[PD] Abort request if transfer fails (#6504)

This commit is contained in:
Byron Hsu
2025-05-21 21:44:25 -07:00
committed by GitHub
parent 7513558074
commit 3bde101099
5 changed files with 84 additions and 4 deletions

View File

@@ -38,6 +38,7 @@ from sglang.srt.disaggregation.utils import (
kv_to_page_indices,
kv_to_page_num,
poll_and_all_reduce,
prepare_abort,
)
from sglang.srt.managers.schedule_batch import FINISH_LENGTH, Req, ScheduleBatch
@@ -157,7 +158,18 @@ class PrefillBootstrapQueue:
if poll == KVPoll.Bootstrapping:
continue
elif poll == KVPoll.Failed:
raise Exception("Bootstrap failed")
error_message = f"Prefill bootstrap failed for request rank={self.tp_rank} {req.rid=} {req.bootstrap_room=}"
try:
req.disagg_kv_sender.failure_exception()
except Exception as e:
error_message += f" with exception {e}"
logger.error(error_message)
prepare_abort(
req, error_message, status_code=HTTPStatus.INTERNAL_SERVER_ERROR
)
self.scheduler.stream_output([req], req.return_logprob)
indices_to_remove.add(i)
continue
# KV.WaitingForInput
num_kv_indices = len(req.origin_input_ids)
@@ -335,7 +347,17 @@ class SchedulerDisaggregationPrefillMixin:
# FIXME: clean up req's data in transfer engine
done_reqs.append(req)
elif poll == KVPoll.Failed:
raise Exception("Transferring failed")
error_message = f"Prefill transfer failed for request rank={self.tp_rank} {req.rid=} {req.bootstrap_room=}"
try:
req.disagg_kv_sender.failure_exception()
except Exception as e:
error_message += f" with exception {e}"
logger.warning(error_message)
self.tree_cache.cache_finished_req(req) # unlock the tree
prepare_abort(
req, error_message, status_code=HTTPStatus.INTERNAL_SERVER_ERROR
)
done_reqs.append(req)
for req in done_reqs:
self.disagg_prefill_bootstrap_queue.req_to_metadata_buffer_idx_allocator.free(