling images cleanup
 HEALTH_CHECK_TIMEOUT = 5
 PORT_QUERY_TIMEOUT = 5
 
 # Retry constants
 HEALTH_CHECK_MAX_ATTEMPTS = 30
 NETWORK_BINDING_MAX_ATTEMPTS = 5
 
+DEFAULT_LIGHTSPEED_LLAMA_STACK_IMAGE_NAME = "lightspeed-llama-stack:local"
+
 
 @pytest.fixture(scope="session")
 def container_runtime() -> str:
@@ -52,6 +57,39 @@ def container_runtime() -> str:
     pytest.skip("No container runtime available")
 
 
+@pytest.fixture(scope="session", autouse=True)
+def cleanup_container_artifacts(container_runtime: str) -> Generator[None]:
+    """Remove container images and dangling layers after all tests complete.
+
+    Parameters
+    ----------
+        container_runtime (str): Container runtime to use.
+
+    Yields
+    ------
+        None
+    """
+    yield
+
+    try:
+        subprocess.run(
+            [container_runtime, "rmi", "-f", DEFAULT_LIGHTSPEED_LLAMA_STACK_IMAGE_NAME],
+            capture_output=True,
+            timeout=IMAGE_CLEANUP_TIMEOUT,
+        )
+    except (subprocess.CalledProcessError, subprocess.TimeoutExpired) as e:
+        warnings.warn(f"Image cleanup failed: {e}")
+
+    try:
+        subprocess.run(
+            [container_runtime, "image", "prune", "-f"],
+            capture_output=True,
+            timeout=DANGLING_IMAGES_CLEANUP_TIMEOUT,
+        )
+    except (subprocess.CalledProcessError, subprocess.TimeoutExpired) as e:
+        warnings.warn(f"Dangling image cleanup failed: {e}")
+
+
 @pytest.fixture(scope="class")
 def managed_container(container_runtime: str) -> Generator[str, None, None]:
     """Start container once for entire test class with strict cleanup.
@@ -102,7 +140,7 @@ class TestContainerBuild:
     """Test container image building with idempotency checks."""
 
     def _get_image_id(
-        self, runtime: str, image_name: str = "lightspeed-llama-stack:local"
+        self, runtime: str, image_name: str = DEFAULT_LIGHTSPEED_LLAMA_STACK_IMAGE_NAME
     ) -> str:
         """Get the unique, immutable Image ID (SHA256).
 
@@ -145,7 +183,7 @@ def test_build_llama_stack_image(self, container_runtime: str) -> None:
 
         # Verify image is listed with correct tag
         result = subprocess.run(
-            [container_runtime, "images", "lightspeed-llama-stack:local"],
+            [container_runtime, "images", DEFAULT_LIGHTSPEED_LLAMA_STACK_IMAGE_NAME],
             capture_output=True,
             text=True,
             timeout=PORT_QUERY_TIMEOUT,
@@ -548,7 +586,12 @@ def test_clean_removes_image_and_container(self, container_runtime: str) -> None
 
         # Verify image is removed
         result = subprocess.run(
-            [container_runtime, "images", "-q", "lightspeed-llama-stack:local"],
+            [
+                container_runtime,
+                "images",
+                "-q",
+                DEFAULT_LIGHTSPEED_LLAMA_STACK_IMAGE_NAME,
+            ],
             capture_output=True,
             text=True,
             timeout=PORT_QUERY_TIMEOUT,