Sync meeting 2023 11 20 with CernVM FS developers on Best Practices for CernVM FS on HPC tutorial

Sync meeting (2023-11-20)

Attending

• CernVM-FS: Laura, Valentin, Jakob
• EESSI/MultiXscale: Alan, Lara, Kenneth

Notes

• 132 people have registered for the online tutorial on Mon 4 Dec'23 so far...
• PR #13: What is CVMFS?
  • terminology moved to separate appendix
  • pages renamed (- not _)
  • links to CVMFS docs for unpacked.cern.ch
  • Ready for final review + merge
• PR #12: EESSI
  • split up into separate pages
  • needs more revision work
• HPC section
  • see also https://cvmfs.readthedocs.io/en/stable/cpt-hpc.html
  • diskless
    • CVMFS cache in ramdisk
  • alien cache is unmanaged (no cache eviction)
    • can be read-only or read/write
    • diskless workernodes => alien cache on shared FS
  • preloaded cache (orthogonal to alien cache)
    • often used with alien cache
  • promote cvmfs-exec
    • requires unpriv user namespaces
    • useful if CVMFS is not installed CVMFS system-wide
  • sync to another FS
    • shrinkwrap
• performance section
  • see notebooks:
    • https://github.com/boegel/cvmfs-tutorial-hpc-best-practices/blob/perf/cvmfs_perf_python_hpcugent.ipynb
    • https://github.com/boegel/cvmfs-tutorial-hpc-best-practices/blob/perf/cvmfs_perf_tensorflow_hpcugent.ipynb
  • test GPFS
  • test CVMFS with cache in ramdisk
• troubleshooting section
  • starting to collect ideas/structure
  • cvmfs_config chkconfig to check client config
  • logging
    • can configure syslog facility
    • CVMFS_USYSLOG to filename
    • access syslog via xattrs (ony if repo is mounted)
  • proxy
    • make sure that proxies are allowed to connect to S1's (in proxy settings)
      • usually targets are restricted
      • common issue, for example if WLCG is already used
  • for revision
    • see xattrs
  • cURL debug command to download .cvmfspublished manifest file
    • cURL treats this as binary file, so be careful
    • -I
  • config
    • order of files being considered
    • default.conf, default.local, config-repo, config for repo
    • showconfig shows actual config + where it was set
  • cache corruption => in monitoring section
• containers
  • TensorFlow container?
  • via CVMFS: no full download, no conversion, managed cache
  • dedup benefit when using multiple container images
• other stuff
  • dedicated URL for full replication of clients
• next sync meeting Tue 28 Nov'23 - 10:00 CET
• last-minute sync on Mon 4 Dec'23 - 10:00 CET