LHCb Core Software Meeting

Date and Location

12-Dec-2012
10:30 - 11:30
CERN (2-R-030)

Attendees

Ben Couturier, Gloria Corti, Illya Shapoval, Joel Closier, Marco Cattaneo, Marco Clemencic (minutes), Markus Frank, Stefan Lohn, Thomas Hartmann (Vidyo)

Subjects

News

• Last week, WLCG meeting presenting the migration from Savannah to JIRA.
  • We need to collect the list of projects to migrate, the history that we want to keep/drop, etc.
  • Gloria have some feedback: improve the management of the privileges (for project admins), a tutorial is needed.
  • Marco Ca.: ROOT is being migrated for testing, we should wait this exercise to be completed before doing anything.

Issues from last meeting

• Coverity nightly build still not working.
  Thomas: testing, but not much progress yet.

Round Table

Marco Cl.

• Doxygen documentation has not been produced for a while (since release of LbScripts v7r3, 1st of November).
  Bug fixed in the release area and committed.
• Progress on CMake configuration
  • committed configuration to Boole
  • ignored Hlt: does not build with latest Phys
  • ignored Moore: depends on Hlt
  • Panoramix is highly non-standard: complex validation
  • Bender: built but not validated
  • Online: complex custom settings
• Progress on Nightly build system based on Jenkins + CDash
  • building a test slot every day since Sunday
  • Jenkins configuration essentially stable
  • the script to checkout and patch all the sources for a stack is mandatory to move on
  • CDash is not perfect: few bugs to be fixed and, possibly, customization of the templates
• Considerations on getpack
  • recursive checkout is extremely slow: LHCb takes 10-15 min, while a simple checkout of the trunk < 1 min
  • we may create global tags like the ones in Gaudi (a script is needed and can be done for old projects too)
  • another possibility is to start from the checkout of the trunk and apply changes on top (TASK:38587)
  • we may cache the official checkouts as tarballs; similar to install_project, but integrated in getpack and without the need for a release
• Added the possibility of checking out using a date-time as version to getpack.

Ben

• Continued the clean up of the tarballs.
• Joel had some problem yesterday, but it seems that it cannot be reproduced anymore.
• Profiling of libm showed that it is really the problem.
• Intel VTune 2013 segfaults on SLC6

Marco Cl.: can we have the slc6-gcc46 LCG tarball.
It was delayed because of CASTOR.

Stefan

• Making VTune profiling results available in Emmanuil's porject.

Gloria

• Sort of regular hick-ups in the release: failing builds.

Marco Ca.: A new release procedure was already agreed (based on the nightlies), and we need to move on with it.

Ben: We need to do something for the short term.

• Problem in the production because of a left-over (--tag_add) for old version of Gauss. Now fixed.

Marco Ca.

• Stefan + WLCG started a campaign to have CVMFS on all the grid nodes: after April, all the sites must have CVMFS (we could even start earlier).

Illya

• Execution time of publishing SQLite files exploded to 30min (publishing ONLINE only) to 1h (ONLINE+LHCBCOND). The cause is that we now have 14GB of SQLite files (11 of which are the global tags). Liang profiled a bit: the main problem is AFS I/O
  • implemented some publishing process optimizations to reduce the execution time to ~3-6min (ONLINE)
• Now concentrating on the two other systems.
• Created a service account for CondDB management tasks.

Marco Ca.: what about compressing?

Markus

• Finished new buffer manager for the Online: client-server approach instead of shared memory
  • performances seem to be enough even for after LS2

Liang

• Added scripts to control the publishing procedure of CondDB. Described in TWiki.

Thomas

• Contract finishing at the end of the year.
• Cleaning up the nightlies scripts and documentation.

---

-- MarcoClemencic - 12-Dec-2012