2.2 Executing grid jobs with ARC commands
In this chapter we assume that the user has installed the personal grid certificate and ARC middleware as described in chapter 1.3. Further we assume that the user has set up the ARC environment with the commands:
cd nordugrid-arc-standalone-11.05-2 source setup.sh
Before you can submit grid jobs, you must create a temporary proxy-certificate. ARC uses this proxy-certificate for authenticating you and checking that you have permission to submit jobs to the FGCI. The proxy certificate is created with the command:
arcproxy -S fgi.csc.fi
The arcproxy command asks for the password you have set for your certificate file. Once the proxy-certificate is created you can start executing other ARC commands. By default the proxy-certificate is valid for 12 hours. After the certificate has expired you can't submit new grid jobs or retrieve results before creating a new proxy-certificate. However, even though your proxy-certificate has expired, the grid jobs you have already submitted will continue running normally in the FGCI environment. You can also refresh your proxy-certificate before the current proxy-certificate expires by running the arcproxy command again. You can modify the validity time of the certificate with option -c validityPeriod. For example the command below would create a proxy-certificate that is valid for 72 hours.
arcproxy -S fgi.csc.fi -c validityPeriod=72h
The status of your proxy-cetificate can be checked with the command:
As jobs can be checked and retrieved easily by generating a new grid-proxy-certificate, it is not recommended to make long validity periods for the grid-proxy-certificate.
If your grid-proxy-certificate is valid, you can submit a job, defined with an xRSL file, with the command:
If no other arcsub options are used, the command first checks, what remote clusters have suitable resources for the job and then submits the job to one of these clusters. By default ARC randomly selects one of the suitable clusters. Option -b FastestQueue makes arcsub submit the job to a cluster were the number of queuing jobs is the smallest.
arcsub -b FastestQueue jobdefinition.xrsl
If you wish to submit the job to a certain FGCI cluster, you can define the the cluster name with the option -c. For example, the following command would send the job to thebe-grid.uef.fi cluster
arcsub -c thebe-grid.uef.fi jobdescription.xrsl
When arcsub has submitted the job, it prints out an identifier for the job (jobid). This identifier is used to monitor the progress of the job and to retrieve the results when the job has finished. The syntax of the job identifier is:
The command arcstat is used to check the status of grid jobs. The status of a singe job can be checked with the command:
You can see the status of all of your FGI jobs by using the option -a.
The status of a grid job can be: Preparing, Queuing, Running, Finishing, Finished or Failed. In addition to arcstat you can also use the command arccat to follow the progress of a grid job. Arccat prints out the standard output, or if you use the option -e, standard error, that the job script has generated so far. The syntax of arccat is:
arccat jobid arccat -e jobid
Once the job is in state Finished or Failed you can use the command arcget to retrieve the results. The syntax of the command is:
Arcget creates a new directory for your results on your local computer and copies the output files defined in the job description file there, as well as the standard output and standard error files produced by the grid job. This directory is named according to the number of the job (the random number in the grid job name) by default. However, if you use the option-J, the result directory is named according to the grid job name defined in the job description file.
If the arcget command runs successfully, it removes all the job related files in the FGI environment. This means that once arcget has downloaded the results, the job no longer exists in the FGI environment and it can't be accessed with arcstator other ARC commands.
You can also cancel a job from FGI before the job is finished. This can be done with the command arckill. The command arcclean removes a finished or failed job from the grid environment without downloading the results to a local computer. The syntax of these commands is:
arckill jobid arcclean jobid
You can cancel and clean all your grid jobs from the FGCI environment by using the option -a with the commands above:
arckill -a arcclean -a
Table 2. Essential ARC commands for running FGCI jobs
Check the standard output and standard error of a running or finished grid job.
Command to remove a finished or failed grid job without downloading the results.
Retrieve the results of a finished grid job.
Cancel an active grid job.
Create proxy certificate.
Check the status of grid jobs.
Command to submit a grid job.
Synchronise the grid job list of the local computer with the FGCI environment.
Below we go through a session where the simple a job, hello.xrsl, described in chapter 2.1, is executed in FGCI. Both the commands and their output are shown. The character ">" represents the command prompt. The commands given by the user are typed with bold-face letters.
First we create a grid proxy certificate and check that all the files that the job uses (job description file, command script and input files) are present in the current working directory.
> arcproxy -S fgi.csc.fi Enter pass phrase for private key: Your identity: /DC=org/DC=terena/DC=tcs/C=FI/O=CSC/CN=Kalle Käyttäjä firstname.lastname@example.org Contacting VOMS server (named fgi.csc.fi): voms.fgi.csc.fi on port: 15003 Proxy generation succeeded Your proxy is valid until: 2015-08-20 03:39:34 > ls file1.txt file2.txt hello.xrsl runhello.shAfter this the job defined in the file hello.xrsl is submitted with the command arcsub:
> arcsub hello.xrsl
ERROR: Conversion failed: @ 3055
ERROR: Conversion failed: : SEVQLVNQRUMwNiBAIEDCoDEyLjIy
Job submitted with jobid: gsiftp://celaeno-grid.lut.fi:2811/jobs/3008913401883521090110523
The output of the arcsub command includes two error messages but they can be ignored. For the future it is good to copy the jobid from the end of the arcsub output to a file for reference. Next, we follow the progress of the job with the commands arcstat and arccat:
> arcstat gsiftp://celaeno-grid.lut.fi:2811/jobs/3008913401883521090110523
State: Queuing (INLRMS:E)
> arcstat gsiftp://celaeno-grid.lut.fi:2811/jobs/3008913401883521090110523
State: Finished (FINISHED)
Exit Code: 0
> arccat gsiftp://celaeno-grid.lut.fi:2811/jobs/3008913401883521090110523
> arcget gsiftp://celaeno-grid.lut.fi:2811/jobs/3008913401883521090110523
3008913401883521090110523 file1.txt file2.txt hello.xrsl runhello.sh
> cd 3008913401883521090110523/
output.txt std.err std.out
When you submit a job with the arcsub command, information about the submitted job is written to file .arc/jobs.dat that locates in the home directory of the computer you are using. The arcstat command uses this local list to resolve the job names, when checking the grid jobs. Thus arcstat does not by default see jobs that you have been submitting from other machines.
To add the jobs submitted from other machines to your local jobs.dat, run command:
Sometimes your local joblist may also contain jobs that you have already retrieved using some other machine or that have been inactive for several weeks and thus automatically cleaned away. In these cases arcstat -a gives error messages like:
WARNING: Job information not found in the information system: gsiftp://usva.fgi.csc.fi:2811/jobs/rTVKDmGyCUhnDJ9eGpfsTMKpABFKDmABFKDmjqKKDmABFKDm8VTkNn
To get rid of these messages, run command:
This command removes the old jobs.xml file and creates a new one, based on the data it collects from the grid environment.
|Previous chapter||One level up||Next chapter|