Ceph/CVMFS/Filer Service Meeting

Europe/Zurich
600/R-001 (CERN)

600/R-001

CERN

15
Show room on map
    • 14:00 14:05
      CVMFS 5m
      Speaker: Enrico Bocchi (CERN)
    • 14:05 14:10
      Ceph Upstream News 5m

      Releases, Tickets, Testing, Board, ...

      Speaker: Dan van der Ster (CERN)
    • 14:10 14:15
      Ceph Backends & Block Storage 5m

      Cluster upgrades, capacity changes, rebalancing, ...
      News from OpenStack block storage.

      Speaker: Theofilos Mouratidis (National and Kapodistrian University of Athens (GR))

      ceph/flax: now 75% bluestore

      ceph/erin: upgraded to mimic

      nfs-ganesha:

      • created a vm with a working example
      • trying to figure out kerberos auth
         
    • 14:15 14:20
      Ceph Disk Management 5m

      OSD Replacements, Liaison with CF, Failure Predictions

      Speaker: Julien Collet (CERN)

      Julien

      • Disk replacement (almost) fully handled by Paul and Remy now
      • Modified scripts to handle updated procedure.
    • 14:20 14:25
      S3 5m

      Ops, Use-cases (backup, DB), ...

      Speakers: Julien Collet (CERN), Roberto Valverde Cameselle (Universidad de Oviedo (ES))

      (Dan)

      • gabe cluster is 80% full -- need to add spare servers to this cluster and balance (change crush rule, enable balancer).
        • default.rgw.meta (obsolete pool) has been deleted (removing 10M objects)
        • script to clean ~13 million obsolete bucket indexes is running now on cephgabe0.
          • all have size=0, num omap keys = 0
          • 347 from 2019, 2975 from 2018, 13.65 M from 2017.
        • Roberto paused cbox backups (60TB used).
        • Kubernetes RECAST Higgs demo using 70TB until mid-May.
        • Deleted 7TB of my test data, 9TB of old Oracle tests.
        • CEPH-697 S3: Contact legacy users, move to openstack accounts
      • S3 outage this weekend: https://cern.service-now.com/service-portal/view-outage.do?n=OTG0049284
        • An OSD had locked up which caused 404's in the traefik lb's. Restarting the OSD unblocked the gateways.
        • (Likely related to the obsolete bucket index cleanup, and compaction in the osd leveldb's. I re-compacted all OSD's after this and resumed the index deletion this morning.)

      Julien

      • Updated TLS certificates (see CEPH-670) with Dan
      • Started S3 multi-region preparation (see CEPH-682) with Roberto
    • 14:25 14:30
      CephFS/Manila/FILER 5m

      Filer Migration, CephFS/Manila status and plans.

      Speaker: Dan van der Ster (CERN)

      (Dan)

      • https://its.cern.ch/jira/browse/CRM-3101: (Fixing a small but which created file /root/1 instead of piping to /dev/null)
      • Ceph/flax remaining FileStore machines clearly more highly loaded now that 75% of cluster is bluestore. I caught some live OSD flapping on Sunday and found that disabling deep scrub on those osds helped.
    • 14:30 14:35
      HPC 5m

      Performance testing, HPC storage status and plans

      Speakers: Alberto Chiusole (Universita e INFN Trieste (IT)), Pablo Llopis Sanmillan (CERN)

      (Dan)

      • Attended HPC Users Workshop: https://indico.cern.ch/event/805674/
        • Users happy to get more perf from kernel mount.
        • One user asked how they could access cephfs from lxbatch. (We explicitly do not want to mount on lxbatch.) They can copy files between via EOS -- they will get back in touch if they need something else.
          (Once we have NFS-Ganesha working, would could enable some low-performance access to the HPC files from outside the cluster.)
    • 14:35 14:40
      HyperConverged 5m
      Speakers: Jose Castro Leon (CERN), Julien Collet (CERN), Roberto Valverde Cameselle (Universidad de Oviedo (ES))

      (Dan)

      • CEPH-698 kelly: redeploy osds with bluestore_min_alloc_size=4096

       

      Julien & Jose

      • CEPH-688: testing of cache configurations is ongoing

       

    • 14:40 14:45
      Monitoring 5m

      Julien

           - Prophetstore: one disk failed "driver error count value" and was not predicted by prophetstore as bad...

           - Currently investigating what had the alarm triggered.

    • 14:45 14:50
      AOB 5m

      Kopano:

      • CephFS will be used for attachments.
        • Currently all attachments stored. They will test dedup.
        • Most attachments around 4kB.
        • Currently 2 levels nesting -- [0-10]/[
      • Current PoC HW is HC machines (14*1TB SSD * 20 servers)
        • CDA want all HW to be in the barn. Need to find space and probably new hardware.
        • Eric Bonfillou will get price for 4x4TB SSD that could go into new arriving batch nodes.
      • Ceph Kelly is now dedicated to Kopano (including DBoD tests)
        • CTA can stay there because they're small, exceptionally.

      ceph/scripts: Change scripts to be closer to Python3 and avoid linter warnings.

       

       

      Julien:

      • Conference call last Thursday with the French Ministere de l'interieur.
        • Want to/are operating a Ceph cluster and are really interested in having tips/help/guidance
        • Planning to come mid-june for a day:
          • They will present they architecture/use-case/etc...
          • We'll probably do a "Ceph at CERN" sort of day
      • Confcall with UNIGE later this week