[PD] Fix resource leak on prealloc/transfer abort and idle check (#28022)

python/sglang/srt/disaggregation/decode.py

@@ -815,6 +815,8 @@ class DecodePreallocQueue(DecodeHiCachePreallocMixin):
                     [decode_req.req],
                     decode_req.req.return_logprob,
                 )
+                decode_req.kv_receiver.clear()
+                decode_req.kv_receiver = None
                 failed_reqs.append(decode_req)
                 indices_to_remove.add(i)
 
@@ -1651,6 +1653,8 @@ class DecodeTransferQueue(DecodeHiCacheTransferMixin):
                     self.scheduler.hisparse_coordinator.request_finished(decode_req.req)
                 # release pre-allocated kv cache, but don't insert into the tree since it's failed
                 release_kv_cache(decode_req.req, self.tree_cache, is_insert=False)
+                decode_req.kv_receiver.clear()
+                decode_req.kv_receiver = None
                 indices_to_remove.add(i)
                 if self.scheduler.metrics_reporter.enable_metrics:
                     self.scheduler.metrics_collector.increment_transfer_failed_reqs()

python/sglang/srt/managers/scheduler.py

@@ -3380,6 +3380,7 @@ class Scheduler(
 
             if self.disaggregation_mode == DisaggregationMode.DECODE:
                 idle &= len(self.disagg_decode_prealloc_queue.queue) == 0
+                idle &= len(self.disagg_decode_prealloc_queue.retracted_queue) == 0
                 idle &= len(self.disagg_decode_transfer_queue.queue) == 0
                 if self.decode_offload_manager is not None:
                     idle &= len(self.decode_offload_manager.ongoing_offload) == 0