Pegasus 4.6.1 Release

with No Comments
We are happy to announce the release of Pegasus 4.6.1.  Pegasus 4.6.1 is a minor release of Pegasus and includes improvements and bug fixes to the 4.6.0 release
New features and Improvements in 4.6.1 are
  • support for MOAB submissions via glite. A new tool called pegasus-configure-glite helps users setup their HTCondor GLite directory for use with Pegasus
  • pegasus-s3 now allows for downloading and uploading folders to and from S3
  • initial support for globus online in pegasus-transfer
  • planner automatically copies the user catalog files into a directory called catalogs in the submit directory.
  • changes to how worker package staging occurs for compute jobs.
  • bug fix for pegasus-remove that prevented jobs in hierarchal workflows to be removed from condor queue

New Features

  1. [PM-1045] – There is a new command line tool pegasus-configure-glite that automatically installs the Pegasus shipped glite local attributes script to the condor glite installation directory
  2. [PM-1044] – Added glite scripts for moab submissions via the Glite interface
  3. [PM-1054] – kickstart has an option to ignore files in lib interpose.
    • This is triggered by setting the environment variables KICKSTART_TRACE_MATCH and KICKSTART_TRACE_IGNORE. The MATCH version only traces files that match the patterns, and the IGNORE version does NOT trace files that match the patterns. Only one of the two can be specified.
  4. [PM-1058] -pegasus can be now installed via homebrew on MACOSX
    • For details refer to documentation at
  5. [PM-1075] – pegasus-s3 to be able to download all files in a folder
    • pegasus-s3 has a –recursive option to allow users to download all files from a folder in S3 or upload all files from a local directory to S3 bucket.
  6. [PM-680] – Add support for GlobusOnline to pegasus-transfer
    • Details on how to configure can be found at
  7. [PM-1043] – Improve CSV file read for Storage Constraints algorithm
  8. [PM-1047] – Pegasus saves all the catalog files in submit dir in a directory named catalogs. This enables for easier debugging later on as everything is saved in the submit directory.


  1. [PM-1043] – Improve CSV file read for Storage Constraints algorithm
  2. [PM-1057] – PegasusLite worker package download improvements
    • Pegasus exposes two additional properties to control behavior of worker package staging for jobs. Users can use these to control whether a PegasusLite job downloads a worker package from the pegasus website or not , in case the shipped worker package does not match the node architecture.
      • pegasus.transfer.worker.package.strict – enforce strict checks against provided worker package. if a job comes with worker package and it does not match fully with worker node architecture , it falls down to pegasus download website. Default value is true.
      • pegasus.transfer.worker.package.autodownload – a boolean property to indicate whether a pegasus lite job is allowed to download from pegasus website. Defaults to true.
  3. [PM-1059] – Implement backup for MySQL databases
  4. [PM-1060] – expose a way to turn off kickstart stat options
  5. [PM-1063] – improve performance for inserts into database replica catalog
  6. [PM-1067] – pegasus-cluster -R should report the finish time and duration, not the start time and duration
  7. [PM-1078] – pegasus-statistics should take comma separated list of values for -s option
  8. [PM-1073] – condor_q changes in 8.5.x will affect pegasus-status
    • pegasus-status was updated to account for changes in the condor_q output in the 8.5 series

Bugs Fixed

  1. [PM-1077] – pegasus-remove on hierarchal workflows results in jobs from the sub workflows still in the condor queue
    • DAGMan no longer condor_rm jobs in a workflow itself. Instead it relies on condor schedd to do it. Pegasus generated sub workflow description files did not trigger this . As a result, pegasus-remove on a top level workflow still resulted in jobs from the sub workflows to be in the condor queue. This is now fixed. Pegasus generated dagman submit files have the right expressions specified.
    [PM-997] – pyOpenSSL v0.13 does not work with new version of openssl (1.0.2d) and El Captain
  2. [PM-1048] – PegasusLite should do a full version check for pre-installed worker packages
    • PegasusLite does a full check ( including the patch version) with the pegasus version installed on the node, when determining whether to use the preinstalled version on the node or not.
  3. [PM-1050] – pegasus-plan should not fail if -D arguments don’t appear first
  4. [PM-1051] – Error missing when nodes, cores, and ppn are all specified
    • In 4.6.0 release, there was a bug where the error message thrown ( when user specified an invalid combination of task requirements) was incorrect. This is fixed, and error messages have been improved to also indicate a reason
  5. [PM-1053] – pegasus-cluster does not know about new Kickstart arguments
  6. [PM-1055] – Interleaved libinterpose records
  7. [PM-1061] – pegasus-analyzer should detect and report on failed job submissions
    • pegasus-monitord did not populate the stampede workflow database with information about job submission failures. As a result, pegasus-analyzer for the cases where a job failed because of job submission errors did not report any helpful information as to why the job failed. This is now fixed.
  8. [PM-1062] – pegasus dashboard shows some workflows twice
    • In the case where HTCondor crashes on a submit node, DAGMan logs may miss a workflow end event. When monitord detects consecutive start events, it creates and inserts a workflow end event. The end event had the same timestamp as the new start event, because of which underlying dashboard query retrieved multiple rows.  This was fixed by setting the timestamp for the artificial end event to be one second less than the second start event.
  9. [PM-1064] – pegasus-transfer prepends to PATH
    • pegasus-transfer used to prepend the system path with other internal determined lookup directories based on environment variables such as GLOBUS_LOCATION. As a result, in some cases, user preferred copy of executables were not picked up. This is now fixed.
  10. [PM-1066] – wget errors because of network issues
    • pegasus-transfer now sets the OSG_SQUID_LOCATION/http_proxy setting only for the first wget attempt
  11. [PM-1068] – monitord fails when trying to open a job error file in a workflow with condor recovery
    • monitord parses the job submit file whenever it notices job submission log by DAGMan. This is done to avoid the case, where because of HTCondor recovery a job may not have a ULOG_ job submission event, because of which the internal state of the job maybe uninitialized.
  12. [PM-1069] – Dashboard invocation page gives an error if the task has no invocation record
    • Dashboard did not display invocation records for Pegasus added auxiliary jobs in the workflow. This was due to a bug in the query that is now fixed.
  13. [PM-1070] – monitord should handle case where jobs have missing JOB_FAILURE/JOB_TERMINATED events
  14. [PM-1072] – Worker package staging issues on OSX
  15. [PM-1081] – pegasus-plan complains if output dir is set but site catalog entry for local site does not storage directory specified
    • pegasus-plan complained if a storage directory was not specified in the site catalog entry for site “local”, even if a user specified a –output-dir option. This is now fixed. The planner will create a default file server based entry for this case.
  16. [PM-1082] – transfer jobs don’t have symlink destination URL even though symlink is enabled
    • In the case, where there are multiple candidate replica locations ( some on preferred site and some on other sites), the destination URL for the transfer jobs did not have a symlink URL. As a result the data was never symlinked even though it was available locally on the preferred site.
  17. [PM-1083] – dashboard user home page requires a trailing /
    • To access a user home page on the dashboard, a trailing / needs to be specified after the username in the URL. Dashboard was updated to handle URL’s without trailing names.
  18. [PM-1084] – credential handling for glite jobs
    • As part of credential handling the environment variable for the staged credential was as environment key instead of the +remote_environment classed key.  As a result transfer jobs running via Glite submission failed as the appropriate environment variable was not set. This is fixed now.
  19. [PM-1085] – -p 0 options for condor_dagman sub dax jobs result in dagman ( 8.2.8) dying
    • Pegasus updated to generate the dagman submit files for sub workflows to be compatible with 8.5.x series.  However, the new arguments added resulted in breaking workflows running with old HTCondor versions. The offending argument is now set only if condor version is more than 8.3.6
  20. [PM-1086] – Never symlink executables
    • Pegasus adds chmod jobs to explicitly set the x bit of the executables staged. If the executable is a symlinked executable, then chmod fails. Symlinking is never triggered for staged executables now.