On 06-02-2020 22:40, Dean Schulze wrote:
I've moved two nodes to a different controller. The nodes are wired and
the controller is networked via wifi. I had to open up ports 6817 and
6818 between the wired and wireless sides of our network to get any
connectivity.
Now when I do
srun -N2 hostname
the jobs show connection timeouts on the nodes:
[2020-02-06T14:24:37.183] launch task 60.0 request from UID:1000
GID:1000 HOST:10.204.18.232 PORT:19602
[2020-02-06T14:24:37.183] lllp_distribution jobid [60] implicit auto
binding: cores, dist 8192
[2020-02-06T14:24:37.183] _task_layout_lllp_cyclic
[2020-02-06T14:24:37.183] _lllp_generate_cpu_bind jobid [60]: mask_cpu,
0x0101
[2020-02-06T14:24:37.184] _run_prolog: run job script took usec=6
[2020-02-06T14:24:37.184] _run_prolog: prolog with lock for job 60 ran
for 0 seconds
[2020-02-06T14:24:45.224] [60.0] error: connect io: Connection timed out
[2020-02-06T14:24:45.224] [60.0] error: IO setup failed: Connection
timed out
[2020-02-06T14:24:45.225] [60.0] error: job_manager exiting abnormally,
rc = 4021
[2020-02-06T14:24:59.538] [60.0] error: _send_launch_resp: Failed to
send RESPONSE_LAUNCH_TASKS: Connection timed out
[2020-02-06T14:24:59.551] [60.0] done with job
That node used port 19602 and the other node was using port 12496. When
I did the srun again the jobs showed two different ports on the nodes
(58040 and 32392).
How can I configure a network if srun is going to grab different ports
each time?
Perhaps the information about firewall setup in my Wiki page can be of use:
https://wiki.fysik.dtu.dk/niflheim/Slurm_configuration#configure-firewall-for-slurm-daemons
/Ole
