Merge 964c4a0884 into 6fa0477d84

litellm/router_rust/invoke/endpoints.py

@@ -358,9 +358,7 @@ PLANS: Final[Mapping[str, Plan]] = MappingProxyType(
             sync=True,
             splits_stream_timeout=True,
             stamp="deployment",
-            total_key=None,
-            success_key=None,
-            fail_key=None,
+            # The sync path updates the same deployment counters as ``acompletion``.
             named=("messages",),
         ),
         "atext_completion": Plan(

tests/unit/test_router/core/test_gaps_l2800.py

@@ -424,8 +424,17 @@ async def test_acompletion_counts_total_success_and_failed_calls_per_deployment_
     assert counts == (2, 1, 1)
 
 
-def test_sync_completion_does_not_count_calls(make_core: MakeCore, host: ScriptedHost) -> None:
-    """Oddity: only the async path updates ``total_calls``/``success_calls``/``fail_calls``."""
-    core: Final = make_core(model_list=[deployment("g", "d1")])
+def test_sync_completion_counts_total_success_and_failed_calls_per_deployment_model(
+    make_core: MakeCore, host: ScriptedHost
+) -> None:
+    core: Final = make_core(model_list=[deployment("g", "d1")], num_retries=0)
     core.completion(model="g", messages=MESSAGES)  # pyright: ignore[reportAttributeAccessIssue]
-    assert core.total_calls["openai/d1"] == 0  # pyright: ignore[reportAttributeAccessIssue]
+    host.answer("openai/d1", rate_limit())
+    with pytest.raises(litellm.RateLimitError):
+        core.completion(model="g", messages=MESSAGES)  # pyright: ignore[reportAttributeAccessIssue]
+    counts: Final = (
+        core.total_calls["openai/d1"],  # pyright: ignore[reportAttributeAccessIssue]
+        core.success_calls["openai/d1"],  # pyright: ignore[reportAttributeAccessIssue]
+        core.fail_calls["openai/d1"],  # pyright: ignore[reportAttributeAccessIssue]
+    )
+    assert counts == (2, 1, 1)

tests/unit/test_router/test_router.py

@@ -19900,3 +19900,40 @@ def test_router_warns_when_a_deployment_enforces_endpoints_without_a_list(caplog
 
     assert "Deployment 'embed-from-config' sets enforce_supported_endpoints without" in caplog.text
     assert "Deployment 'embed-from-api' sets enforce_supported_endpoints without" in caplog.text
+
+
+def test_sync_completion_tracks_router_metrics():
+    """
+    Regression: sync router.completion() must update router.total_calls /
+    router.success_calls / router.fail_calls, same as router.acompletion() does.
+    """
+    router = Router(
+        model_list=[
+            {
+                "model_name": "gpt-4o",
+                "litellm_params": {"model": "openai/gpt-4o", "num_retries": 0},
+            }
+        ],
+        num_retries=0,
+    )
+
+    ## success: total/success track the deployment model
+    router.completion(
+        model="gpt-4o",
+        messages=[{"role": "user", "content": "hi"}],
+        mock_response="Hello",
+    )
+    assert router.total_calls["openai/gpt-4o"] == 1
+    assert router.success_calls["openai/gpt-4o"] == 1
+    assert router.fail_calls.get("openai/gpt-4o", 0) == 0
+
+    ## failure: fail tracks the deployment model
+    with pytest.raises(Exception):
+        router.completion(
+            model="gpt-4o",
+            messages=[{"role": "user", "content": "hi"}],
+            mock_response="litellm.InternalServerError",
+        )
+    assert router.fail_calls["openai/gpt-4o"] == 1
+    assert router.total_calls["openai/gpt-4o"] == 2
+    assert router.success_calls["openai/gpt-4o"] == 1