A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
[NIXL] Improve request_finished() debug logs (#25665)
Signed-off-by: Mark McLoughlin <markmc@redhat.com>
Mark McLoughlin committed
Oct 16, 2025 at 14:55 UTC
4a510ab487a4c66dccf1fb642215d463c2753779
1 file changed
+9
-1
vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py
+9
-1
@@ -463,7 +463,9 @@ class NixlConnectorScheduler:
463
464
params = request.kv_transfer_params
465
logger.debug(
466
- "NIXLConnector request_finished, request_status=%s, kv_transfer_params=%s",
466
+ "NIXLConnector request_finished(%s), request_status=%s, "
467
+ "kv_transfer_params=%s",
468
+ request.request_id,
469
request.status,
470
params,
471
)
@@ -495,6 +497,12 @@ class NixlConnectorScheduler:
497
498
if delay_free_blocks:
499
# Prefill request on remote. It will be read from D upon completion
500
+ logger.debug(
501
+ "NIXLConnector request_finished(%s) waiting for %d seconds "
502
+ "for remote decode to fetch blocks",
503
+ request.request_id,
504
+ envs.VLLM_NIXL_ABORT_REQUEST_TIMEOUT,
505
+ )
506
self._reqs_need_send[request.request_id] = (
507
time.perf_counter() + envs.VLLM_NIXL_ABORT_REQUEST_TIMEOUT
508
)