Merge pull request #39804 from BerriAI/litellm_/e2e-wildcard-probe-and-embedding-allowlist

test(e2e): repair the wildcard readiness probe and the semantic auto-router spend assertion

tests/e2e/models.py

@@ -851,6 +851,15 @@ class ModelListEntry(BaseModel):
     id: str
 
 
+class ModelsListParams(BaseModel):
+    """Query for GET /v1/models. A wildcard route such as ``openai/gpt-5.4*`` is
+    listed only under ``return_wildcard_routes``; without it the route is dropped
+    and only its expansions remain, so a readiness poll for the pattern itself
+    never resolves."""
+
+    return_wildcard_routes: bool = True
+
+
 class ModelsListResponse(BaseModel):
     """GET /v1/models on the data plane: the deployments the gateway can actually
     serve right now. Used to confirm a freshly created model has propagated from

tests/e2e/proxy_client.py

@@ -55,6 +55,7 @@ from models import (
     ModelMode,
     ModelNewBody,
     ModelNewResponse,
+    ModelsListParams,
     ModelsListResponse,
     ModelUpdateBody,
     OcrBody,
@@ -336,7 +337,7 @@ class ProxyClient:
             lambda poll_timeout: self.transport.get(
                 "/v1/models",
                 headers=headers,
-                params=NoBody(),
+                params=ModelsListParams(),
                 response_type=ModelsListResponse,
                 timeout=poll_timeout,
             ),

tests/e2e/router/test_auto_router_regressions_e2e.py

@@ -597,9 +597,20 @@ class TestSemanticAutoRouterResponses:
             )
         )
         assert answer.id, "/v1/responses through the semantic auto-router returned no response id"
-        rows: Final = proxy.poll_logs_for_key(key, min_rows=1)
+        rows: Final = proxy.poll_logs_for_key(
+            key,
+            min_rows=2,
+            predicate=lambda logged: any(row.model == EMBEDDING_MODEL for row in logged),
+        )
+        embedding_rows: Final = tuple(row for row in rows if row.model == EMBEDDING_MODEL)
+        assert embedding_rows, (
+            "the routing embedding was not billed to the caller's key; "
+            f"spend logs show {tuple(row.model for row in rows)}"
+        )
         _assert_served_only_by(
-            rows, CHEAP_SERVED | {semantic_auto_router.target}, "semantic auto-router /v1/responses string input"
+            [row for row in rows if row.model != EMBEDDING_MODEL],
+            CHEAP_SERVED | {semantic_auto_router.target},
+            "semantic auto-router /v1/responses string input",
         )