AFS Phaseout - Coordinator meeting

Wednesday 1 Mar 2017, 15:00 → 18:00 Europe/Zurich

31/3-004 - IT Amphitheatre (CERN)

Discuss plans and share experiences around the AFS phaseout in the various areas at CERN.

AFS Phaseout TWiki

(see slides)

"AFS phaseout - overview + discussion"

• Disconnection test summary: still waiting for feedback from experiments and sites. Will repeat test once new EOSFUSE is there and hopefully some major users are gone, and will extend to 3..7 days. After a quick discussion, it was clarified that this should not affect BE or the CCC, and hence does not need to be tied to accelerator stops
• Account management (short-term, GID-by-default): do group admins now have to ask users whether they need AFS? Same as now, for some E-groups the AFS account gets created automatically, for others still need to create explicitly via self-service.
• Discussion on BE Java developers and their use cases (compilation with heavy dependency on AFS); new system does not seem to offer immediate advantages, and testing effort is hard to find. Also, would not want to re-find already-known issues (BE-ABP), and LS2 in general will be very busy for BE. IT expects that "development" is a generic use case between BE and experiments so can be checked upfront, but specifics indeed require separate test, effort for this shuoldbe discussed with hierarchy.
• Performance impact of updating large files - can the recycling bin feature ("backup") be tuned? Can turn of/off (after discussion), but cannot tune.
• will data be kept after AFS is shut off (e.g 6 months after start of RUN3, sometimes needed 1 year-old-data in the past)? yes, plan to keep AFS data around after shut-off (to be decided for how long exactly).
• how to indicate that for a project space the data has been copied, but will be kept on AFS until end of AFS? Just add comment in JIRA, but advise against (hides legacy use cases).
• how are AFS phaseout and EOS plans linked? /work phaseout needs new EOSFUSE, and the new EOS namespace will be required to hold all files currently in AFS
• is there an upfront specification for EOSFUSE (POSIX calls it should support, performance targets)? No, iteratively fixing use cases, but also running a POSIX compatibility test suite. Implementation goal for now is "same as AFS".

EP-SFT status and plans + discussion

• For the AFS phaseout to proceed the software would need to be actively removed at some stage from AFS, instead of old releases being left there. Per-product timescale to be agreed with the experiments.

PLUS & BATCH plans

• Will dot-files work on "lxplus-eos"? yes (but need to copy manually into EOS, and synchronizing to desktop still is being looked at).
• Timescale is for "lxplus-eos" is a few days after new EOSFUSE implementation is available. This is a test cluster to flush out non-working usecases (i.e not the future LXPLUS), testing effort is appreciated.

AOB

• for personal workstations, there is ongoing effort to integrate EOSFUSE (and CVMFS) - see e.g Linux Support "locmap" on CC7.3. However, the recommended solution is to use CERNBox and sync back files/folders into EOS.
• Next meeting?
  • the meeting was overall felt to be useful, a further meeting will be scheduled in 3months time (~June).

Action items:

• documentation for migrating websites from AFS to EOS (incl test)
• documentation for "which solution for which use case", incl desktop setups, xrootd access vs FUSE. To be adapted over time.
• clarify EOS(USER) availability on the TN - RQF0609864
• EOSFUSE: performance problem with appending to a large file - reproduce, make sure it is fixed after rewrite

 

15:00 → 15:30 AFS phaseout - overview + discussion

Speaker: Jan Iven (CERN)

PhaseoutMeeting-20170301.pdf

15:30 → 16:00 EP-SFT status and plans + discussion

current status of the AFS phaseout for SFT, and a proposal for stopping support for AFS-based software completely

Speaker: Patricia Mendez Lorenzo (CERN)

afs_phaseout.pdf

16:00 → 16:10 PLUS & BATCH plans

Speaker: Gavin McCance (CERN)

AFS-plus-batch.pdf

16:10 → 16:40 AOB