Page Comparison

Vector Read updates to Echo

/wiki/spaces/GRIDPP/pages/188153912

Post-mortem to be held today

Next steps for WN deployment

Move to 5.5.4-2 (core) + 5.5.4-3 (xroot-ceph-buffered):

https://github.com/stfc/xrootd/tree/v5.5.4patched

https://github.com/stfc/xrootd-ceph/releases/tag/5.5.4-3

Status:

lcg2268 running with that config. n_connections from xrdceph file close metrics show similar numbers compared to other nodes in 5.3.3-2 without readv

Image Added

Image Added

No failures of lhcb jobs were observed on the test WN so far.

EBUSY in readV requests

Discussion on merging bufferedIO into master.
Also to discuss pushing changes to “upstream” (xrootd/xroot-ceph”

https://github.com/stfc/xrootd-ceph/pull/44

Needs testing for ‘correctness’

Also some discussion ongoing on xrootd “issues” on the xrootd-ceph sub-module:
https://github.com/xrootd/xrootd/pull/2008

SEGV investigations with -S multi-stream flags

Ian Johnson Preparing script for ‘git-bisect run <script>’ to run, automating the search of a range of commits. This should identify the commit that introduced the unexpected corrective action that now allows multiple streams in XrdCeph without the SEGV. Guilherme commented in https://github.com/xrootd/xrootd/issues/1821#issuecomment-1558733567 about reading release notes for a clue; I tried and didn’t find anything relevant.

CMSD status

Thomas, Jyothish (STFC,RAL,SC) identified that xrdcp via the redirector using root protocol requires two “/” in the LFN to work correctly.

• Although this is the nominally correct path construction for xroot-based paths; existing transfers work with just a single-/ and may break from some corner-case person/VO.

Possible solution is Assuming we want a solution which maintains backwards compatibility, we may need to explicitly add:

all.export cms #and other pool/ # already existing line
# add in the lines
all.export cms
all.export store #special cms 'namespace' 
# and for other pools

What’s the impact? Do we need “/” export now. Should we consider the authdb file logic too ?
To understand exactly how the export / authdb / “// | /" finally interact ?

Transfers of 0-byte files

Observed Dune transfer failures using 0-byte files/

Thomas, Jyothish (STFC,RAL,SC) identified FTS not supporting 0-byte transfers, hence the cause of the reported failures.

Ian Johnson noted that the “classic” RADOS striper implementation doesn’t appear to create an object until data is written to it; is there a use case which might call for the creation of 0-byte files?