2023-06-08 Meeting Notes

Next steps for WN deployment

Planning for rollout; Anticipated for next week - Needs @Thomas Birkett to confirm.
June 2023: Vector Read rollout (with XCache)

Tentative start date of June 7th, rollout is simpler compared to previous attempt (no drain required)

• Final steps:

  • Created RPM with final updates ?

  • tested Docker / xrootd configurations ?

  • other concerns ?

Alex: concern: to ensure that configuration is correct.
Confirmed on the one WN: lcg2268. (based on GitHub PRhttps://github.com/stfc/grid-workernode/pull/38/files

http://aquilon.gridpp.rl.ac.uk/sandboxes/diff.php?sandbox=workernode-xrootd

EBUSY in readV requests

https://stfc.atlassian.net/browse/XRD-64

Actions to analysis the existing behaviour with the read request (where EBUSY) is already handled.

SEGV investigations with -S multi-stream flags

@Ian Johnson set ‘git bisect run’ going with a shell script to compile, install and restart XRootD, and test a multi-stream transfer. This identified a problem in re-using the the initial build environment at each step, should be easy to resolve.

CMSD status

https://stfc.atlassian.net/browse/XRD-41

single slash behaviour with cmsd redirector

“/” vs “//” issue:

• Characterise the behaviour

• Understand if this is something we need to provide some mitigating action on?

Options:

1. Insist clients use “filesystem” like access (i.e. //) for root

2. Allow mixed use filesystem and object id access. Add all.export <pool> lines for each pool

   1. can even remove the “/” export and add explicitly:
      (/store would stay as /store)

      all.export <pool:> all.export /<pool:>

3. Understand / create / fix(?) oss.statlib functionality to behave as expected from xrd-dev perspective