Reporting of jobs to SGAS by site - NDGF-T1_BOINC
From Skype, Maiken: @andrii - We have some jobs running in an Oracle cloud using Atlas@home on the NDGF-T1_BOINC queue in NDGF-T1 site. We would like that the jobs are accounted to NDGF/Oslo, and for this to work some tweaking must be done. This is because that CE runs jobs for many sites.
Is it possible to modify each record to set the real "site" (BOINC user) who ran the job? Or do you have some suggestions if what we need/would like can be achieved?
From Skype, Andrii: OMG, one CE that belong to several sites is definitely not in typical organizational and accounting model :) Maybe I missunderstood the architecture, because it is not completely clear what ARC CE should publish it.
If I understand correctly you are publishing to APEL and have records that belongs to different sites on the same CE.
Technically site name is added by Jura during the time it creates SSM records from A-REX accounting logs. The name is set according to 'gocdb_name' value in arc.conf and this is only one name for all records.
If you want to hack it, the automatic way I can imagine is to modify 'ssmsend' script to parse stuff in /var/spool/arc/ssm and conditionally substitute values in XML files there just before doing the intended payload (sending to APEL broker).
From Skype, David: Just to explain a bit more - if someone wants to contribute resources to ATLAS@home from their grid site, they register a username in BOINC which is the official name of their grid site (eg NDGF-T1). When a job finishes, the BOINC LRMS script adds this username to the job.id.diag file and this is then propagated back to ATLAS monitoring systems so the resources can be accounted in ATLAS accounting. What Maiken would like is for this to appear in official SGAS/APEL accounting too.
It should be ok to hack the xml in ssmsend, but this depends on the diag file being available at that moment, and I'm not sure at which point this is deleted.
All the jobs for all these different "sites" run through two CEs which currently do not publish any accounting info. I suppose there would need to be some filters for which sites should go to which accounting systems, otherwise they would be flooded with records from random site names