Status of slurm
sinfo sinfo -R squeue
we have draining nodes, but no jobs running on them
Reviving draining node (as root)
scontrol update NodeName=octopus05 State=DOWN Reason="undraining" update NodeName=octopus05 State=RESUME show node octopus05
Kill time can lead to drain state
scontrol show config | grep kill UnkillableStepProgram = (null) UnkillableStepTimeout = 60 sec
check valid configuration with `slurmd -C` and update nodes with
scontrol reconfigure
So we create a script that can deploy files from octopus01 (head node). Unfortunately ids in passwd do no match, so we can't copy those yet.
See /etc/nodes for script and ssh files, sudoers (etc)
Basically the root user can copy across.
To avoid `./scratch/script.sh: Permission denied` on `device_file`:
- `sudo bash` - `ls /scratch -l` to check where `/scratch` is - `vim /etc/fstab` - replace `noexec` with `exec` for `device_file` - `mount -o remount [device_file]` to remount the partition with its new configuration.
Some notes:
root@tux09:~# mkdir -p /var/lib/nfs/statd root@tux09:~# systemctl enable rpcbind Synchronizing state of rpcbind.service with SysV service script with /lib/systemd/systemd-sysv-install. Executing: /lib/systemd/systemd-sysv-install enable rpcbind root@tux09:~# systemctl list-unit-files | grep -E 'rpc-statd.service' rpc-statd.service static -
network-online.target x-systemd.device-timeout= 10.0.0.110:/export/3T /mnt/3T nfs nofail,x-systemd.automount,x-systemd.requires=network-online.target,x-systemd.device-timeout=10 0 0
Current nodes in the pool have:
munge --version
munge-0.5.13 (2017-09-26)
sbatch --version
slurm-wlm 18.08.5-2
To install `munge`, go to `octopus01` and run:
guix package -i munge@0.5.14 -p /export/octopus01/guix-profiles/slurm systemctl status munge # to check if the service is running and where its service file is
We need to setup the rights for `munge`:
sudo bash addgroup -gid 900 munge adduser -uid 900 -gid 900 --disabled-password munge sed 's,/home/munge:/bin/bash,/var/lib/munge:/usr/sbin/nologin,g' /etc/passwd -i mkdir -p /var/lib/munge chown munge:munge /var/lib/munge/ mkdir -p /etc/munge # copy `munge.key` (from a working node) to `/etc/munge/munge.key` chown -R munge:munge /etc/munge mkdir -p /run/munge chown munge:munge /run/munge mkdir -p /var/log/munge chown munge:munge /var/log/munge mkdir -p /var/run/munge # todo: not sure why it needs such a folder chown munge:munge /var/run/munge # copy `munge.service` (from a working node) to `/etc/systemd/system/munge.service` systemctl daemon-reload systemctl enable munge systemctl start munge systemctl status munge
To test the new installation, go to `octopus01` and then:
munge -n | ssh tux08 /export/octopus01/guix-profiles/slurm-2-link/bin/unmunge
If you get `STATUS: Rewound credential (16)`, it means that there is a difference between the encoding and decoding times. To fix it, go into the new machine and fix the time with
sudo date MMDDhhmmYYYY.ss
To install `slurm`, go to `octopus01` and run:
guix package -i slurm@18.08.9 -p /export/octopus01/guix-profiles/slurm
We need to setup the rights for `slurm`:
sudo bash addgroup -gid 901 slurm adduser -uid 901 -gid 901 --no-create-home --disabled-password slurm sed 's,/home/slurm:/bin/bash,/var/lib/slurm:/bin/bash,g' /etc/passwd -i mkdir -p /var/lib/slurm chown munge:munge /var/lib/slurm/ mkdir -p /etc/slurm # copy `slurm.conf` to `/etc/slurm/slurm.conf` # copy `cgroup.conf` to `/etc/slurm/cgroup.conf` chown -R slurm:slurm /etc/slurm mkdir -p /run/slurm chown slurm:slurm /run/slurm mkdir -p /var/log/slurm chown slurm:slurm /var/log/slurm # copy `slurm.service` to `/etc/systemd/system/slurm.service` /export/octopus01/guix-profiles/slurm-2-link/sbin/slurmd -f /etc/slurm/slurm.conf -C | head -n 1 >> /etc/slurm/slurm.conf # add node configuration information systemctl daemon-reload systemctl enable slurm systemctl start slurm systemctl status slurm
On `octopus01` (the master):
sudo bash # add the new node to `/etc/slurm/slurm.conf` systemctl restart slurmctld # after editing /etc/slurm/slurm.conf on the master