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
[misc] log info messages by default for hanging / busy / idle (#25627)
Signed-off-by: youkaichao <youkaichao@gmail.com>
youkaichao committed
Sep 25, 2025 at 21:14 UTC
6c340da4df81487e1fdbfd78eb87f55c9023b9b8
1 file changed
+6
-4
vllm/distributed/device_communicators/shm_broadcast.py
+6
-4
@@ -390,10 +390,11 @@ class MessageQueue:
390
# if we wait for a long time, log a message
391
if (time.monotonic() - start_time
392
> VLLM_RINGBUFFER_WARNING_INTERVAL * n_warning):
393
- logger.debug(
393
+ logger.info(
394
("No available shared memory broadcast block found"
395
" in %s seconds. This typically happens when some"
396
- " processes are hanging."),
396
+ " processes are hanging, doing some time-consuming"
397
+ " work (e.g. compilation), or sitting idle."),
398
VLLM_RINGBUFFER_WARNING_INTERVAL,
399
)
400
n_warning += 1
@@ -454,10 +455,11 @@ class MessageQueue:
455
# if we wait for a long time, log a message
456
if (time.monotonic() - start_time
457
> VLLM_RINGBUFFER_WARNING_INTERVAL * n_warning):
457
- logger.debug(
458
+ logger.info(
459
("No available shared memory broadcast block found"
460
" in %s seconds. This typically happens when some"
460
- " processes are hanging."),
461
+ " processes are hanging, doing some time-consuming"
462
+ " work (e.g. compilation), or sitting idle."),
463
VLLM_RINGBUFFER_WARNING_INTERVAL,
464
)
465
n_warning += 1