[Fix] Include disagg prefill waiting queue in FPM (#32122)

Signed-off-by: hongkuanz <hongkuanz@nvidia.com>
2026-07-23 09:30:27 -07:00
parent b98a577fbe
commit d0b9689805
2 changed files with 5 additions and 2 deletions
@@ -232,11 +232,12 @@ class TestForwardPassMetrics(unittest.TestCase):
            self.scheduler._fpm_publisher.metrics[0].wall_time, 0.035, places=4
        )

-    def test_disagg_prefill_queued_metrics(self):
+    def test_disagg_prefill_queued_metrics_include_compute_waiting_queue(self):
        self.scheduler.disaggregation_mode = DisaggregationMode.PREFILL
        self.scheduler.disagg_prefill_bootstrap_queue = types.SimpleNamespace(
-            queue=[_FakeReq(100), _FakeReq(200), _FakeReq(50)],
+            queue=[_FakeReq(100)],
        )
+        self.scheduler.waiting_queue = [_FakeReq(200), _FakeReq(50)]
        batch = self._make_batch()

        with patch(