Skip to content
Snippets Groups Projects

Software Shifter Summary 03/02/20 - 16/02/20

    • Closed Issue created by Ross John Hunter @rjhunter

    Shifter: Ross Hunter, University of Warwick (@rjhunter)

    N.B. These notes are based upon the notes I took during my shift, which ended on the evening of Friday 14th. I have tried to take account of all updates since, but I may have missed a couple of changes in the status of MRs (e.g. a couple of this list may have since been merged), but I expect it to be ~95% correct.

    Summary

    The shift went fairly smoothly during the last two weeks. The nightlies went down for one night after the MooreAnalysis project was added to the builds. This failed because Jenkins tried to check it out from svn rather than gitlab by default - this has now hopefully been fixed (symptom + cause), thanks to @apearce and @clemenci!

    Across LHCb, Lbcom, Rec, Brunel, Phys and Moore, 61 MRs were merged to master since this shift was started.

    The master branch is healthy for the main RTA projects on x86_64-centos7-gcc9 and x86_64-centos7-clang8, but there are a few issues on the x86_64+avx2 platform.

    Thanks to @rmatev for all his help as maintainer, and to all the developers who helped me a lot by providing good descriptions and efficient and effective discussions. @peilian is the new shifter.

    Shifter's highlights

    • The lhcb-test-efficiency slot was added to the nightly builds, which is a place for dedicated testing of the MRs that relate Allen with the rest of the LHCb stack. For instance, various MRs to do with Gaudi-Allen integration (e.g. Allen!281 (merged)) and MRs related to the HltEfficiencyChecker tool (e.g. mooreanalysis!1) have been included in this slot so far.

    • @olupton's custom memory pools (see LHCb#73 (closed)) were merged to master, which required a big effort from a group of people in testing and debugging obscure compiler issues. There was a satisfying frenzy of merging at the end, and we expect to see big performance improvements from this.

    • DecReport keys now end in "Decision" rather than "Line" for historical reasons (e.g. "Hlt1SingleHighPtMuonLineDecision" - the "Line" is to be removed.

    • Implementation of a second loop in the SciFiTrackForwarding algorithm for extending VeloUT tracks to the SciFi, achieves around 8% gain in reconstruction efficiency of long tracks.

    Summary of Open MRs

    There are currently 115 open MRs targeting master, with 66 marked as WIP. Approx. 15 are ready for merging.

    Ready to Merge

    (all assigned to maintainer for merging unless explicitly noted otherwise)

    MRs in nightly testing (see here

    MRs needing review from coding/physics expert

    MRs which need testing

    MRs that need releases (the release each one needs is indicated)

    Needs new references

    Open MRs adding new references

    Active non-WIP MRs to keep an eye on.

    Thinks to not forget about

    • Errors in LbNightlyTools that lead to the nightlies failures: 1) the summary page looks for when things "end" to give you status. The bit that tells you it has ended is at the end of the script. So if the nightly build fails (e.g. at the point it is checking-out the projects), the summary page will tell you it is "running" rather than "failed". 2) If there is an issue in checking-out, it shouldn't crash. It should exit gracefully the failing MR/project and try to do the best it can with the other projects. See lhcb-core/LbNightlyTools#55.
    • Moore test_hlt2_example.py times out a lot. May have already been fixed in !400 (merged).
    • Should we have specific references for the Skylake and avx2 platforms? There are a few extra failures in these platforms when compared to the vanilla x86_64. See last night's build here for instance.
    • I don't think that WIP MRs show up in the maintainer-app if you put nightlies labels on them - this is a bug if it is always the case. Certainly one that I put there didn't show up.
    Edited

    Designs

      Child items ...

      2. Activity

      Please register or sign in to reply