Versions Compared


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

Table of Contents

July 2016

July 8th, 2016

  • Pegasus development
    • waiting for LIGO to check the support for changes for OSG, where pegasuslite URLs are converted to file URL if the staging site and compute site are same
    • pegasuslite signal handling
      • mats updated it. LIGO reported cases, where jobs got killed before the outputs were staged back . But the jobs themselves were not marked as failures.
    • pegasus-init R example
      • R example will not run on OSG because of module load issues
    • 4.6.2 after changes are verified.
  • Pegasus Virtual Summer School

July 1st, 2016

  • Mats has moved bamboo to a new RHEL7 VM
    • migrated all the tests to it.
    • there were issues with CondorC tests that are resolved now. because of path issues
  • pegasus-init R
    • Rafael will integrate Samrat's R example workflow
    • Samrat is also working on a bioconductor example workflow
  • rajiv made minor dashboard query changes

May 2016

May 13th, 2016

  • Pegasus development
    • kickstart wrappers
      • process explosion.
      • eventually we would want it to be in the workflow.
        • handle these wrappers as credentials in the workflow. 
        • what are class of files that are always required.
      • KICKSTART_WRAPPER in kickstart
        • was done for the PAPI stuff originally.
    • pegasus-init for OSG
      • pegasus-init 
    • R examples?
      • rafael will do it in june.
    • job held scenarios
      • open with htcondor admin .. a job should never goto the held state
      • maybe pegasus should do quick retry for small workflows
        • for large workflows retries should happen at a longer delay
      • for workflows less than 100 nodes held duration should be small, and failures maybe should be triggered earlier
      • not for large workflows
    • revisit whether clustered jobs should be based on size of the cluster or the number of jobs
      • mats no longer likes the idea of having fixed number of transfers
    • deep directory structure for the workflows
      • can splinter move to using them?
        • right now they are condor io
        • on the data side it deep directory structure will only work 
      • Mats tried with condor 8.5.4 on comet.