Hi -
Very delayed response to this, as I'm working my way through a backlog
of slurm-user posts. If this error is intermittent, it's likely a
hardware issue. Recently I ran into an problem where a host with 8 GPUs
was spontaneously rebooting a couple of minutes after a user would start
an 8 GPU process. The same task on the same machine with 7 GPUs worked
fine. It turned out that one of the power supplies was partially
unseated, and this was causing the problem. Once the power supply was
properly re-installed, the problem went away.
On 2/22/23 01:38, Eunsong Goh wrote:
Hi,
I'm using slurm cluster with one master node and two worker node.
When I run sbatch job with SBATCH parameter shell file, job allocated
each node.
With nvidia-smi command, I figure out job was appropriately allocated
two node.
But mean a while, one node job suddenly killed and occured error
mesage in /var/log/slurm/slurmd.log as below
After that, job continued remainded one node.
What is the problem?
[2023-02-22T07:12:54.214] pyxis: version v0.11.1
[2023-02-22T07:12:54.215] slurmd version 22.05.2 started
[2023-02-22T07:12:54.226] slurmd started on Wed, 22 Feb 2023 07:12:54 +0000
[2023-02-22T07:12:54.299] CPUs=40 Boards=1 Sockets=2 Cores=10 Threads=2
Memory=257609 TmpDisk=100220 Uptime=104407 CPUSpecList=(null)
FeaturesAvail=(null) FeaturesActive=(null)
[2023-02-22T07:13:53.613] error: bad node index (-1 > 1)
[2023-02-22T07:14:00.966] epilog for job 1056 ran for 7 seconds
[2023-02-22T07:16:08.691] task/affinity: task_p_slurmd_batch_request:
task_p_slurmd_batch_request: 127
[2023-02-22T07:16:08.692] task/affinity: batch_bind: job 127 CPU input
mask for node: 0x0000000003
[2023-02-22T07:16:08.692] task/affinity: batch_bind: job 127 CPU final
HW mask for node: 0x0000100001
[2023-02-22T07:16:08.951] [127.extern] task/cgroup: _memcg_initialize:
job: alloc=0MB mem.limit=257609MB memsw.limit=unlimited
[2023-02-22T07:16:08.951] [127.extern] task/cgroup: _memcg_initialize:
step: alloc=0MB mem.limit=257609MB memsw.limit=unlimited
[2023-02-22T07:16:08.956] Launching batch job 127 for UID 1002
[2023-02-22T07:16:08.976] [127.batch] task/cgroup: _memcg_initialize:
job: alloc=0MB mem.limit=257609MB memsw.limit=unlimited
[2023-02-22T07:16:08.976] [127.batch] task/cgroup: _memcg_initialize:
step: alloc=0MB mem.limit=257609MB memsw.limit=unlimited
[2023-02-22T07:16:09.036] launch task StepId=127.0 request from UID:1002
GID:1003 HOST:27.122.137.19 PORT:32996
[2023-02-22T07:16:09.036] task/affinity: lllp_distribution: JobId=127
implicit auto binding: cores, dist 8192
[2023-02-22T07:16:09.036] task/affinity: _task_layout_lllp_block:
_task_layout_lllp_block
[2023-02-22T07:16:09.036] task/affinity: _lllp_generate_cpu_bind:
_lllp_generate_cpu_bind jobid [127]: mask_cpu, 0x0000100001
[2023-02-22T07:16:09.148] [127.0] task/cgroup: _memcg_initialize: job:
alloc=0MB mem.limit=257609MB memsw.limit=unlimited
[2023-02-22T07:16:09.149] [127.0] task/cgroup: _memcg_initialize: step:
alloc=0MB mem.limit=257609MB memsw.limit=unlimited
[2023-02-22T07:16:53.880] error: bad node index (-1 > 1)
[2023-02-22T07:16:55.569] [127.0] get_exit_code task 0 died by signal: 9
[2023-02-22T07:16:55.578] [127.0] done with job
[2023-02-22T07:17:01.412] epilog for job 1056 ran for 8 seconds
[2023-02-22T07:19:35.335] [127.batch] done with job
[2023-02-22T07:19:35.350] [127.extern] done with job
[2023-02-22T07:19:53.067] error: bad node index (-1 > 1)
[2023-02-22T07:20:00.154] epilog for job 1056 ran for 7 seconds
[2023-02-22T07:22:53.294] error: bad node index (-1 > 1)
[2023-02-22T07:23:00.463] epilog for job 1056 ran for 7 seconds
[2023-02-22T07:25:53.485] error: bad node index (-1 > 1)
[2023-02-22T07:26:00.612] epilog for job 1056 ran for 7 seconds
[2023-02-22T07:28:53.709] error: bad node index (-1 > 1)
[2023-02-22T07:29:00.841] epilog for job 1056 ran for 7 seconds
[2023-02-22T07:31:53.975] error: bad node index (-1 > 1)
[2023-02-22T07:32:01.083] epilog for job 1056 ran for 8 seconds
[2023-02-22T07:34:53.165] error: bad node index (-1 > 1)
This message is from an external sender. Learn more about why this
matters.
<https://ut.service-now.com/sp?id=kb_article&number=KB0011401>
