A few utilities for use on (hopefully) any SLURM cluster.
The only requirements are python 2.6+ and that the standard slurm commands
(squeue,sinfo,...) are available through PATH.
If the hostlist module is available to python gnodes will use it to expand
lists of nodes, otherwise it will use an included function with some
limitations.
The two scripts are self-contained, so installation is done by downloading the latest version and then marking the two files as executable.
wget https://raw.githubusercontent.com/birc-aeh/slurm-utils/master/gnodes
wget https://raw.githubusercontent.com/birc-aeh/slurm-utils/master/jobinfo
chmod +x gnodes jobinfo
mv gnodes jobinfo ~/.local/bin/ # or another folder in your $PATH
The gnodes script gives a visual representation of your cluster.
It shows you available memory and allocated/loaded cores for every every node
in each partition.
The layout is adjusted to the terminal width, but might get very tall if you
have a lot of nodes of course.
It optionally also takes multiple search parameters that can be usernames, job ids or node names. Nodes that are running either the mentioned job or any job from the user will be highlighted.
Example output:
+- normal - 16 cores & 23GB ------+---------------------------------+
| norm_54 23G ........_____OOO | norm_70 23G ................ |
| norm_55 23G ........_____OOO | norm_71 23G ................ |
| norm_56 23G ........_____OOO | norm_72 23G ................ |
| norm_57 23G ............___O | norm_73 23G ................ |
| norm_58 23G ........____OOOO | norm_74 23G ................ |
| norm_59 23G ........____OOOO | norm_75 23G ................ |
| norm_60 0G _______________O | norm_76 23G ................ |
| norm_61 23G ........____OOOO | norm_77 23G ................ |
| norm_62 23G ......_____OOOOO | norm_78 23G ................ |
| norm_63 23G ................ | norm_79 23G ................ |
| norm_64 23G ................ | norm_80 23G ........!!!!!!!! |
| norm_65 23G ................ | norm_81 23G ........!!!!!!!! |
| norm_66 23G ................ | norm_82 23G ................ |
| norm_67 23G ................ | norm_83 23G ........____OOOO |
| norm_68 23G ................ | norm_84 23G ........____OOOO |
| norm_69 23G ................ | norm_85 23G ........____OOOO |
+---------------------------------+---------------------------------+
+- fancyfancy - 40 cores & 62GB --------------------------+
| fancy_0 0G ____________________________OOOOOOOOOOOO |
| fancy_1 0G ___________________________OOOOOOOOOOOOO |
| fancy_2 0G ____________________________OOOOOOOOOOOO |
| fancy_3 0G __________________________OOOOOOOOOOOOOO |
| fancy_4 0G ____________________________OOOOOOOOOOOO |
| fancy_5 0G ___________________________OOOOOOOOOOOOO |
+---------------------------------------------------------+
Unallocated cores are marked with ., allocated cores with no load
are marked with _, loaded cores are marked with O and if the load
goes above 1.5 times the allocated number of cores it is marked with !.
The jobinfo script tries to collect information for a full job combining
information from the SLURM accounting system and live stats from sstat if the
job is still running.
Example output:
[aeh@fe1 ~]$ jobinfo 11983512
Name : bash
User : aeh
Partition : normal
Nodes : s02n[45-48,51-53]
Cores : 50
State : FAILED
Submit : 2015-10-12T21:21:18
Start : 2015-10-12T21:21:23
End : 2015-10-12T21:24:14
Reserved walltime : 2-00:00:00
Used walltime : 00:02:51
Used CPU time : 00:00:59
% User (Computation): 83.22%
% System (I/O) : 16.78%
Mem reserved : 100M/node
Max Mem used : 25.18M (s02n45,s02n47,s02n48,s02n51,s02n53)
Max Disk Write : 16.00M (s02n45)
Max Disk Read : 2.00M (s02n45)
It has mostly been tested on batch jobs without any sub-steps so please send feedback.