Thursday Morning Meeting 28 September 2017

  • DIAL-IN NUMBERS & PASSCODES:
  • IP: 192.33.117.12##8110
  • Phone: (434) 817-6524

Attendance

  • Socorro:
  • CV:
  • Garching:
  • SCO:

News / Meetings / Visitors

Build, Release

  • 5.1.0 is released!
    • Issue for VLA pipeline - CAS-10771 - linear,nearestflag needs to be set in the VLA pipeline, fixed by Brian K, not in 5.1

Verification Testing

  • Akeem is back! Welcome Akeem!

Validation Testing

  • Current validation tally
    • 122 Under Validation, 87 Ready to Validate.
    • 8 tickets went RtV this week so far. 7 tickets went through Validation to Resolved this week so far.
    • Sept 8 - 14: 14 tickets assigned for validation. 11 tickets resolved.
    • Sept 15 - 21: 28 tickets assigned for validation. 23 tickets resolved.
  • Discussion item: running PL runs regularly during the build + test stage with each build to catch problems earlier? Amanda has picked two PL runs that don't take an age (one 7m and one 12m)
  • Current Testing Efforts for 5.2 deliverables
    • update from HPC/Bjorn?
  • Current Testing Efforts for 5.3 deliverables
    • imaging issues/autoboxing refinements
      • CAS-10701, making the mosaics bigger is not fixing issues with finding components in edges anymore, primary beam masking seems to be turned off?
      • CAS-10692, tclean is waffling in some cases around the threshold in 1 channel (0.001\%) for umpteen major cycles
    • statwt2 (CAS-10530)
      • Josh has been finding and reporting bugs
    • systematic tests of mstransform
      • tickets other than CAS-5174?
    • plone/CASAdocs
      • need to make some decisions on CAS-10595: last call for feedback on deprecating tasks, and I'll start moving text into plone for the rest
    • further parallelization testing (non-ALMA pipeline specific)
    • polarization calibration
    • testing snapshot -- please see https://safe.nrao.edu/wiki/bin/view/Software/CASAUserTestingfortheup-to-date list of tickets and testing status
      • testing ongoing:
        • plotants tickets (Pam, a couple last ones, including the parent CAS-10598)
        • plotms/plotbandpass capability (Pam, CAS-9053)
        • imhistory tickets (Dave, several)
        • phase center shift in plotms (CAS-8431)
        • plotms seg fault (CAS-10727)
        • cal library parser error (CAS-10736)
        • ms tool deprecation warnings (CAS-10604)
      • "next up"/looking for testers:
        • CAS-9538 Perley & Butler 2017 flux density scale
        • CAS-8270 Implement remaining plotcal functionality in plotms
        • CAS-10329 rename plot ms tool from 'plotms' to 'pmtool'
        • CAS-10696 make imfit give fit result also in pixel coordinates

Architecture

  • Warnings Summary
    • This still uses the SVN repository and is therefore frozen in time. Ville is looking in to cloning this work so that it can be updated to use git until a replacement for this script is available.
  • New warnings summary: * Warnings Summary

HPC

  • For 5.2
    • CAS-10672: RuntimeError: Error in running Major Cycle : CountedPtr: null dereference error - Under verification
    • CAS-10538 : The two imager errors are not reproduceable, and a test has been added to try to trigger one of them. RESOLVED
    • CAS-10453 : Applycal errors are a harmless(?) manifestation of using subMSs. We need to understand why the SEVERE messages are not filtered out from the logs as in the past. Maybe something in the pipeline is resetting the log filter, which then exposes the messages to the logger. If not solved in 5.2, at least it will be added to list of "differences" between serial parallel runs.
  • For 5.2 (pipeline)
    • CAS-10662: "MPI buffer error", fixed by Stewart in pipeline r40843 (we'll need this included in 5.2). Validation throughout the next tests for 5.2. The pipeline was sending over parts of the context.
  • For 5.3
    • CAS-10697 : parallel uvcontsub failure on subMS. This does not affect the ALMA pipeline, therefore strictly speaking, it should go to 5.3. Old uvcontsub was parallelised a few years ago but needs a better error handling to advise the user that when combine=‘spw’, all the Sub-MSs need to have all the selected spws. If the MMS is partitioned with the default options, most likely the Sub-MSs will not contain all the spws. The best implementation means, adding a new parameter called keepmms, in order to align uvcontsub with the other tasks such as split, which also can create an MMS.
    • CAS-10615 : Info about partitions for the pipeline. This was not requested by the pipeline. It is a convenient way to show the contents of an MMS when partition/importasdm finishes to run. There is already a task called listpartition which does this.
  • Discussions with Bjorn on his VLA HPC testing. His tests show slowdown when running gaincal and setjy, both in serial mode on MMS. TBV
  • Preparing for tests with CASA 5.2 prereleases: https://open-confluence.nrao.edu/display/CASA/ALMA+Parallel+Pipeline+Testing. Adapting analysisUtils/picklePipeRun to compare parallel/serial runs.
    • Expanding performance metrics and stats to identify main slowdown points in parallel mode, including non-CASA-tasks pipeline stuff.
    • Approximate priority ranking:
      • 1 hif.heuristics.imageparams_base, doing long sequences of "setDataOnThisMS() + advise()", can slowdown from ~1h to ~11h (dataset T008), or even from ~4 to 3d+ (dataset T012)
      • 2 infrastructure.displays.sky and (to a much lesser extent) qa.scorecalculator. This is for producing weblogs. Example, displays.sky, from 2h20m to 23h10m, T025.
      • Lesser slowdown-ers: infrastructure.basetask, infrastructure.mpihelpers, hif.tasks.findcont.findcont, and others: hif.heuristics.cleanbox hif.tasks.tclean.cleanbase.
    • Next days: create a small number (2-4) of focused tickets for discussion, with aim fixVersion=5.2, if feasible.

Development

  • CASA 5.2 : 8 issues being tracked. 5 CASA-specific ones are below. ( Need not discuss if already covered by the HPC group's report )
    • CAS-10672, CAS-10538 : The two imager errors are not reproduceable, and a test has been added to try to trigger one of them.
    • CAS-10453 : Applycal errors are a harmless(?) manifestation of using subMSs,
    • CAS-10697 : parallel uvcontsub failure on subMS ?
    • CAS-10615 : Info about partitions for the pipeline ?

  • CASA 5.3 :
    • Iteration control + Autoboxing : Plans have been set up for a clear division of work for R&D vs production, as well as how to handle overlapping features. This is to avoid the issues we have had over the past year on this topic.
    • statwt2 : Some dev in response to testing. Some updates to interface requirements.
    • ALMA has done it's planned changes to the SDM (new tables - CalAntennaSolutions, Pulsar, more cols to the CalBandpass table + a channel column to the flag table (i.e. a new Flag.xml tag which will require edits to our parser used by imports and flagger). Bob - any surprises yet ?
      • No. But I haven't looked in detail. It will take some time before this progresses through the ALMA acceptance tests and shows up in the main trunk. Also, ALMA's move to git is happening now. So, I'm going to wait awhile (probably about a month) for the dust to settle before I look at this. The related CASA tickets (CAS-10386, and CAS-10422) are for 5.3 with the requirement that they be available for testing in a 5.3 pre-release by Feb. 2018. I see no problems with meeting that timeline at this point.
    • Polarization Calibration dev update from George ?
    • Imager : new bugs to be investigated : mosaic sidelobe/wrapping, two spectral coordinate system generation issues.
    • Anything else ?

Pipeline

  • CASA 5.1 / Pipeline
    • Out the door
  • CASA 5.2 / Pipeline
    • Pipeline plans to branch for 5.2 after today's meeting
    • First pipeline HPC issue fixed on pipeline trunk, will be backported to HPC branch
  • CASA 5.4 / Pipeline
    • Development has resumed
    • ALMA heuristics development plan is conservative and evolutionary
    • ALMA heuristics ticket will go into jira shortly, framework tickets are already there
    • VLASS heuristics for single epoch imaging will require effort

AOB

Developer Reports

Thursday Meeting
  • Sanjay Bhatnagar
  • Sandra Castro
  • Lindsey Davis
    • Heuristics planning meeting with PWG
    • Ongoing CASA callibrary tests, problems with data / callibrary selection
    • Pipeline refactoring, code cleanup work
  • Bjorn Emonts
  • Pam Ford
    • 3 tickets under validation (CAS-10727 plotms add/close plot seg fault, CAS-8270 plotms axis label bug with cal table overplot, CAS-10736 cal library parser bug)
    • 1 pull request merged (CAS-10750 plotms constant name conflict)
    • 2 tickets responded to validation bug reports (CAS-9053 atm/tsky overlay, CAS-10604 ms tool deprecation inline help)
    • Developer builds on OSX 10.11 (cloned git code), OSX 10.12 (upgraded laptop from 10.11)
  • Enrique Garcia
  • Bob Garwood
    • CAS-10738 - Review the existing ephemeris handling code in asdm2MS and fix found bugs. I'm now testing these fixes to make sure that the examples I know of behave as expected. An entire use case in the code was blocked by poor command line argument handling and it may not exist in any data to date, but it's also not excluded by the ASDM so I've revived that code. Any suggestions for ephemeris data to test should be sent to me.
    • Discussions on the planned move of the ASDM and generated and shared code to a separate model. The full move is being deferred until ALMA developers are more comfortable with git and related changes to the work flow. In the interim I will reorganize the code within the ALMA code tree so that when re revisit this issue next year the actual move to a separate package and git submodule will be relatively simple. This reorganization will also involve the work on removing the boost dependency in the shared code (at least for CASA builds) - CAS-10693.
    • CAS-10732 and CAS-10665 - both dealing with parts of importasdm being slow. I'll likely spend more time on this in the next few weeks trying to understand where the filler spends most of it's time and looking for low-hanging fruit.
    • Some GBT work: accepted some contributed gbtgridder code that adds a new "eqweight" option. Some anonymous person created a new JIRA user and submitted a GBT post processing ticket saying "did you know anyone can create a new user account and submit a ticket?". Discussions ongoing on whether that's a feature or a bug and if it's a bug, what to do about. GBTPP-15.
  • Kumar Golap
  • David Mehringer
  • George Moellenbrock
    • Minor newsletter article revisions
    • Finished building CASA on loaner OSX10.12 laptop! Relatively smooth compared to previous experience. Thanks Ville!
    • MSv3 suggestions submitted to Kumar and Martin
    • Cal library discussions with Lindsey
    • CAS-9108 (linear feed Xf refactor and related): work continuing
    • CAS-10772 (repeated use of bandpass+applycal causes creeping flags; see also CAS-10771): testing fix now, soon ready to verify and validate
    • Some travel prep (@JIVE/ASTRON next week for CASA/VLBI workshop)
  • Dirk Petry
  • Martin Pokorny
  • Federico M Pouzols
    • Timing of pipeline tasks in serial versus parallel test runs.
    • New test scripts and adapting analysisUtils/picklePipeRun to compare parallel/serial runs.
  • Urvashi Rao
    • Numerous emails/discussions on various topics (primarily R&D vs production issues w.r.to the PLWG)
    • Began work on CAS-10768 : step 1 of a resource predictor for imager.
  • Darrell Schiebel
  • Ville Suoranta
  • Takahiro Tsutsumi
    • Started timing analysis of some of underlying functions used in automask (CAS-10743)
    • Some science time (gathering information needed for responding the referee’s report for a paper that I am co-authoring)
Friday NAOJ Meeting
  • Kanako Sugimoto
  • Wataru Kawasaki
  • Masaya Kuniyoshi
  • Takeshi Nakazato
  • Renaud Miel

-- MorganGriffith - 2017-09-20
Topic revision: r18 - 2017-10-03, VilleSuoranta
This site is powered by FoswikiCopyright © by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding NRAO Public Wiki? Send feedback