make async timeline failure log errors

PiperOrigin-RevId: 889305628

Author: s-noghabi

tests/perf/experimental/timeline_test.py

@@ -234,8 +234,19 @@ def fail_wait(waitlist, success, failure):
 
     self.mock_async_wait.side_effect = fail_wait
 
-    with self.assertRaisesRegex(RuntimeError, "failed"):
-      t.span("failed", 1.0, ["wait"])
+    with mock.patch.object(timeline.logging, "error") as mock_log_error:
+      t.span("failed_op", 1.0, ["wait"])
+      # Exception is caught and logged, no exception is raised to the caller.
+      mock_log_error.assert_called_once()
+      args, kwargs = mock_log_error.call_args
+      format_str, name, span_id, err = args
+      self.assertEqual(format_str, "Timeline span '%s' (id=%d) failed: %s")
+      self.assertEqual(name, "failed_op")
+      self.assertEqual(span_id, 0)
+      self.assertIsInstance(err, RuntimeError)
+      self.assertEqual(str(err), "failed")
+      self.assertIn("exc_info", kwargs)
+      self.assertIsInstance(kwargs["exc_info"], RuntimeError)
 
   def test_wait_pending_spans_clears_threads(self):
     t = timeline.AsyncTimeline("test_tl", 0.0)

tunix/perf/experimental/timeline.py

@@ -262,7 +262,10 @@ def on_success() -> None:
     def on_failure(e: Exception) -> None:
       # TODO(noghabi):Capture the span even if it fails, but add a tag that it
       # failed and process accordingly.
-      raise e
+      # Metrics are best effort and should not raise exceptions.
+      logging.error(
+          "Timeline span '%s' (id=%d) failed: %s", name, span_id, e, exc_info=e
+      )
 
     if not waitlist:
       on_success()