help
help > RE: Cluster computing - SLURM jobmanager test fails
May 6, 2021 11:05 PM | Alfonso Nieto-Castanon - Boston University
RE: Cluster computing - SLURM jobmanager test fails
Hi Amy,
More importantly, it is missing the SCRIPT portion (!) which specifies the script that you would like to submit, so that explains why sbatch would simply wait/hang (when no script is specified, sbatch will wait and read the script from the standard input). In any case, you can fix that cmd_submit value from the GUI or from the command-line. if working from the GUI, go to Tools. HPC options. Configuration and edit the line below "command used to submit a job" to read:
sbatch --job-name=JOBLABEL --error=STDERR --output=STDOUT OPTS SCRIPT
(this is the default settings, the JOBLABEL, STDERR, etc. words are keywords that are filled-in automatically by CONN with the correct values, so generally there is no need to modify this line at all). Then, if you want to add other options, like your --qos option, add those options to the line below "additional submit settings (optional) : in-line", e.g.:
-t 12:00:00 --mem=8Gb --qos=blanca-ics
and then click 'Save' to save those changes (if you are the owner of the conn distribution files in your cluster/system, you may use the 'All users' option when saving so these settings apply to all users, otherwise use the 'Current user' option so these settings apply only to your user)
Similarly, if you prefer to do the above two steps using the command-line, you can do so using the syntax:
>> conn_jobmanager options cmd_submit 'sbatch --job-name=JOBLABEL --error=STDERR --output=STDOUT OPTS SCRIPT'
>> conn_jobmanager options cmd_submitoptions '-t 12:00:00 --mem=8Gb --qos=blanca-ics' save current
Let me know if that seems to fix those issues
Best
Alfonso
Originally posted by Amy Hegarty:
More importantly, it is missing the SCRIPT portion (!) which specifies the script that you would like to submit, so that explains why sbatch would simply wait/hang (when no script is specified, sbatch will wait and read the script from the standard input). In any case, you can fix that cmd_submit value from the GUI or from the command-line. if working from the GUI, go to Tools. HPC options. Configuration and edit the line below "command used to submit a job" to read:
sbatch --job-name=JOBLABEL --error=STDERR --output=STDOUT OPTS SCRIPT
(this is the default settings, the JOBLABEL, STDERR, etc. words are keywords that are filled-in automatically by CONN with the correct values, so generally there is no need to modify this line at all). Then, if you want to add other options, like your --qos option, add those options to the line below "additional submit settings (optional) : in-line", e.g.:
-t 12:00:00 --mem=8Gb --qos=blanca-ics
and then click 'Save' to save those changes (if you are the owner of the conn distribution files in your cluster/system, you may use the 'All users' option when saving so these settings apply to all users, otherwise use the 'Current user' option so these settings apply only to your user)
Similarly, if you prefer to do the above two steps using the command-line, you can do so using the syntax:
>> conn_jobmanager options cmd_submit 'sbatch --job-name=JOBLABEL --error=STDERR --output=STDOUT OPTS SCRIPT'
>> conn_jobmanager options cmd_submitoptions '-t 12:00:00 --mem=8Gb --qos=blanca-ics' save current
Let me know if that seems to fix those issues
Best
Alfonso
Originally posted by Amy Hegarty:
Hi Alfonso,
You are correct that the test does not progress, or show the second line 'Waiting for grid/cluster jobs to finish'.
As for the suggested tests:
>> conn_jobmanager options cmd_submit
sbatch --job-name=conntest --qos=blanca-ics
** so this is missing the --error and --output you showed in your example.
>> conn_jobmanager options cmd_submitoptions
-t 12:00:00 --mem=8Gb
>>conn_jobmanager options cmd_submitoptions_infile
** empty
Also, I did double check sbatch is not aliased and points to /usr/bin/sbatch.
Thanks again!
Amy
You are correct that the test does not progress, or show the second line 'Waiting for grid/cluster jobs to finish'.
As for the suggested tests:
>> conn_jobmanager options cmd_submit
sbatch --job-name=conntest --qos=blanca-ics
** so this is missing the --error and --output you showed in your example.
>> conn_jobmanager options cmd_submitoptions
-t 12:00:00 --mem=8Gb
>>conn_jobmanager options cmd_submitoptions_infile
** empty
Also, I did double check sbatch is not aliased and points to /usr/bin/sbatch.
Thanks again!
Amy
Threaded View
| Title | Author | Date |
|---|---|---|
| Amy Hegarty | Apr 28, 2021 | |
| Alfonso Nieto-Castanon | May 3, 2021 | |
| Amy Hegarty | May 4, 2021 | |
| Alfonso Nieto-Castanon | May 4, 2021 | |
| Amy Hegarty | May 6, 2021 | |
| Alfonso Nieto-Castanon | May 6, 2021 | |
| Amy Hegarty | May 7, 2021 | |