Difference between revisions of "A quick guide to HTCondor"

From GridPP Wiki
Jump to: navigation, search
 
(15 intermediate revisions by one user not shown)
Line 2: Line 2:
 
== Basic commands ==
 
== Basic commands ==
  
Firstly, some basic HTCondor commands are as follows. To submit a job, type:
+
Firstly, some basic HTCondor commands are as follows. To submit a job type:
  
 
<pre>
 
<pre>
Line 8: Line 8:
 
</pre>
 
</pre>
  
To list your running and idle jobs, type:
+
To list running and idle jobs type:
  
 
<pre>
 
<pre>
Line 14: Line 14:
 
</pre>
 
</pre>
  
To list completed jobs, type:
+
To list completed jobs type:
  
 
<pre>
 
<pre>
 
condor_history
 
condor_history
 +
</pre>
 +
 +
Note that lcgui03 and lcgui04 each have their own HTCondor schedd daemon, i.e. job queue. This is means that <code>condor_q</code> and <code>condor_history</code> will only show jobs which were submitted on the host where you ran the query. However, if you run this command:
 +
<pre>
 +
condor_q -global <username>
 +
</pre>
 +
it will show idle and running jobs submitted to any schedd. For example:
 +
<pre>
 +
-bash-4.1$ condor_q -global alahiff
 +
 +
 +
-- Schedd: lcgui04.gridpp.rl.ac.uk : <130.246.181.132:25365?...
 +
ID      OWNER            SUBMITTED    RUN_TIME ST PRI SIZE CMD
 +
8212.0  alahiff        2/26 14:11  0+00:00:00 I  0  0.0  script.sh 1000
 +
 +
1 jobs; 0 completed, 0 removed, 1 idle, 0 running, 0 held, 0 suspended
 +
 +
 +
-- Schedd: lcgui03.gridpp.rl.ac.uk : <130.246.180.41:33754?...
 +
ID      OWNER            SUBMITTED    RUN_TIME ST PRI SIZE CMD
 +
91961.0  alahiff        2/26 14:10  0+00:00:09 R  0  0.0  script.sh 1000
 +
 +
1 jobs; 0 completed, 0 removed, 0 idle, 1 running, 0 held, 0 suspended
 
</pre>
 
</pre>
  
Line 24: Line 47:
 
Create a file called <code>simplejob.sub</code> containing:
 
Create a file called <code>simplejob.sub</code> containing:
 
<pre>
 
<pre>
cmd=script.sh
+
executable=script.sh
 
arguments=10
 
arguments=10
 
output=job.$(cluster).$(process).out
 
output=job.$(cluster).$(process).out
Line 31: Line 54:
 
should_transfer_files = YES
 
should_transfer_files = YES
 
when_to_transfer_output = ON_EXIT
 
when_to_transfer_output = ON_EXIT
RequestMemory=100
+
request_memory=100
 
queue
 
queue
 
</pre>
 
</pre>
Line 39: Line 62:
 
sleep $1
 
sleep $1
 
hostname
 
hostname
</pre>
 
Make sure it's executable:
 
<pre>
 
chmod 755 script.sh
 
 
</pre>
 
</pre>
  
Line 64: Line 83:
 
</pre>
 
</pre>
  
 +
Explanation of the content of <code>script.sh</code>:
 +
* <code>executable=script.sh</code>: the job will execute the script <code>script.sh</code>
 +
* <code>arguments=10</code>: the argument <code>10</code> will be passed to the executable when it is run
 +
* <code>request_memory=100</code>: request 100MB memory for the job
 +
* <code>should_transfer_files = YES</code>: tells HTCondor to transfer files to/from the worker node
 +
* <code>when_to_transfer_output = ON_EXIT</code>: tells HTCondor transfers any output files only when the job has completed
 +
* <code>output=job.$(cluster).$(process).out</code>: tells HTCondor the path (and name) of the file containing the job's stderr on the submit machine
 +
* <code>error=job.$(cluster).$(process).err</code>: tells HTCondor the path (and name) of the file containing the job's stdout on the submit machine
 +
* <code>log=job.$(cluster).$(process).log</code>: tells HTCondor where to write the job event log and the name of the file on the submit machine
  
The job will:
+
Once the job has completed there will be 3 files visible, containing the log, stdout and stderr:
- run the script "script.sh" with argument "10"
+
<pre>
- 100 MB of memory will be requested
+
- the executable (script.sh in this case) will be automatically transferred to the worker nodes
+
- stdout from the job will end up in the file job.91959.0.out (once it completes)
+
- stderr from the job will end up in the file job.91959.0.err (once it completes)
+
- the log file job.91959.0.log will give some information about the status of the job
+
 
+
Checking the status:
+
 
+
Once the job is completed
+
 
+
 
-bash-4.1$ ls -lt *91959*
 
-bash-4.1$ ls -lt *91959*
 
-rw-r--r-- 1 alahiff esc 1032 Feb 26 13:09 job.91959.0.log
 
-rw-r--r-- 1 alahiff esc 1032 Feb 26 13:09 job.91959.0.log
 
-rw-r--r-- 1 alahiff esc  24 Feb 26 13:09 job.91959.0.out
 
-rw-r--r-- 1 alahiff esc  24 Feb 26 13:09 job.91959.0.out
 
-rw-r--r-- 1 alahiff esc    0 Feb 26 13:08 job.91959.0.err
 
-rw-r--r-- 1 alahiff esc    0 Feb 26 13:08 job.91959.0.err
 +
</pre>
 +
The log file will in fact be generated as soon as the job is submitted to HTCondor and will be updated regularly.
  
Note that with this example job description file, all files generated by the job will be automatically transferred back to the machine where you submitted the job (i.e. lcgui03 or lcgui04). You can prevent this from happening by adding:
+
== Input & output files ==
  
 +
We don't use s shared filesystem on the worker nodes. Instead we rely on HTCondor to transfer files between the submit machine (e.g. lcgui03 or lcgui04) and the worker node. By default the executable will always be transferred automatically to the worker nodes.
 +
 +
Note that with the example job description file above, all files generated by the job will be automatically transferred back to the machine where you submitted the job. You can prevent this from happening by adding the following to the job description file:
 +
 +
<pre>
 
+TransferOutput=""
 
+TransferOutput=""
 +
</pre>
 +
 +
If there are specific output files you want copied back to the submit machine you can specify these using <code>transfer_output_files</code>, for example:
 +
 +
<pre>
 +
transfer_output_files = outputfile1,outputfile2
 +
</pre>
  
 
If the job needs additional files, you can add a line something like this:
 
If the job needs additional files, you can add a line something like this:
  
 +
<pre>
 
transfer_input_files = input1.dat,input2.dat
 
transfer_input_files = input1.dat,input2.dat
 +
</pre>
  
and they will be copied.
 
  
The official documentation is here:
+
== Official documentation ==
  
 
http://research.cs.wisc.edu/htcondor/manual/v8.4/index.html
 
http://research.cs.wisc.edu/htcondor/manual/v8.4/index.html
  
Regards,
+
Information about submitting jobs http://research.cs.wisc.edu/htcondor/manual/v8.4/condor_submit.html#man-condor-submit
Andrew.
+

Latest revision as of 14:25, 26 February 2016

Basic commands

Firstly, some basic HTCondor commands are as follows. To submit a job type:

condor_submit <file>

To list running and idle jobs type:

condor_q

To list completed jobs type:

condor_history

Note that lcgui03 and lcgui04 each have their own HTCondor schedd daemon, i.e. job queue. This is means that condor_q and condor_history will only show jobs which were submitted on the host where you ran the query. However, if you run this command:

condor_q -global <username>

it will show idle and running jobs submitted to any schedd. For example:

-bash-4.1$ condor_q -global alahiff


-- Schedd: lcgui04.gridpp.rl.ac.uk : <130.246.181.132:25365?...
 ID      OWNER            SUBMITTED     RUN_TIME ST PRI SIZE CMD
8212.0   alahiff         2/26 14:11   0+00:00:00 I  0   0.0  script.sh 1000

1 jobs; 0 completed, 0 removed, 1 idle, 0 running, 0 held, 0 suspended


-- Schedd: lcgui03.gridpp.rl.ac.uk : <130.246.180.41:33754?...
 ID      OWNER            SUBMITTED     RUN_TIME ST PRI SIZE CMD
91961.0   alahiff         2/26 14:10   0+00:00:09 R  0   0.0  script.sh 1000

1 jobs; 0 completed, 0 removed, 0 idle, 1 running, 0 held, 0 suspended

Job submission

Create a file called simplejob.sub containing:

executable=script.sh
arguments=10
output=job.$(cluster).$(process).out
error=job.$(cluster).$(process).err
log=job.$(cluster).$(process).log
should_transfer_files = YES
when_to_transfer_output = ON_EXIT
request_memory=100
queue

and a script called script.sh containing:

#!/bin/sh
sleep $1
hostname

Submit the job:

-bash-4.1$ condor_submit simplejob.sub
Submitting job(s).
1 job(s) submitted to cluster 91959.

Checking the status of the job:

-bash-4.1$ condor_q


-- Schedd: lcgui03.gridpp.rl.ac.uk : <130.246.180.41:33754?...
 ID      OWNER            SUBMITTED     RUN_TIME ST PRI SIZE CMD
91959.0   alahiff         2/26 13:08   0+00:00:07 R  0   0.0  script.sh 10

1 jobs; 0 completed, 0 removed, 0 idle, 1 running, 0 held, 0 suspended

Explanation of the content of script.sh:

  • executable=script.sh: the job will execute the script script.sh
  • arguments=10: the argument 10 will be passed to the executable when it is run
  • request_memory=100: request 100MB memory for the job
  • should_transfer_files = YES: tells HTCondor to transfer files to/from the worker node
  • when_to_transfer_output = ON_EXIT: tells HTCondor transfers any output files only when the job has completed
  • output=job.$(cluster).$(process).out: tells HTCondor the path (and name) of the file containing the job's stderr on the submit machine
  • error=job.$(cluster).$(process).err: tells HTCondor the path (and name) of the file containing the job's stdout on the submit machine
  • log=job.$(cluster).$(process).log: tells HTCondor where to write the job event log and the name of the file on the submit machine

Once the job has completed there will be 3 files visible, containing the log, stdout and stderr:

-bash-4.1$ ls -lt *91959*
-rw-r--r-- 1 alahiff esc 1032 Feb 26 13:09 job.91959.0.log
-rw-r--r-- 1 alahiff esc   24 Feb 26 13:09 job.91959.0.out
-rw-r--r-- 1 alahiff esc    0 Feb 26 13:08 job.91959.0.err

The log file will in fact be generated as soon as the job is submitted to HTCondor and will be updated regularly.

Input & output files

We don't use s shared filesystem on the worker nodes. Instead we rely on HTCondor to transfer files between the submit machine (e.g. lcgui03 or lcgui04) and the worker node. By default the executable will always be transferred automatically to the worker nodes.

Note that with the example job description file above, all files generated by the job will be automatically transferred back to the machine where you submitted the job. You can prevent this from happening by adding the following to the job description file:

+TransferOutput=""

If there are specific output files you want copied back to the submit machine you can specify these using transfer_output_files, for example:

transfer_output_files = outputfile1,outputfile2

If the job needs additional files, you can add a line something like this:

transfer_input_files = input1.dat,input2.dat


Official documentation

http://research.cs.wisc.edu/htcondor/manual/v8.4/index.html

Information about submitting jobs http://research.cs.wisc.edu/htcondor/manual/v8.4/condor_submit.html#man-condor-submit