BlueJeans: https://bluejeans.com/426716450

Attending

Ian Sullivan , Unknown User (cmorrison) , Krzysztof Findeisen , John Swinbank , Unknown User (emorganson), Unknown User (gkovacs)

Table Round

  • Ian Sullivan
    • Profiling of DCR code; 80% of the time goes to numpy.shift(), which Ian Sullivan is unsurprised by.
    • Good news that we are not wasting time in e.g. Butler.get().
    • Used gprof2dot in conjunction with --profile on command line tasks.
  • Unknown User (cmorrison)
    • Waiting for Gabor review on DIAForcedSource.
  • Krzysztof Findeisen
    • Nothing to report.
  • Unknown User (emorganson)
    • Been gone for two weeks.
    • Did some general profiling of the AP pipeline.
    • On newer systems (similar to Ops) see about ~100s per DECam CCD.
    • No single tall pole to focus optimization on.
    • Convolving is the most expensive single operation.
    • Discussion of the -j parallelization option.
  • Unknown User (gkovacs)
    • Working on False +ve sprint.
    • Dealing with hard-wired assumptions on ap_pipe while processing DECam instcals.
    • Wrapping ap_association to run from the command line.
    • Expect a presentation or report of some sort late this week.

Metrics

  • Observe a significant increase in ProcessCcd runtime, and an overall slight slowdown across ap_verify.
  • Krzysztof Findeisen wonders if this is a systematic Jenkins issue, rather than being related to changes in AP code. He will investigate other Jenkins jobs to see if everything is getting slower, and, if so, follow up on #dm-square.
  • Colin Slater wonders if there is a temporary issue at LDF.
  • Given the above considerations, we agreed to review this issue next week and see if it continues before taking further action.

AOB