Fix PD bootstrap failure handling (#24772)

Co-authored-by: Cheng Wan <chwan@rice.edu>
2026-05-10 19:02:47 +08:00
parent 8cc16c9974
commit bd0aa22309
1 changed files with 3 additions and 0 deletions
@@ -608,6 +608,8 @@ class CommonKVReceiver(BaseKVReceiver):

        self.prefill_dp_rank = prefill_dp_rank
        self._setup_bootstrap_infos()
+        if self.conclude_state == KVPoll.Failed:
+            return
        self.kv_mgr.update_status(self.bootstrap_room, KVPoll.WaitingForInput)

    def _setup_bootstrap_infos(self):
@@ -650,6 +652,7 @@ class CommonKVReceiver(BaseKVReceiver):
                            self.kv_mgr.update_status(
                                self.bootstrap_room, KVPoll.Failed
                            )
+                            self.bootstrap_infos = None
                            return

                self.bootstrap_infos = bootstrap_infos