RAL Tier1 weekly operations castor 21/06/2019
From GridPP Wiki
Revision as of 09:50, 21 June 2019 by Rob Appleyard 7f7797b74a (Talk | contribs)
Contents
Standing agenda
1. Achievements this week
2. Problems encountered this week
3. What are we planning to do next week?
4. Long-term project updates (if not already covered)
5. Special topics
6. Actions
7. Review Fabric tasks
1. Link
8. AoTechnicalB
9. Availability for next week
10. On-Call
11. AoOtherB
Achievements this week
- LHCb decommissioning ongoing.
- Wrinkle over files which are on both d1t0 and tape.
- Odd behaviour of xrdadler32 for files that have become corrupt on disk explained.
- If there's a checksum in the extended attributes, it will return that instead of actually checksumming the file.
- Agreed to stop the on-disk checksumming.
Operation problems
- ATLAS are periodically submitting SAM tests that impact availability and cause pointless callouts - Currently with TA
- New Facilities headnodes called out. They are not supported 24/7, and weren't in production, so this was doubly wrong!
- The reasons were twofold:
- CASTOR team asked for the machines to have the same Nagios config as Tier 1 headnodes, but neglected to mention that these machines should not call out.
- The person who created the config forgot to put them in downtime.
- The reasons were twofold:
Plans for next few weeks
- Proposal: Move VCert into the Facilities domain so we have a facilities test instance.
- Decommissioned lhcbDst; hardware awaiting retirement.
- Kevin has done some storageD functional tests with the new tape robot
- Castor tape testing has started again for CEDA
Long-term projects
- New CASTOR WLCGTape instance.
- Migration of name server to VMs on 2.1.17-xx is waiting until aliceDisk is decommissioned.
- CASTOR disk server migration to Aquilon.
- Need to work with Fabric to get a stress test (see above)
- SL7 VM headnodes need changes to their personalities for the facilities.
- SL7 headnodes are being tested by GP
- Implementing DUNE on Spectralogic robot is paused.
- Migrate VCert to VMWare.
Actions
- AD wants us to make sure that experiments cannot write to that part of namespace that was used for d1t0 data: namespace cleanup/deletion of empty dirs.
- Some discussion about what exactly is required and how this can be actually implemented.
- CASTOR team proposal is either:
- to switch all of these directories to a fileclass with a requirement for a tape copy but no migration route; this will cause an error whenever any writes are attempted.
- to run a recursive nschmod on all the unneeded directories to make them read only.
- CASTOR team split over the correct approach.
- Problem with functional test node using a personal proxy which runs out some time in July.
- RA met with JJ, requested an appropriate certificate.
- Follow up with JJ or ST next week
Staffing
- GP is back on the 24th.
AoB
On Call
RA on call (tentative)