ADC Weekly

Tuesday 8 Oct 2013, 15:40 → 17:30 Europe/Zurich

3162/2-E01 (CERN)

Alessandro Di Girolamo (CERN), I Ueda (Department of Particle Physics-University of Tokyo)

15:40 → 15:45 possible delay

15:45 → 16:00 Hot topics

15:45 Managing Panda Resources at Tier-1s

Announcement to Tier-1s Managing_Panda_Resources_at_Tier1s.txt

Mail forwarded to cloud-all https://groups.cern.ch/group/atlas-adc-cloud-all/Lists/Archive/Flat.aspx?RootFolder=%2Fgroup%2Fatlas-adc-cloud-all%2FLists%2FArchive%2FManaging%20Panda%20Resources%20at%20Tier-1s

• Changing the main panda resource at a T1 needs a careful procedure.
• The information is related only to T1s
• The message is forwarded to the cloud support so that they are aware of it

15:50 GDP

Speakers: Andrej Filipcic (Jozef Stefan Institute (SI)), Dr Rodney Walker (Ludwig-Maximilians-Univ. Muenchen (DE))

Slides adcgdp-20131008.pdf

16:00 → 16:15 AMOD/ADCoS report

Speakers: Alexey Sedov (Universitat Autònoma de Barcelona (ES)), Helmut Wolters (LIP Coimbra, Portugal)

Slides 131001_AMODreport.pdf

• The issue reported as BNL proxy expiration was caused by the CERN VOMS server issues. The slide will be corrected.
• Renewal of VOMS proxy may need to include a step for checking the validity of new proxy before replacing the old one?
  • D.Cameron: rather than putting in such a check, keeping a backup and rolling-back manually would be better.
  • It is rather a problem in voms-proxy-init. It should not return an invalid proxy. A fix to be requested to the developers.

16:15 → 16:30 Monitoring jobs failing-over to FAX

Speaker: Ilija Vukotic (University of Chicago (US))

ADC Weekly, June 18 https://indico.cern.ch/conferenceDisplay.py?confId=256535

Slides FAXpanda_monitoring_for_ADC.pdf FAXpanda_monitoring_for_ADC.pptx

• At the last s&c week, it was announced that the FAX team has activated the input file fail-over to FAX for some selected sites. It triggered a discussion and it was agreed that FAX team will prepare instructions/procedures. The presentation is supposed to address this.
• Slide 2: "we would suggest all the sites and all the queues to have it on"
  • "we" means FAX team.
  • ADC-ops does not recommend/suggest to activate the failing-over to FAX before seeing results from stress tests. It is not up to the sites/clouds/FAX team to decide to switch it on
    • Joining to FAX is on a voluntary basis and up to the sites/clouds, but activating the fail-over could affect other sites, especially T1s, and should be a decision by ADC-ops.
    • The level of stress needs to be agreed offline in a dedicated discussion by mail.
• When a SE is down, what happens to the output?
  • Ilija: need to be written to another site
  • This means a need for a dev in panda/pilot
• HC (AFT/PFT) should exclude queues when SE is down, so the FAX fail-over cannot be used for this case.
  • AFT/PFT should ignore this flag.
• we need more numbers to monitor
  • total number of successful jobs, total number of failed jobs, number of successful jobs because of the fail-over, number of failed jobs despite the fail-over
  • we should also monitor the negative impacts; eg. if fax takes long time and if jobs fail after unsuccessful fail-over, it is a loss of computing resources.

16:30 → 16:35 AOB