Merge remote-tracking branch 'origin/feat/vLLM-GPU-runners' into feat/vLLM-GPU-runners

Author: skamenan7

.github/workflows/pypi.yml

@@ -1025,6 +1025,21 @@ jobs:
           echo "::error::${PACKAGE_NAME}==${VERSION} did not appear on test PyPI after 10 minutes"
           exit 1
 
+      - name: Generate config labels
+        id: labels
+        run: |
+          # Embed distribution configs as OCI labels (see scripts/generate-config-labels.sh).
+          # generate-config-labels.sh emits alternating "--label"/"key=value" lines for
+          # docker CLI; build-push-action wants bare "key=value" lines, so drop the flags.
+          LABELS=$(bash scripts/generate-config-labels.sh \
+            "${{ matrix.distro }}" "${{ needs.compute-version.outputs.version }}" \
+            | grep -v '^--label$')
+          {
+            echo "labels<<EOF"
+            echo "$LABELS"
+            echo "EOF"
+          } >> "$GITHUB_OUTPUT"
+
       - name: Build and push Docker image
         uses: docker/build-push-action@f9f3042f7e2789586610d6e8b85c8f03e5195baf # v7.2.0
         with:
@@ -1033,6 +1048,7 @@ jobs:
           platforms: ${{ matrix.platforms }}
           push: true
           tags: ${{ steps.meta.outputs.tags }}
+          labels: ${{ steps.labels.outputs.labels }}
           build-args: |
             DISTRO_NAME=${{ matrix.distro }}
             INSTALL_MODE=${{ steps.meta.outputs.install_mode }}

src/ogx/providers/remote/vector_io/qdrant/qdrant.py

@@ -133,6 +133,12 @@ async def query_vector(
         if filters is not None:
             raise NotImplementedError("Qdrant provider does not yet support native filtering")
 
+        # Collections are created lazily on first insert, so a search against a
+        # store that has never had chunks added has no collection yet. Treat that
+        # as an empty result rather than letting Qdrant raise a 404.
+        if not await self.client.collection_exists(self.collection_name):
+            return QueryChunksResponse(chunks=[], scores=[])
+
         results = (
             await self.client.query_points(
                 collection_name=self.collection_name,
@@ -184,6 +190,12 @@ async def query_keyword(
         if filters is not None:
             raise NotImplementedError("Qdrant provider does not yet support native filtering")
 
+        # Collections are created lazily on first insert, so a search against a
+        # store that has never had chunks added has no collection yet. Treat that
+        # as an empty result rather than letting Qdrant raise a 404.
+        if not await self.client.collection_exists(self.collection_name):
+            return QueryChunksResponse(chunks=[], scores=[])
+
         try:
             # Use scroll for keyword-only search since query_points requires a query vector
             # Scroll allows filtering without a query vector
@@ -264,6 +276,12 @@ async def query_hybrid(
         if filters is not None:
             raise NotImplementedError("Qdrant provider does not yet support native filtering")
 
+        # Collections are created lazily on first insert, so a search against a
+        # store that has never had chunks added has no collection yet. Treat that
+        # as an empty result rather than letting Qdrant raise a 404.
+        if not await self.client.collection_exists(self.collection_name):
+            return QueryChunksResponse(chunks=[], scores=[])
+
         try:
             query_words = query_string.lower().split()
             if not query_words:

src/ogx/providers/utils/memory/openai_vector_store_mixin.py

@@ -1121,14 +1121,8 @@ async def openai_search_vector_store(
             )
 
         except Exception as e:
-            # Log the error and return empty results
-            logger.error("Error searching vector store", vector_store_id=vector_store_id, error=str(e))
-            return VectorStoreSearchResponsePage(
-                search_query=request.query if isinstance(request.query, list) else [request.query],
-                data=[],
-                has_more=False,
-                next_page=None,
-            )
+            logger.error("Failed to search vector store", vector_store_id=vector_store_id, error=str(e))
+            raise
 
     def _build_reranker_params(
         self,

tests/integration/vector_io/test_openai_vector_stores.py

@@ -3585,7 +3585,7 @@ def test_openai_vector_store_with_chunks(
     filtered_search = compat_client.vector_stores.search(
         vector_store_id=vector_store.id,
         query="artificial intelligence",
-        filters={"topic": "ai"},
+        filters={"type": "eq", "key": "topic", "value": "ai"},
         max_num_results=5,
     )
 

tests/unit/providers/vector_io/test_vector_io_stores_config.py

@@ -213,6 +213,32 @@ async def mock_query_chunks(*args, **kwargs):
     assert result.search_query == ["test query"]  # Original query preserved
 
 
+async def test_search_vector_store_propagates_backend_errors(vector_io_adapter):
+    """Test that exceptions from the vector store backend propagate to the caller."""
+    vector_store_id = "test_store_error"
+    vector_io_adapter.openai_vector_stores[vector_store_id] = {
+        "id": vector_store_id,
+        "name": "Test Store",
+        "description": "",
+        "vector_store_id": "test_db",
+        "embedding_model": "test/embedding",
+    }
+
+    async def mock_query_chunks(*args, **kwargs):
+        raise KeyError("chunk_content")
+
+    vector_io_adapter.query_chunks = mock_query_chunks
+
+    from ogx_api import OpenAISearchVectorStoreRequest
+
+    request = OpenAISearchVectorStoreRequest(query="test query", max_num_results=5)
+    with pytest.raises(KeyError, match="chunk_content"):
+        await vector_io_adapter.openai_search_vector_store(
+            vector_store_id=vector_store_id,
+            request=request,
+        )
+
+
 async def test_create_gin_index_executes_correct_sql():
     from ogx.providers.remote.vector_io.pgvector.config import PGVectorHNSWVectorIndex
     from ogx.providers.remote.vector_io.pgvector.pgvector import PGVectorIndex