Difference between revisions of "Submitting CPU Jobs"

From UFAL AIC
(→‎Advanced usage)
(→‎Selected submit options)
 
(35 intermediate revisions by 3 users not shown)
Line 1: Line 1:
−
== Resource specification ==
+
The CPU jobs should be submitted to <code>cpu</code> partition.
−
== Monitoring and interaction ==
 
  
−
== Job monitoring ==
+
You can submit a non-interactive job using the '''sbatch''' command.
−
We should be able to see what is going on when we run a job. Following examples shows typical usage of the command <code>qstat</code>:
+
To submit an interactive job, use the '''srun''' command:
−
* <code>qstat</code> - this way we inspect all our jobs (both waiting in the queue and scheduled, i.e. running).
 
−
* <code>qstat -u '*' | less</code> - this shows the jobs of all users.
 
−
* <code>qstat -j 121144</code> - this shows detailed info about the job with this number (if it is still running).
 
  
−
== Output monitoring ==
+
srun --pty bash
−
If we need to see output produced by our job (suppose the ID is 121144), we can inspect the job's output (in our case stored in <code>job_script.sh.o121144</code>) with:<br>
 
−
<code>less job_script.sh.o*</code><br>
 
−
''Hint:'' if the job is still running, press '''F''' in <code>less</code> to simulate <code>tail -f</code>.
 
  
−
=== How to read output epilog ===
+
== Resource specification ==
−
The epilog section contains some interesting pieces of information. However this it can get confusing sometimes.
 
  
−
======= EPILOG: Tue Jun 4 12:41:07 CEST 2019
+
You should specify the memory and CPU requirements (if higher than the defaults) and don't exceed them.
−
== Limits: 
+
If your job needs more than one CPU (thread) (on a single machine) for most of the time, reserve the given number of CPU threads with the <code>--cpus-per-task</code> and memory with the <code>--mem</code> options.   
−
== Usage:    cpu=00:00:00, mem=0.00000 GB s, io=0.00000 GB, vmem=N/A, maxvmem=N/A
 
−
  == Duration: 00:00:00 (0 s)
 
−
== Server name: cpu-node13
 
  
−
* ''Limits'' - on this line you can see job limits specified through <code>qsub</code> options
+
srun -p cpu --cpus-per-task=4 --mem=8G --pty bash
−
* ''Usage'' - resource usage during computation
+
−
** ''cpu=HH:MM:SS'' - the accumulated CPU time usage
+
This will give you an interactive shell with 4 threads and 8G RAM on the ''cpu'' partition.
−
** ''mem=XY GB'' - gigabytes of RAM used times the duration of the job in seconds, so don't be afraid XY is usually a very high number (unlike in this toy example)
 
−
** ''io=XY GB'' - the amount of data transferred in input/output operations in GB
 
−
** ''vmem=XY'' - actual virtual memory consumption when the job finished
 
−
** ''maxvmem=XY'' - peak virtual memory consumption
 
−
* ''Duration'' - total execution time
 
−
* ''Server name'' - name of the executing server
 
  
−
== Advanced usage ==
+
== Monitoring and interaction ==
−
<code>qsub '''-q''' cpu.q</code>
 
−
This way your job is submitted to the CPU queue which is the default. If you need GPU use <code>gpu.q</code> instead.
 
−
 
 
−
<code>qsub '''-l''' ...</code>
 
−
See <code>man complex</code> (run it on aic) for a list of possible resources you may require (in addition to <code>mem_free</code> etc. discussed above).
 
−
 
 
−
<code>qsub '''-p''' -200</code>
 
−
Define a priority of your job as a number between -1024 and 0. Only SGE admins may use a number higher than 0. Default is set to TODO. You should ask for lower priority (-1024..-101) if you submit many jobs at once or if the jobs are not urgent. SGE uses the priority to decide when to start which pending job in the queue (it computes a real number called <code>prior</code>, which is reported in <code>qstat</code>, which grows as the job is waiting in the queue). Note that once a job is started, you cannot ''unschedule'' it, so from that moment on, it is irrelevant what was its priority.
 
−
 
 
−
<code>qsub '''-o** LOG.stdout **-e''' LOG.stderr</code>
 
−
redirect std{out,err} to separate files with given names, instead of the defaults <code>$JOB_NAME.o$JOB_ID</code> and <code>$JOB_NAME.e$JOB_ID</code>.
 
−
 
 
−
<code>qsub '''-@''' optionfile</code>
 
−
Instead of specifying all the <code>qsub</code> options on the command line, you can store them in a file (you can use # comments in the file). See also [[#In-script options]].
 
  
−
<code>qsub '''-a''' 12312359</code>
+
=== Job monitoring ===
−
Execute your job no sooner than at the given time (in <code>[YY]MMDDhhmm'' format). An alternative to ''sleep 3600 && qsub ... &</code>.
+
We should be able to see what is going on when we run a job. Following examples shows usage of some typical commands:
 +
* <code>squeue -a</code> - this shows the jobs in all partitions.
 +
* <code>squeue -u user</code> - print a list of running/waiting jobs of a given user
 +
* <code>squeue -j<JOB_ID></code> - this shows detailed info about the job with given JOB_ID (if it is still running).
 +
* <code>sinfo</code> - print available/total resources
  
−
<code>qsub '''-b''' y</code>
+
=== Job interaction ===
−
Treat <code>script.sh'' (or whatever is the name of the command you execute) as a binary, i.e. don't search for [[#in-script options]] within the file, don't transfer it to the qmaster and then to the execution node. This makes the execution a bit faster and it may prevent some rare but hard-to-detect errors caused SGE interpreting the script. The script must be available on the execution node via NFS, Lustre (which is our case), etc. With ''-b y'' (shortcut for ''-b yes''), ''script.sh'' can be a script or a binary. With ''-b n'' (which is the default for ''qsub''), ''script.sh</code> must be a script (text file).
+
* <code>scontrol show job JOBID</code> - this shows details of running job with JOBID
 +
* <code>scancel JOBID</code> - delete job from the queue
  
−
<code>qsub '''-M** popel@ufal.mff.cuni.cz,rosa@ufal.mff.cuni.cz **-m''' beas</code>
+
=== Selected submit options ===
−
Specify the emails where you want to be notified when the job has been '''b** started, **e** ended, **a** aborted or rescheduled, **s''' suspended.
+
The complete list of available options for the commands <code>srun</code> and <code>sbatch</code> can be found in [https://slurm.schedmd.com/man_index.html SLURM documentation]. Most of the options listed here can be entered as a command parameters or as an SBATCH directive inside of a script.
−
The default is now <code>-m a'' and the default email address is forwarded to you (so there is no need to use ''-M''). You can use ''-m n</code> to override the defaults and send no emails.
 
  
−
<code>qsub '''-hold_jid** 121144,121145'' (or ''qsub **-hold_jid''' get_src.sh,get_tgt.sh</code>)
+
  -J helloWorld        # name of job
−
The current job is not executed before all the specified jobs are completed.
+
  --chdir /job/path/   # path where the job will be executed
 +
  -p gpu                # name of partition or queue (if not specified default partition is used)
 +
  -q normal            # QOS level (sets priority of the job)
 +
  -c 4                  # reserve 4 CPU threads
 +
  --gres=gpu:1          # reserve 1 GPU card
 +
  -o script.out        # name of output file for the job
 +
  -e script.err        # name of error file for the job
  
−
<code>qsub '''-now''' y</code>
+
== Array jobs ==
−
Start the job immediately or not at all, i.e. don't put it as pending to the queue. This is the default for <code>qrsh'', but you can change it with ''-now n'' (which is the default for ''qsub</code>).
+
If you need to submit rather large number of jobs which are similar (i.e. processing a large number of input files) you should consider launching an ''array job''.
  
−
<code>qsub '''-N''' my-name</code>
+
For example, one might need to process 1000 files named <code>file_N.txt</code> (where N is a number between 1-1000).
−
By default the name of a job (which you can see e.g. in <code>qstat'') is the name of the ''script.sh</code>. This way you can override it.
+
A program that can process one file is called <code>crunchFile</code> and it takes only one argument - the name of the file to process. Instead of calling 1000x:
 +
  sbatch crunchFile file_N.txt
  
−
<code>qsub '''-S''' /bin/bash</code>
+
we can write a wrapper script <code>crunchScript.sh</code> referring to the SLURM variable <code>SLURM_ARRAY_TASK_ID</code>:
−
The hashbang (<code>!#/bin/bash'') in your ''script.sh'' is ignored, but you can change the interpreter with ''-S''. I think ''/bin/bash'' is now (2017/09) the default (but it used to be ''csh</code>).
 
  
−
<code>qsub '''-v''' PATH[=value]</code>
+
  #!/bin/bash
−
Export a given environment variable from the current shell to the job.
+
  #SBATCH -p CPU
 +
  #SBATCH --mem 2G
 +
 
 +
  crunchFile name_${SLURM_ARRAY_TASK_ID}.txt
  
−
<code>qsub '''-V'''</code>
+
and submit all the jobs at once as an ''array job'':
−
Export all environment variables. (This is not so needed now, when bash is the default interpreter and it seems your <code>~/.bashrc</code> is always sourced.)
 
  
−
<code>qsub '''-soft** -l ... **-hard''' -l ... -q ...</code>
+
  sbatch --array=1-1000%20 crunchScript.sh
−
By default, all the resource requirements (specified with <code>-l'') and queue requirements (specified with ''-q'') are //hard//, i.e. your job won't be scheduled unless they can be fulfilled. You can use ''-soft'' to mark all following requirements as nice-to-have. And with ''-hard</code> you can switch back to hard requirements.
 
  
−
<code>qsub '''-sync''' y</code>
+
Where the option <code>--array 1-1000%20</code> means that we want SLURM to:
−
This causes qsub to wait for the job to complete before exiting (with the same exit code as the job). Useful in scripts.
+
* launch 1000 instances of <code>crunchScript.sh</code>
 +
* each instance will be launched with <code>SLURM_ARRAY_TASK_ID</code> set to a number in the specified range
 +
* there will be at most 20 parallel tasks running at once. This is useful for a larger number of tasks - this way we ensure that we do not flood the cluster with requests.
  
−
<code>'''qalter'''</code>
+
You can read more about ''array jobs'' from the [https://slurm.schedmd.com/job_array.html SLURM documentation].
−
You can change some properties of already submitted jobs (both waiting in the queue and running). Changeable properties are listed in <code>man qsub</code>.
 

Latest revision as of 15:49, 23 April 2024

The CPU jobs should be submitted to cpu partition.

You can submit a non-interactive job using the sbatch command. To submit an interactive job, use the srun command: 

srun --pty bash

Resource specification

You should specify the memory and CPU requirements (if higher than the defaults) and don't exceed them. If your job needs more than one CPU (thread) (on a single machine) for most of the time, reserve the given number of CPU threads with the --cpus-per-task and memory with the --mem options. 

srun -p cpu --cpus-per-task=4 --mem=8G --pty bash

This will give you an interactive shell with 4 threads and 8G RAM on the cpu partition.

Monitoring and interaction

Job monitoring

We should be able to see what is going on when we run a job. Following examples shows usage of some typical commands:

  • squeue -a - this shows the jobs in all partitions.
  • squeue -u user - print a list of running/waiting jobs of a given user
  • squeue -j<JOB_ID> - this shows detailed info about the job with given JOB_ID (if it is still running).
  • sinfo - print available/total resources

Job interaction

  • scontrol show job JOBID - this shows details of running job with JOBID
  • scancel JOBID - delete job from the queue

Selected submit options

The complete list of available options for the commands srun and sbatch can be found in SLURM documentation. Most of the options listed here can be entered as a command parameters or as an SBATCH directive inside of a script. 

 -J helloWorld         # name of job
 --chdir /job/path/    # path where the job will be executed
 -p gpu                # name of partition or queue (if not specified default partition is used)
 -q normal             # QOS level (sets priority of the job)
 -c 4                  # reserve 4 CPU threads
 --gres=gpu:1          # reserve 1 GPU card
 -o script.out         # name of output file for the job 
 -e script.err         # name of error file for the job

Array jobs

If you need to submit rather large number of jobs which are similar (i.e. processing a large number of input files) you should consider launching an array job.

For example, one might need to process 1000 files named file_N.txt (where N is a number between 1-1000). A program that can process one file is called crunchFile and it takes only one argument - the name of the file to process. Instead of calling 1000x: 

  sbatch crunchFile file_N.txt

we can write a wrapper script crunchScript.sh referring to the SLURM variable SLURM_ARRAY_TASK_ID: 

  #!/bin/bash
  #SBATCH -p CPU
  #SBATCH --mem 2G
  
  crunchFile name_${SLURM_ARRAY_TASK_ID}.txt

and submit all the jobs at once as an array job: 

 sbatch --array=1-1000%20 crunchScript.sh

Where the option --array 1-1000%20 means that we want SLURM to:

  • launch 1000 instances of crunchScript.sh
  • each instance will be launched with SLURM_ARRAY_TASK_ID set to a number in the specified range
  • there will be at most 20 parallel tasks running at once. This is useful for a larger number of tasks - this way we ensure that we do not flood the cluster with requests.

You can read more about array jobs from the SLURM documentation.

Retrieved from "https://aic.ufal.mff.cuni.cz/index.php?title=Submitting_CPU_Jobs&oldid=120"