AFS weekly meeting

Monday 5 Dec 2016, 14:00 → 15:00 Europe/Zurich

31/1-012 (CERN)

AFS service/operation meeting

AFS ops meeting 2016-12-05

present: Kuba, Jan, Massimo

excused: Dan

 

Issues

• Wed morning: fileservers and DB servers updated to 1.6.7cern3 (fixes a crash in )
• OpenAFS-1.6.20 released, yet another info disclosure via uninitialized buffers being sent back to the client (low-prio)
• 3 VM (aux) migrated by cloud people (JvE), no issues with migration, but "ROGER --duration" reverted to some totally un-alarmed state from a few months back. Ticket opened.
• SELinux AVC alarms (finally) gone, configured via puppet, manually re-labelled all affected filesystems

Discussion:

• may not be able to keep current HW all the way to 2020. Jan would like to go for VMs+CEPH in case we need to move. Alternative is to use much fewer much bigger physical machines.
  • need to make sure VMs work - OK experience with afs910; test with 10VMs and track time spent.
    • if SLC6 on VM - can use current production openafs version
    • if CC7 on VM - need to either forward port Rainer rescheduling patch (broken from 1.6.7 to 1.6.9), or go upstream (no more rescheduling)
  • VM size constraints: should "fit" 1024 threads, ~12GB VIRT mem  ; figure out req. number of partition == block devices (constrained by I/O ops; parallel salvage; but fewer means less fragmentation wastage)