Quick Condor Tutorial

You may perform the examples in the tutorial by manually typing them in from the text below, or by using the tutorial command on login-el7.uscms.org. It's your choice; the tutorial is the same either way.

Pretyped setup

To save some typing, you can install the tutorial into your home directory from login.uscms.org. This is highly recommended to ensure that you don't encounter transcription errors during the tutorials.

[login ~]$ tutorial
usage: tutorial list                 - show available tutorials
       tutorial info <tutorial-name> - show details of a tutorial
       tutorial <tutorial-name>      - set up a tutorial

Currently available tutorials:
quickstart ............ Basic HTCondor job submission tutorial
Enter "tutorial name-of-tutorial" to clone and try out a tutorial.

Now, run the quickstart tutorial:

$ tutorial quickstart 
Installing quickstart (master)...
Tutorial files installed in ./tutorial-quickstart.
Running setup in ./tutorial-quickstart...
$ cd ~/tutorial-quickstart 

Manual setup

Alternatively, if you want the full manual experience, create a new directory for the tutorial work:

$ mkdir tutorial-quickstart 
$ cd tutorial-quickstart 

Tutorial jobs

Job 1: A simple, nonparallel job

Create a workload

Inside the tutorial directory that you created or installed previously, let's create a test script to execute as your job:

$ nano short.sh
file: short.sh
# short.sh: a short discovery job
printf "Start time: "; /bin/date
printf "Job is running on node: "; /bin/hostname
printf "Job running as user: "; /usr/bin/id
printf "Job is running in directory: "; /bin/pwd
echo "Working hard..."
sleep ${1-15}
echo "Science complete!"

To close nano, hold down Ctrl and press X. Press Y to save, and then Enter

Now, make the script executable.

$ chmod +x short.sh 

If you used the tutorial command, all files are already in your workspace.

Run the job locally

When setting up a new job type, it's important to test your job outside of Condor before submitting into the grid.

$ ./short.sh 
Start time: Fri Oct 23 14:01:45 CDT 2015
Job is running on node: login.cms.ci-connect.net
Job running as user: uid=24239(khurtado) gid=1000(users) groups=1000(users),5532(@connect),5782(@osg),6220(@cms-org-nd),6635(@HTCC),7180(@cms-org-colorado),7366(@cms),8368(@cms-org-und),8376(@cms-org-unl),8414(@cms-org-ucsd),8416(@cms-org-fnal)
Job is running in directory: /home/khurtado/examples/tutorial-quickstart

Working hard...
Science complete!

Create an HTCondor submit file

So far, so good! Let's create a simple (if verbose) HTCondor submit file.

$ nano tutorial01
file: tutorial01
# The UNIVERSE defines an execution environment. You will almost always use VANILLA. 
Universe = vanilla 

# EXECUTABLE is the program your job will run It's often useful 
# to create a shell script to "wrap" your actual work. 
Executable = short.sh 

# ERROR and OUTPUT are the error and output channels from your job
# that HTCondor returns from the remote host.
Error = job.error
Output = job.output

# The LOG file is where HTCondor places information about your 
# job's status, success, and resource consumption. 
Log = job.log

# +ProjectName is the name of the project reported to the OSG accounting system.
# If not specified, the default project in $HOME/.ciconnect/defaultproject is used.
# +ProjectName="cms-org-nd"

# QUEUE is the "start button" - it launches any jobs that have been 
# specified thus far. 
Queue 1

Submit the job

Submit the job using condor_submit.

$ condor_submit tutorial01
Submitting job(s). 
1 job(s) submitted to cluster 823.

Check job status

The condor_q command tells the status of currently running jobs. Generally you will want to limit it to your own jobs:

$ condor_q netid
-- Submitter: login.cms.ci-connect.net : <> : login.cms.ci-connect.net
 823.0   netid           8/21 09:46   0+00:00:06 R  0   0.0  short.sh
1 jobs; 0 completed, 0 removed, 0 idle, 1 running, 0 held, 0 suspended

If you want to see all jobs running on the system, use condor_q without any extra parameters.

You can also get status on a specific job cluster:

$ condor_q 823
-- Submitter: login.cms.ci-connect.net : <> : login.cms.ci-connect.net
 823.0   netid           8/21 09:46   0+00:00:10 R  0   0.0  short.sh
1 jobs; 0 completed, 0 removed, 0 idle, 1 running, 0 held, 0 suspended

Note the ST (state) column. Your job will be in the I state (idle) if it hasn't started yet. If it's currently scheduled and running, it will have state R (running). If it has completed already, it will not appear in condor_q.

Let's wait for your job to finish – that is, for condor_q not to show the job in its output. A useful tool for this is watch – it runs a program repeatedly, letting you see how the output differs at fixed time intervals. Let's submit the job again, and watch condor_q output at two-second intervals:

$ condor_submit tutorial01
Submitting job(s). 
1 job(s) submitted to cluster 824
$ watch -n2 condor_q netid 

When your job has completed, it will disappear from the list.

To close watch, hold down Ctrl and press C.

Job history

Once your job has finished, you can get information about its execution from the condor_history command:

$ condor_history 823
 823.0   netid            8/21 09:46   0+00:00:12 C   8/21 09:46 /home/netid/

You can see much more information about your job's final status using the -long option.

Check the job output

Once your job has finished, you can look at the files that HTCondor has returned to the working directory. If everything was successful, it should have returned:

  • a log file from Condor for the job cluster: jog.log
  • an output file for each job's output: job.output
  • an error file for each job's errors: job.error

Read the output file. It should be something like this:

$ cat job.output
Start time: Wed Aug 21 09:46:38 CDT 2013
Job is running on node: appcloud01
Job running as user: uid=58704(osg) gid=58704(osg) groups=58704(osg)
Job is running in directory: /var/lib/condor/execute/dir_2120
Sleeping for 10 seconds...
Et voila!

Job 2: Submitting jobs concurrently

What do we need to do to submit several jobs simultaneously? In the first example, Condor returned three files: outerror, and log. If we want to submit several jobs, we need to track these three files for each job. An easy way to do this is to add the $(Cluster) and $(Process) macros to the HTCondor submit file. Since this can make our working directory really messy with a large number of jobs, let's tell HTCondor to put the files in a directory called log. Here's what the second (less verbose) submit file looks like:

file: tutorial02
Universe = vanilla 
Executable = short.sh 
Error = log/job.error.$(Cluster)-$(Process) 
Output = log/job.output.$(Cluster)-$(Process) 
Log = log/job.log.$(Cluster) 
Queue 10 

Before submitting, we also need to make sure the log directory exists.

$ mkdir -p log

You'll see something like the following upon submission:

$ condor_submit tutorial02
Submitting job(s)..........
10 job(s) submitted to cluster 837.

Job 3: Passing arguments to executables

Sometimes it's useful to pass arguments to your executable from your submit file. For example, you might want to use the same job script for more than one run, varying only the parameters. You can do that by adding {Arguments to your submission file. Let's try that with tutorial03.

We want to run many more instances for this example: 100 instead of only 10. To ensure that we don't collectively overwhelm the scheduler let's also dial down our sleep time from 15 seconds to 5.

file: tutorial03
Universe = vanilla 
Executable = short.sh 
Arguments = 5 # to sleep 5 seconds 
Error = log/job.err.$(Cluster)-$(Process) 
Output = log/job.out.$(Cluster)-$(Process) 
Log = log/job.log.$(Cluster) 
Queue 10

And let's submit:

$ condor_submit tutorial03
Submitting job(s)....................................................................................................
10 job(s) submitted to cluster 938. 

Workload Analysis

You can see your jobs through the CMS Dashboard: http://dashboard.cern.ch/cms/.

Please, refer to this Twiki for an introduction on how to use CMS Dashboard.

Removing jobs

On occasion, jobs will need to be removed for a variety of reasons (incorrect parameters, errors in submission, etc.). In these instances, the condor_rm command can be used to remove an entire job submission or just particular jobs in a submission. The condor_rm command accepts a cluster id, a job id, or username and will remove an entire cluster of jobs, a single job, or all the jobs belonging to a given user respectively. E.g. if a job submission generates 100 jobs and is assigned a cluster id of 103, then condor_rm 103.0 will remove the first job in the cluster. Likewise, condor_rm 103 will remove all the jobs in the job submission and condor_rm username will remove all jobs belonging to the user. The condor_rm documenation has more details on using condor_rm including ways to remove jobs based on other constraints.

Getting help

If anything here didn't work, please email cms-connect-support@cern.ch