Node states are unknown

[percyfal]

Hi,

I'm using docker-centos7-slurm to test a workflow manager. It has been a while since updating, but when trying out the most recent version, I notice that only one node (c1) is up in the container. I am currently testing this in my fork (see pr #1). Briefly, I parametrized test_job_can_run to pass partition to the --partition option. The normal partition works as expected, but debug fails.

If one enters the latest image with

docker run -it -h slurmctl giovtorres/docker-centos7-slurm:latest

running sinfo yields

NODELIST   NODES PARTITION       STATE CPUS    S:C:T MEMORY TMP_DISK WEIGHT AVAIL_FE REASON              
c1             1   normal*        idle 1       1:1:1   1000        0      1   (null) none                
c2             1   normal*    unknown* 1       1:1:1   1000        0      1   (null) none                
c3             1     debug    unknown* 1       1:1:1   1000        0      1   (null) none                
c4             1     debug    unknown* 1       1:1:1   1000        0      1   (null) none                

See github action resuts, where I added some print statements to see what was going on (nevermind that the test actually passed; I was simply looking at the erroneous slurm output file). I consistently get the feedback that the required nodes are not available; it would seem node c1 is the only node available to sbatch.

Are you able to reproduce this?

Cheers,

Per

[giovtorres]

Yes, I can reproduce in 21.08, but 20.11.8 appears to be working. Something related to networking changed in Slurm between those releases and the slurm.conf probably is missing something to accommodate. I saw errors about NodeHostName and NodeAddr being the same for all 4 nodes.

Help is appreciated 😄

[percyfal]

So I actually found a solution; removing redundant NodeHostName statements in slurm.conf seems to work:

NodeName=c1 NodeHostName=slurmctl NodeAddr=127.0.0.1 RealMemory=1000
NodeName=c2 NodeAddr=127.0.0.1 RealMemory=1000
NodeName=c3 NodeAddr=127.0.0.1 RealMemory=1000 Gres=gpu:titanxp:1
NodeName=c4 NodeAddr=127.0.0.1 RealMemory=1000 Gres=gpu:titanxp:1

Not that I could find anything in the docs that would suggest this solution; I basically reacted to this error in the logs:

cat /var/log/slurm/slurmctld.log | grep Duplicated
[2022-02-17T08:33:22.239] error: Duplicated NodeHostName slurmctl in the config file
[2022-02-17T08:33:22.239] error: Duplicated NodeHostName slurmctl in the config file
[2022-02-17T08:33:22.239] error: Duplicated NodeHostName slurmctl in the config file

I'm currently running the test on my fork (see changeset). I'll submit a pr on success.

[giovtorres]

Fixed in #38