WLCG Accounting Task Force Meeting

Europe/Zurich
513/R-068 (CERN)

513/R-068

CERN

19
Show room on map

Attended:

Jarka, Luca Magnoni, Ivan, Steve, Antonio, Adrian, Gavin, Julia

 

Discussion after Gavin's presentation:

The fix for Condor related bug  to be provided by the end of September.

Julia asked, whether we need to resent fixed data to APEL. Gavin told that old data before May is not any more available.

There was no strong opinion about necessity of fixing old data neither from ATLAS (Ivan) nor from CMS (Antonio).

Ivan and Antonio will check with experiments community whether it is required.

CERN accounting data is available in the EGI portal, but also in the internal CERN accounting portal. Ivan asked whether it is possible to split accounting by various CERN cluster categories (cloud, condor, etc...). Gavin told that 'yes' it is possible. Julia told that she remembered that this is also possible  in the EGI portal. Julia checked after the meeting and confirmed, that yes, it is possible, using 'Submit host' attribute. Currently it shows 3 categories: cloud, condor and lsf', for example boinc and slurm hpc are currently not there.

Julia asked Jarka, whether now the crosscheck with ATLAS accounting is performed on regular basis. Jarka confirmed that it is. She will send the link where the result of this comparison can been seen, the link should be added to the twiki page of the accounting task force.

Antonio mentioned that at PIC they experienced the problem of not taking into account  some condor jobs , in particular those which are killed by the VO. Gavin supposed that they might be excluded from the consideration since these jobs have a particular status in the condor-history file.

Ivan told  that since beginning of June after introducing new submission mechanism by ATLAS via Harvester, there should be no more empty pilots any more and asked Jarka and Gavin to inform ATLAS in case they still see such a problem. Jarka told that she would add number of jobs metric to her comparison scripts, like this it would be possible to detect the problem in case it shows up.

Discussion after Steve's presentation:

Julia asked Steve, whether she understood correctly that his parser was working with condor batch history file and therefore should work regardless of the type of the CE which is run by the site. Steve confirmed that this was the case.

Julia told that then the scripts of Steve (when he feels that the scripts are ready to be used by other people) can be advertised to the sites which run condor batch system.

Julia told that so far we did not have clear recommendations for the sites how to enable accounting for sites with HTCondor CEs and we need to work on it.  Adrian mentioned that there are two different pending pull requests addressing HTCondor accounting, one from PIC another one from Kirsti. The second one also originated from the PIC one,  but has been improved. Would be nice if these solutions are evaluated and we have one single consolidated solution which we could advertise to the sites.  Steve volunteered to look into it.

The next meeting will take place most probably on the 25th of October and we will review the situation with HTCondor accounting.

 

 

 

 

 

There are minutes attached to this event. Show them.