LHCb Core Software Meeting
Date and Location
09-May-2012
10:30 - HH:MM
CERN (2-R-030)
Attendees
Ben Couturier,
Gloria Corti,
Joel Closier,
Marco Cattaneo,
Marco Clemencic (minutes),
Markus Frank,
Thomas Hartmann (Vidyo)
Subjects
News
Round Table
Marco Cl.
- Reported ROOT bug #94227
: was causing the failure in Erasmus reported by Thomas last week
- the fix caused other failures in Erasmus, which have been fixed
- A problem in VeloDDDB has been found after a report on a mailing list. Patched version already released.
- Problem with jobs failing on old versions of ONLINE CondDB partition.
- the failure was due to the fact that the working directory of the pilot job was used to run a prompt reconstruction (downloading ONLINE) on one day, then it was re-used two days later to run a normal reconstruction (no update), which picked up the old ONLINE snapshot already there.
- there are 3 possible solutions:
- require "use SQLDDDB" in every job, but we do not need/want it
- remove "use SQLDDDB" from every job, but we have problems with the prompt reconstruction
- purge the local MYSITEROOT between two jobs in the pilot, but it's more complicated
- once we have the new CVMFS volume for CondDB, we can have a more frequent refresh, so we can get rid of "use SQLDDDB"
Ben
- Problem with a release of Erasmus, because of mkproject did not report the failure.
- it is not clear why mkproject does not behave, because the code seems correct
- New version of LbScripts, because there are 2 new packages that can be released/installed
- install_project uses a dynamic list, so it is not a problem to add packages, but other tools use an hard-coded one, and they need to be fixed
- Progress in changing the default CMTCONFIG
- the default comes from a prioritized list in Platforms.py, but a change there means a change for everybody
- to test at CERN, we can modify the CERN group login (ugly, but easy)
- we cannot change yet the default, because Ganga does not work with gcc 4.6
- Investigating the performances of Brunel, discovered a ~5% fluctuation in the CPU time used (CPU ticks), while it should be more deterministic
- a possible cause of this effect is differences in the access to the memory in different executions (because of cache misses)
- will investigate further
Gloria
- Discovered (with Marco Cl.) the reason of segfaults in the generation of configurables in GaussRICH
- service where retrieved in constructors, thus, after the splitting of GaudiSvc, invalidating iterators in genconf
- GaussRICH has been fixed (to be validated by Sajan). Marco Cl. says he can had a proper error message in genconf, instead of the segfault.
- There are still compilations problems with gcc 4.6
Marco Ca.
- Hard limits on memory on grid sites have been hitting Stripping jobs, see bug #94439
.
- We need to think about possible improvements at the framework level.
--
MarcoClemencic - 09-May-2012