(r1) CafQA < Sandbox

---++Basic description and usage of the automatized crab job submission script for the IBs.

---+++IDEA
Run a script triggered by a cronjob, which sends some 'standard' CMSSW jobs on real data to the grid. 

---+++OPERATION
The files are located here: http://cmssw.cvs.cern.ch/cgi-bin/cmssw.cgi/UserCode/bbozsogi/CRABCAF/

The cronjob starts =Tester/standaloneTester-caf.py= every 12h. Next to the usual statusFile (containing the time it started, finished), I set up a global statusFile (**now it's in my AFS area**), preventing to start new CAF jobs before the latest has finished.

**important**:%BR%
For the script to run, cmssw and grid environmental variables has to be set (done by standaloneTester).%BR%
For the IB runs, we have to use custom CRAB build, since they're not supported by default. I set up one on vocms101 (/build/bbozsogi/CRAB_2_7_5), currently the script runs on this machine.

Basically the script:
   * creates a crabCAF directory
   * checks out http://cmssw.cvs.cern.ch/cgi-bin/cmssw.cgi/CMSSW/Configuration/PyReleaseValidation/data/cmsDriver_standard_hlt.txt
   * searches for jobs meant to be run on the caf
   * sets up directories for each of them
   * runs cmsDriver.py with --no_exec option
   * creates the crab.cfg file
   * send the jobs
   * waits for the results
   * copy the logs and a pickled dictionary with some basic parsed info about the jobs to the IB's AFS area, under _cafQAlogs/<job_name>/_
   * sends a mail containing all the output from crab and the parsed info (crabCAF.log)

---+++USAGE
I'll try to keep the =crabRun.py --help= up-to-date :) %BR%
By running the script without parameters, it does the things listed above. %BR%

You can change the default crab.cfg parameters (see CafQADefaults) by: 
<verbatim>
crabRun.py -c 'USER:email=bbozsogi@cern.ch,CMSSW:events_per_job=2000'
</verbatim>

You can send custom jobs to the CAF by using option =-inputCmd='cmsDriver.py ...' -d='<datasetpath>'= . %BR%
In this case it runs the cmsDriver.py with -no_exec option and use that config file as an input. For this, you also have to 
<verbatim>
crabRun.py -d '/Electron/Run2010B-WZEG-v2/RAW-RECO' -i 'cmsDriver.py step2 -s RAW2DIGI,L1Reco,RECO,ALCA:SiStripCalZeroBias+SiStripCalMinBias,DQM 
--data --datatier RECO --eventcontent RECO --conditions auto:com10 --scenario pp --no_exec --magField AutoFromDBCurrent 
--process reRECO --customise Configuration/DataProcessing/RecoTLR.py --cust_function customisePPData'
</verbatim>


---+++PICKLE
The pickle file contains a dictionary(=logData=) with the most important infos about the job.%BR%
it has a dictionary for every logFile in a stucture like:
<verbatim>
logData[logName] = {
   'JobExitCode' = exitcode
   'files' = ['file1.root', 'file2.root', ...]
   'SE_PATH' = '/castor/.../'
   'MSG' = [((startNum, endNum), 'errorMessage'), ...]
}
</verbatim>

The 'MSG' part is the most important, it's basically every block in the logs between =%MSG= tags.

A way to parse it (used it for the crabCAF.log):
<verbatim>
        for key, value in logData.items():
            self.log.write(key+': {\n')
            for key2, value2 in value.items():
                if key2 == 'MSG':
                    map(lambda x: self.log.write('---'+key2+': '+'('+str(x[0][0])+'-'+str(x[0][1])+'):\n'+x[1]), value2)
                elif key2 == 'files':
                    self.log.write('---'+key2+': '+', '.join(value2).split()[0])
                    self.log.write('\n')
                else:
                    self.log.write('---'+key2+': '+value2)
                    self.log.write('\n')
            self.log.write('}\n')
            self.log.flush()
</verbatim>

---+++ISSUES
Now, the cronjob is running under my user. We have to port it at some point to cmsbuild maybe. %BR%
I have sometimes IOError on the statusFiles, need to check why.%BR%
Some more error protection...



-- Main.BalazsBozsogi - 17-Nov-2010
Topic revision: r1 - 2010-11-17 - BalazsBozsogi
Account
- Log In
- Cern Search
- TWiki Search
- Google Search
Sandbox All webs
~~Edit~~
~~Attach~~
Copyright &© 2008-2021 by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
or Ideas, requests, problems regarding TWiki? use Discourse or Send feedback