Versions Compared

Key

  • This line was added.
  • This line was removed.
  • Formatting was changed.

...

  • Primary conflict - feeling that community was "promissed ad hoc user batch "
    • could wrap any code in a pipeline task which grabs from butler puts fits on disk and runs the user code.
    • it does not have to python if the binary is in the container.
  • Colin would seperate out bullet 4 from 223.
    • for provenance you have to use pipeline frame work (Tim agrees)
    • if you want to run over bunch of files just use slurm - no provenance and stuff
    • Richard PanDA can do this and things like simulation (that will use all the processing).
    • All user batch must go through TAC .. this leads to better support in our framework - they need SLAC account etc. Frossie - small birthright is not useful. So all you get be default is notebook.
  • KT arbitrary code requires arbitrary access to data
    • we could force them to wrap in pipeline task
    • Frossie points out we do not have support for arbitrary stuff - should prototype pipeline task as a wrapper.
    • staff tools may not be portable to external users
    • provide fund for googel resource ..
    • bulk export to another center - fairly rigid
    • forcing to pipeline task does not save us much in terms of the arbitrary code and authentication ...
  • SRP - whats the requirement - only one seems to be user provided code
    • lots of problems with user provided code - more restrictive better .. 
    • Support is actually not that high for big stuff - the users do this. (May not be the case for first timers from undrepresented institutes)
      • the documentation is really really good though ..
  • Frossie - TAC will decide what the user can do even simulations.
    • need to separate what allows us return the requirements
    • impression on the community (the slide decks of promises)
    • should go for the pipeline wraper ..
    • would be great to get the data to TACC or some place.
    • 10% and relative breakdown .. not fungible in hybrid model .
    • We provide adhoc arbitrary RSP type acess birtghright
      • user batch has to be in our DRP framework AT SLAC
      • access to outputs should be provided
  •  run sextractor inside pipeline task Wil O'Mullane ?
  •  Present PST to agree on User Batch uses data holdings and pipeline task ... Wil O'Mullane 

...