MUPJ

--Chollet 11:41, 27 janvier 2010 (CET)

Support of Multi-user pilot jobs (MUPJ) for LHC VOs

In December 09, the WLCG Management Board asked the WLCG Technical Forum to provide recommendations on the use of multi-user pilot jobs, in particular on the deployment and configuration of glexec and its usage by the experiment frameworks.

• Document that describes what pilot jobs are, their benefits, disadvantages and other aspects that sites and experiments may need to be aware of, is available here : https://wlcg-tf.hep.ac.uk/wiki/Multi_User_Pilot_Jobs (IGTF certificate required)

• "Traceability and Logging" and "Grid Pilot Jobs" policies available at http://www.jspg.org/wiki/JSPG_Docs

• Questionnaire for site admins : https://wlcg-tf.hep.ac.uk/wiki/Multi_User_Pilot_Jobs#Questionnaire

• See T2-T3 Technical Meeting, January 22
  • Pierre Girard's presentation (restricted access to t2t3 group)
  • Collective answer from sitessent to Maarten Litmaath has been attached to the meeting agenda as well (restricted access to t2t3 group)

• Status report@March 2010 GDB

- MUPJ in principle supported
- Most of the sites/resources compatible with glexec setuid 
    o many sites require it in the medium terms
    o strongly preferred by CMS and ATLAS
- Many sites prohibit bypassing glexec on internal errors (with some tolerance for the time being)
- Site concerns :
     Batch jobs clean-up (complications of), some incompatibilities with afs in setuid mode, 
     how can local users gets special share/priorities ?, strong concerns about MUPJ concept expresseb bu NorduGrid 
- Proposed actions :
    o sites should configure glexec with SCAS/Argus/GUMS backend and advertise it with GlueCECapability : glexec
      glexec/SCAS testing chain in validation - new ops/role=pilot - will be integrated in the next release of national nagios
    o experiments should try out their glexec workflows at such sites as they become available
    o sites that do not support the setuid mode would no longer receive MUPJ 
    o others technologies (such as virtualization) may affect medium-term strategies
- In conclusion, MB agrreed that a detailed plan for glexec deployment in setuid mode in WLCG sites is required
- Milestone for T0/T1 sites

Links

• Glasgow wiki page (glexec and SCAS installation and configuration)

https://www.scotgrid.ac.uk/wiki/index.php/Glasgow_GLite_gLExec_installation_and_configuration

• http://glite.web.cern.ch/glite/packages/R3.2/sl5_x86_64/deployment/glite-GLEXEC_wn/glite-GLEXEC_wn.asp