RFE: better LXPLUS-specific docs that do not rely on AFS?

[jmuf]

(CERN AFS support here..) We occasionally have CMS users asking for more AFS space for running HiggsDNA. Could you perhaps document a "best practice" for your users at CERN that

• does (as much as possible) not rely on conda or pip installations in their homedirectory?
  • since we've seen that parallel access from batch jobs will kill the AFS server, and get the user's account blocked
  • instead please promote the use of some suitable CVMFS area, and/or request the shared tools to be installed there
• directly uses EOS for storing any input+output data (parquet files, ROOT etc)
  • in particular could use https://batchdocs.web.cern.ch/local/file_xfer_plugin.html or the CMS Tier3 /store
• does not rely on having HTCondor submit files+logs (on AFS) in the same place as the code and data files

If this is not possible, could you please internally escalate to CMS computing? Thanks in advance

[jmuf]

Just to say that while the tutorial still promotes conda (i.e filling up and overloading a shared filesystem homedirectory), the HIggsDNA docs do (nowadays) indeed have alternatives:

• https://higgs-dna.readthedocs.io/en/latest/installation.html#docker-singularity for using ready-made images (nice!)
• https://higgs-dna.readthedocs.io/en/latest/jobs.html#lxplus-vanilla-submitter for running in sandbox mode (data on EOS, HTCondor logs to be fetched on demand)