Merge pull request #469 from tisnik/lcore-466-llama-stack-version-in-info-endpoint

LCORE-466: Llama Stack version in /info endpoint

Author: tisnik

docs/openapi.json

@@ -63,7 +63,15 @@
                             }
                         },
                         "name": "Service name",
-                        "version": "Service version"
+                        "service_version": "Service version",
+                        "llama_stack_version": "Llama Stack version"
+                    },
+                    "500": {
+                        "description": "Internal Server Error",
+                        "detail": {
+                            "response": "Unable to connect to Llama Stack",
+                            "cause": "Connection error."
+                        }
                     }
                 }
             }
@@ -1606,28 +1614,40 @@
                             "Lightspeed Stack"
                         ]
                     },
-                    "version": {
+                    "service_version": {
                         "type": "string",
-                        "title": "Version",
+                        "title": "Service Version",
                         "description": "Service version",
                         "examples": [
                             "0.1.0",
                             "0.2.0",
                             "1.0.0"
                         ]
+                    },
+                    "llama_stack_version": {
+                        "type": "string",
+                        "title": "Llama Stack Version",
+                        "description": "Llama Stack version",
+                        "examples": [
+                            "0.2.1",
+                            "0.2.2",
+                            "0.2.18"
+                        ]
                     }
                 },
                 "type": "object",
                 "required": [
                     "name",
-                    "version"
+                    "service_version",
+                    "llama_stack_version"
                 ],
                 "title": "InfoResponse",
-                "description": "Model representing a response to an info request.\n\nAttributes:\n    name: Service name.\n    version: Service version.\n\nExample:\n    ```python\n    info_response = InfoResponse(\n        name=\"Lightspeed Stack\",\n        version=\"1.0.0\",\n    )\n    ```",
+                "description": "Model representing a response to an info request.\n\nAttributes:\n    name: Service name.\n    service_version: Service version.\n    llama_stack_version: Llama Stack version.\n\nExample:\n    ```python\n    info_response = InfoResponse(\n        name=\"Lightspeed Stack\",\n        service_version=\"1.0.0\",\n        llama_stack_version=\"0.2.18\",\n    )\n    ```",
                 "examples": [
                     {
+                        "llama_stack_version": "1.0.0",
                         "name": "Lightspeed Stack",
-                        "version": "1.0.0"
+                        "service_version": "1.0.0"
                     }
                 ]
             },

docs/openapi.md

@@ -48,6 +48,7 @@ Returns:
 | Status Code | Description | Component |
 |-------------|-------------|-----------|
 | 200 | Successful Response | [InfoResponse](#inforesponse) |
+| 500 | Internal Server Error |  |
 ## GET `/v1/models`
 
 > **Models Endpoint Handler**
@@ -882,21 +883,24 @@ Model representing a response to an info request.
 
 Attributes:
     name: Service name.
-    version: Service version.
+    service_version: Service version.
+    llama_stack_version: Llama Stack version.
 
 Example:
     ```python
     info_response = InfoResponse(
         name="Lightspeed Stack",
-        version="1.0.0",
+        service_version="1.0.0",
+        llama_stack_version="0.2.18",
     )
     ```
 
 
 | Field | Type | Description |
 |-------|------|-------------|
 | name | string | Service name |
-| version | string | Service version |
+| service_version | string | Service version |
+| llama_stack_version | string | Llama Stack version |
 
 
 ## JsonPathOperator

docs/output.md

@@ -48,6 +48,7 @@ Returns:
 | Status Code | Description | Component |
 |-------------|-------------|-----------|
 | 200 | Successful Response | [InfoResponse](#inforesponse) |
+| 500 | Internal Server Error |  |
 ## GET `/v1/models`
 
 > **Models Endpoint Handler**
@@ -872,21 +873,24 @@ Model representing a response to an info request.
 
 Attributes:
     name: Service name.
-    version: Service version.
+    service_version: Service version.
+    llama_stack_version: Llama Stack version.
 
 Example:
     ```python
     info_response = InfoResponse(
         name="Lightspeed Stack",
-        version="1.0.0",
+        service_version="1.0.0",
+        llama_stack_version="0.2.18",
     )
     ```
 
 
 | Field | Type | Description |
 |-------|------|-------------|
 | name | string | Service name |
-| version | string | Service version |
+| service_version | string | Service version |
+| llama_stack_version | string | Llama Stack version |
 
 
 ## JsonPathOperator

src/app/endpoints/info.py

@@ -3,13 +3,15 @@
 import logging
 from typing import Annotated, Any
 
-from fastapi import APIRouter, Request
+from fastapi import APIRouter, HTTPException, Request, status
 from fastapi import Depends
+from llama_stack_client import APIConnectionError
 
 from auth.interface import AuthTuple
 from auth import get_auth_dependency
 from authorization.middleware import authorize
 from configuration import configuration
+from client import AsyncLlamaStackClientHolder
 from models.config import Action
 from models.responses import InfoResponse
 from version import __version__
@@ -23,7 +25,14 @@
 get_info_responses: dict[int | str, dict[str, Any]] = {
     200: {
         "name": "Service name",
-        "version": "Service version",
+        "service_version": "Service version",
+        "llama_stack_version": "Llama Stack version",
+    },
+    500: {
+        "detail": {
+            "response": "Unable to connect to Llama Stack",
+            "cause": "Connection error.",
+        }
     },
 }
 
@@ -49,4 +58,24 @@ async def info_endpoint_handler(
     # Nothing interesting in the request
     _ = request
 
-    return InfoResponse(name=configuration.configuration.name, version=__version__)
+    try:
+        # try to get Llama Stack client
+        client = AsyncLlamaStackClientHolder().get_client()
+        # retrieve version
+        llama_stack_version_object = await client.inspect.version()
+        llama_stack_version = llama_stack_version_object.version
+        return InfoResponse(
+            name=configuration.configuration.name,
+            service_version=__version__,
+            llama_stack_version=llama_stack_version,
+        )
+    # connection to Llama Stack server
+    except APIConnectionError as e:
+        logger.error("Unable to connect to Llama Stack: %s", e)
+        raise HTTPException(
+            status_code=status.HTTP_500_INTERNAL_SERVER_ERROR,
+            detail={
+                "response": "Unable to connect to Llama Stack",
+                "cause": str(e),
+            },
+        ) from e

src/models/responses.py

@@ -84,13 +84,15 @@ class InfoResponse(BaseModel):
 
     Attributes:
         name: Service name.
-        version: Service version.
+        service_version: Service version.
+        llama_stack_version: Llama Stack version.
 
     Example:
         ```python
         info_response = InfoResponse(
             name="Lightspeed Stack",
-            version="1.0.0",
+            service_version="1.0.0",
+            llama_stack_version="0.2.18",
         )
         ```
     """
@@ -100,18 +102,24 @@ class InfoResponse(BaseModel):
         examples=["Lightspeed Stack"],
     )
 
-    version: str = Field(
+    service_version: str = Field(
         description="Service version",
         examples=["0.1.0", "0.2.0", "1.0.0"],
     )
 
+    llama_stack_version: str = Field(
+        description="Llama Stack version",
+        examples=["0.2.1", "0.2.2", "0.2.18"],
+    )
+
     # provides examples for /docs endpoint
     model_config = {
         "json_schema_extra": {
             "examples": [
                 {
                     "name": "Lightspeed Stack",
-                    "version": "1.0.0",
+                    "service_version": "1.0.0",
+                    "llama_stack_version": "1.0.0",
                 }
             ]
         }

tests/unit/app/endpoints/test_info.py

@@ -1,7 +1,10 @@
 """Unit tests for the /info REST API endpoint."""
 
 import pytest
-from fastapi import Request
+from fastapi import Request, HTTPException, status
+
+from llama_stack_client import APIConnectionError
+from llama_stack_client.types import VersionInfo
 
 from app.endpoints.info import info_endpoint_handler
 from configuration import AppConfig
@@ -11,6 +14,9 @@
 @pytest.mark.asyncio
 async def test_info_endpoint(mocker):
     """Test the info endpoint handler."""
+    mock_authorization_resolvers(mocker)
+
+    # configuration for tests
     config_dict = {
         "name": "foo",
         "service": {
@@ -36,6 +42,14 @@ async def test_info_endpoint(mocker):
     cfg = AppConfig()
     cfg.init_from_dict(config_dict)
 
+    # Mock the LlamaStack client
+    mock_client = mocker.AsyncMock()
+    mock_client.inspect.version.return_value = VersionInfo(version="0.1.2")
+    mock_lsc = mocker.patch("client.AsyncLlamaStackClientHolder.get_client")
+    mock_lsc.return_value = mock_client
+    mock_config = mocker.Mock()
+    mocker.patch("app.endpoints.models.configuration", mock_config)
+
     # Mock configuration
     mocker.patch("configuration.configuration", cfg)
 
@@ -50,4 +64,62 @@ async def test_info_endpoint(mocker):
     response = await info_endpoint_handler(auth=auth, request=request)
     assert response is not None
     assert response.name is not None
-    assert response.version is not None
+    assert response.service_version is not None
+    assert response.llama_stack_version == "0.1.2"
+
+
+@pytest.mark.asyncio
+async def test_info_endpoint_connection_error(mocker):
+    """Test the info endpoint handler."""
+    mock_authorization_resolvers(mocker)
+
+    # configuration for tests
+    config_dict = {
+        "name": "foo",
+        "service": {
+            "host": "localhost",
+            "port": 8080,
+            "auth_enabled": False,
+            "workers": 1,
+            "color_log": True,
+            "access_log": True,
+        },
+        "llama_stack": {
+            "api_key": "xyzzy",
+            "url": "http://x.y.com:1234",
+            "use_as_library_client": False,
+        },
+        "user_data_collection": {
+            "feedback_enabled": False,
+        },
+        "customization": None,
+        "authorization": {"access_rules": []},
+        "authentication": {"module": "noop"},
+    }
+    cfg = AppConfig()
+    cfg.init_from_dict(config_dict)
+
+    # Mock the LlamaStack client
+    mock_client = mocker.AsyncMock()
+    mock_client.inspect.version.side_effect = APIConnectionError(request=None)
+    mock_lsc = mocker.patch("client.AsyncLlamaStackClientHolder.get_client")
+    mock_lsc.return_value = mock_client
+    mock_config = mocker.Mock()
+    mocker.patch("app.endpoints.models.configuration", mock_config)
+
+    # Mock configuration
+    mocker.patch("configuration.configuration", cfg)
+
+    mock_authorization_resolvers(mocker)
+
+    request = Request(
+        scope={
+            "type": "http",
+        }
+    )
+    auth = ("test_user", "token", {})
+
+    with pytest.raises(HTTPException) as e:
+        await info_endpoint_handler(auth=auth, request=request)
+        assert e.value.status_code == status.HTTP_500_INTERNAL_SERVER_ERROR
+        assert e.detail["response"] == "Unable to connect to Llama Stack"