[router] Fix flaky router e2e tests (#13306)

sgl-router/py_test/integration_mock/test_circuit_breaker.py

@@ -112,6 +112,9 @@ def test_circuit_breaker_disable_flag(router_manager, mock_workers):
             "disable_retries": True,
         },
     )
+
+    saw_500 = False
+    for _ in range(8):
         r = requests.post(
             f"{rh.url}/v1/completions",
             json={
@@ -122,7 +125,15 @@ def test_circuit_breaker_disable_flag(router_manager, mock_workers):
             },
             timeout=3,
         )
-    assert r.status_code == 500
+        if r.status_code == 500:
+            # Worker starts, continue to check
+            saw_500 = True
+            break
+        assert (
+            r.status_code == 503
+        ), "Should only see 503 when waiting for worker to start"
+
+    assert saw_500
 
 
 @pytest.mark.integration

sgl-router/py_test/integration_mock/test_retries.py

@@ -47,6 +47,8 @@ def test_disable_retries_surfaces_failure(router_manager, mock_workers):
         },
     )
 
+    saw_500 = False
+    for _ in range(8):
         r = requests.post(
             f"{rh.url}/v1/completions",
             json={
@@ -57,5 +59,13 @@ def test_disable_retries_surfaces_failure(router_manager, mock_workers):
             },
             timeout=5,
         )
-    assert r.status_code == 500
+        if r.status_code == 500:
+            # Worker starts, continue to check
+            saw_500 = True
+            break
+        assert (
+            r.status_code == 503
+        ), "Should only see 503 when waiting for worker to start"
+
+    assert saw_500
     # mock_workers fixture handles cleanup