ctrl_bps v23.0.1 2022-02-02

New Features

  • Check early in submission process that can import WMS service class and run any pre-submission checks provided by the WMS plugin. (DM-32199)
    • Large tasks (> 30k jobs) splitted into chunks
    • Updated iDDS API usage for the most recent version
    • Updated iDDS API initialization to force PanDA proxy using the IAM user name for submitted workflow
    • Added limit on number of characters in the task pseudo inputs (DM-32675)
    • New panda_auth command for handling PanDA authentication token. Includes status, reset, and clean capabilities.
    • Added early check of PanDA authentication token in submission process. (DM-32830)

Other Changes and Additions

    • Changed printing of submit directory early.
    • Changed PanDA plugin to only print the numeric id when outputing the request/run id.
    • Set maximum number of jobs in a PanDA task (maxJobsPerTask) to 70000 in config/bps_idf.yaml. (DM-32830)

ctrl_bps v23.0.0 2021-12-10

New Features

    • Added bps htcondor job setting that should put jobs that get the signal 7 when exceeding memory on hold. Held message will say: “Job raised a signal 7. Usually means job has gone over memory limit.” Until bps has the automatic memory exceeded retries, you can restart these the same way as with jobs that htcondor held for exceeding memory limits (condor_qedit and condor_release).

    • Too many files were being written to single directories in job/<label>. There is now a template for it defined in yaml:

      subDirTemplate: "{label}/{tract}/{patch}/{visit.day_obs}/{exposure.day_obs}/{band}/{subfilter}/{physical_filter}/{visit}/{exposure}"

      To revert back to previous behavior, in your submit yaml set:

      subDirTemplate: "{label}"

    • bps now has defaults so submit yamls should be a lot simpler and require less changes when bps or pipetask changes. For default values see ${CTRL_BPS_DIR}/python/lsst/ctrl/bps/etc/bps_defaults.yaml. See ${CTRL_BPS_DIR}/doc/lsst.ctrl.bps/pipelines_check.yaml for an example of much simpler submit yaml.

      Values in bps_defaults.yaml are overridden by values in submit yaml (be careful of scoping rules e.g., values in a pipetask section override the global setting).

      STRONGLY recommend removing (commenting out) settings in the submit yaml that are set in the default yaml (i.e., the settings that are same across runs across repos, …)

      It would be helpful to know in what cases submit yamls have to override default settings, in particular the command lines.

    • With the above defaults one can more easily append options to the pipetask command lines as variables in submit yaml:

      • extraQgraphOptions: Adds given string to end of command line for creating QuantumGraph (e.g., for specifying a task wit -t)
      • extraInitOptions: Adds given string to end of pipetaskInit command line
      • extraRunQuantumOptions: Adds given string to end of the pipetask command line for running a Quantum (e.g., --no-versions)

      These can also be specified on the command line (see bps submit --help).

      • --extra-qgraph-options TEXT
      • --extra-init-options TEXT
      • --extra-run-quantum-options TEXT

      Settings on command line override values set in submit yaml.

      The default commands no longer include --no-versions or saving a dot version of the QuantumGraph. Use the appropriate new variable or command-line option to add those back.

    • Can specify some pipetask options on command line (see bps submit --help):

      • -b, --butler-config TEXT
      • -i, --input COLLECTION ...
      • -o, --output COLL
      • --output-run COLL
      • -d, --data-query QUERY
      • -p, --pipeline FILE
      • -g, --qgraph TEXT

      Settings on command line override values set in submit yaml.

    • bps now saves yaml in run’s submit directory. One is just a copy of the submit yaml (uses original filename). And one is a dump of the config after combining command-line options, defaults and submit yaml (<run>_config.yaml).

    • If pipetask starts reporting errors about database connections (e.g., remaining connection slots are reserved for non-replication superuser connections) ask on #dm-middleware-support about using execution butler in bps. This greatly reduces the number of connections to the central database per run. It is not yet the default behavior of bps, but one can modify the submit yaml to use it. See ${CTRL_BPS_DIR}/doc/lsst.ctrl.bps/pipelines_check_execution_butler.yaml

    The major differences visible to users are:

    • bps report shows new job called mergeExecutionButler in detailed view. This is what saves the run info into the central butler repository. As with any job, it can succeed or fail. Different from other jobs, it will execute at the end of a run regardless of whether a job failed or not. It will even execute if the run is cancelled unless the cancellation is while the merge is running. Its output will go where other jobs go (at NCSA in jobs/mergeExecutionButler directory).
    • See new files in submit directory:
      • EXEC_REPO-<run>: Execution butler (yaml + initial sqlite file)
      • execution_butler_creation.out: output of command to create execution butler
      • final_job.bash: Script that is executed to do the merging of the run info into the central repo.
      • final_post_mergeExecutionButler.out: An internal file for debugging incorrect reporting of final run status. (DM-28653)
    • Add numberOfRetries option which specifies the maximum number of retries allowed for a job.
    • Add memoryMultiplier option to allow for increasing the memory requirements automatically between retries for jobs which exceeded memory during their execution. At the moment this option is only supported by HTCondor plugin. (DM-29756)
    • bps report
      • Columns now are as wide as the widest value/heading and some other minor formatting changes.
      • Detailed report (--id) now has an Expected column that shows expected counts per PipelineTask label from the QuantumGraph. (DM-29893)

  • Create list of node ids for the pipetask --init-only job. (DM-31541)

  • Add a new configuration option, preemptible, which indicates whether a job can be safely preempted. (DM-31841)

  • Add user-defined dimension clustering algorithm. (DM-31859)

  • Add --log-label option to bps command to allow extra information to be injected into the log record. (DM-31884)

  • Make using an execution butler the default. (DM-31887)

  • Change HTCondor bps plugin to use HTCondor curl plugin for local job transfers. (DM-32074)

Bug Fixes

    • Fix issue with accessing non-existing attributes when creating the final job.
    • Fix issue preventing bps report from getting the run name correctly. (DM-31541)
  • Fix issue with job attributes not being set. (DM-31841)
    • Fix variable substitution in merge job commands.
    • Fix bug where final job doesn’t appear in report.
    • Fix bug in HTCondor plugin for reporting final job status when –id <path>. (DM-31887)
  • Fix single concurrency limit splitting. (DM-31944)
    • Fix AttributeError during submission if explicitly not using execution butler.
    • Fix bps report summary PermissionsError caused by certain runs with previous version in queue. (DM-31970)
  • Fix the bug in the formula governing memory scaling. (DM-32066)
  • Fix single quantum cluster missing node number. (DM-32074)
  • Fix execution butler with HTCondor plugin bug when output collection has period. (DM-32201)
  • Fix issues with bps commands displaying inaccurate timings (DM-32217)
  • Disable HTCondor auto detection of files to copy back from jobs. (DM-32220)
    • Fixed bug when not using lazy commands but using execution butler.
    • Fixed bug in htcondor_service.py that overwrote message in bps report. (DM-32241)
    • Fixed bug when a pipetask process killed by a signal on the edge node did not expose the failing status. (DM-32435)

Performance Enhancement

  • Cache values by labels to reduce number of config lookups to speed up multiple submission stages. (DM-32241)

Other Changes and Additions

  • Complain about missing memory limit only if memory autoscaling is enabled. (DM-31541)
  • Persist bps DAG attributes across manual restarts. (DM-31944)
  • Change outCollection in submit YAML to outputRun. (DM-32027)
  • Change default for bpsUseShared to True. (DM-32201)
  • Switch default logging level from WARN to INFO. (DM-32217)
  • Provide a cleaned up version of default config yaml for PanDA-pluging on IDF (DM-31476)
  • Rolled back changes in BpsConfig that were added for flexibility when looking up config values (e.g., snake case keys will no longer match camel case keys nor will either match lower case keys). This also removed dependence on third-party inflection package. (DM-32594)