fix: e2e tests and deployment

.dockerignore

@@ -2,6 +2,8 @@
 
 !opengaterag/
 !opengaterag/api/**
+!scripts/
+!scripts/**
 !.github/
 !pyproject.toml
 !config.example.yml

.github/.trivyignore

@@ -0,0 +1,9 @@
+# perl-archive-tar: perl-archive-tar: Path traversal via 
+# crafted symlinks allows arbitrary file access
+# https://avd.aquasec.com/nvd/cve-2026-42496
+CVE-2026-42496 exp:2026-09-01
+
+# perl: Perl: Heap buffer overflow when compiling regular 
+# expressions on 32-bit builds...
+# https://avd.aquasec.com/nvd/cve-2026-8376
+CVE-2026-8376 exp:2026-09-01

.github/badges/coverage.json

@@ -1 +1 @@
-{"schemaVersion":1,"label":"coverage","message":"55.92%","color":"red"}
+{"schemaVersion":1,"label":"coverage","message":"73.96%","color":"red"}

.github/tests/compose.yml

@@ -33,7 +33,7 @@ services:
     environment:
       - ALBERT_ADMIN_API_KEY=${ALBERT_ADMIN_API_KEY}
     volumes:
-      - ".github/tests/config.yml:/config.yml:ro"
+      - "./config.opengatellm.yml:/config.yml:ro"
     healthcheck:
         test: [ "CMD-SHELL", "curl -sf http://localhost:8000/health || exit 1" ]
         interval: 4s

.github/tests/config.yml → .github/tests/config.opengatellm.yml

@@ -1,7 +1,7 @@
 # -------------------------------- dependencies ---------------------------------
 dependencies:
   postgres:
-    url: postgresql+asyncpg://postgres:changeme@localhost:5432/postgres
+    url: postgresql+asyncpg://postgres:changeme@postgres:5432/postgres
     echo: False
     pool_size: 10
     max_overflow: 5
@@ -14,7 +14,7 @@ dependencies:
       command_timeout: 60
 
   redis:
-    url: redis://:changeme@localhost:6379
+    url: redis://:changeme@redis:6379
     max_connections: 200
     socket_connect_timeout: 5
     retry_on_timeout: True
@@ -27,25 +27,24 @@ dependencies:
     number_of_shards: 1
     number_of_replicas: 1
     refresh_interval: 1s
-    hosts: http://localhost:9200
+    hosts: http://elasticsearch:9200
     basic_auth:
-      - elasticsearch
+      - elastic
       - changeme
 
 # ---------------------------------- settings -----------------------------------
 settings:
-  vector_store_model: BAAI/bge-m3
+  vector_store_model: openweight-embeddings
 
 # ----------------------------------- models ------------------------------------
 models:  
-  - name: BAAI/bge-m3
+  - name: openweight-embeddings
     type: text-embeddings-inference
-    aliases: ["openweight-embedding"]
     cost_prompt_tokens: 10
     cost_completion_tokens: 30
     providers:
       - type: albert
-        key: ${ALBERT_API_ADMIN_KEY}
+        key: ${ALBERT_ADMIN_API_KEY}
         timeout: 120
         model_name: BAAI/bge-m3
         model_total_params: 8

.github/tests/e2e/conftest.py

@@ -12,6 +12,25 @@
 from opengaterag.api.utils.configuration import configuration
 
 
+def refresh_elasticsearch_index() -> None:
+    """Force Elasticsearch to make recently indexed chunks visible to search."""
+
+    async def _refresh() -> None:
+        kwargs = configuration.dependencies.elasticsearch.model_dump()
+        index_name = kwargs.pop("index_name")
+        kwargs.pop("index_language")
+        kwargs.pop("number_of_shards")
+        kwargs.pop("number_of_replicas")
+        kwargs.pop("refresh_interval")
+        client = AsyncElasticsearch(**kwargs)
+        try:
+            await client.indices.refresh(index=index_name)
+        finally:
+            await client.close()
+
+    asyncio.run(_refresh())
+
+
 class AuthenticatedTestClient:
     """Wrap a shared TestClient and inject auth headers per request."""
 
@@ -98,3 +117,15 @@ def admin_client(test_client: TestClient) -> AuthenticatedTestClient:
     """Test client authenticated as an admin user."""
     api_key = _validate_opengatellm_api_key("OPENGATELLM_ADMIN_API_KEY", "admin")
     return AuthenticatedTestClient(client=test_client, api_key=api_key)
+
+
+@pytest.fixture(scope="function")
+def es_refresh():
+    """
+    Callable fixture: call `es_refresh()` after writes so searches see fresh data.
+
+    We return the callable (instead of auto-refresh) so tests can refresh exactly
+    after indexing (documents/chunks) and avoid timing-based sleeps.
+    """
+
+    return refresh_elasticsearch_index