Pegasus 3.1.0 Released

with No Comments

We are pleased to annouce Pegasus 3.1. This is a major release of Pegasus with

  • Support for workflow and job level notifications
  • A redesigned backend database to store information about workflows and jobs.
  • Updated and redesigned auxillary tools in the release namely
    • pegasus-status
    • pegasus-statistics
    • pegasus-plots

The user guide has been reorganized and now has a new user walkthrough

and reference guide for all command line tools.

For more information:

Note for Existing Users:
There has been a change on how locations of properties files are passed to the planner and tools. From Pegasus 3.1 release onwards, support has been dropped for the following properties that were used to signify the location of the properties file
Instead, users should use the –conf option for the tools. The –conf option should appear after any -Dproperty=value specified to the tools.
More details can be found here
  1. Support for Notifications
    • This release of Pegasus has support for workflow level and job notifications. Currently, the user can annotate the DAX to specify  what notifications they want associated with the workflow or/and  individual jobs. Associating a notification with job entails  specifying the condition when notification should be sent, the   executable that needs to be invoked and the arguments with which it   needs to be invoked. All notifications are invoked on the submit  host by the monitoring daemon pegasus-monitord. The release comes  bundled with default notification scripts that users can use for  notifications.
      The DAX API’s have also been updated to allow for associating notifications with the jobs and the workflow.
      More details about how notifications can be found here
  2. Workflow and Jobs Database
    • The backend database schema to which pegasus-monitord populates runtime information about jobs in the workflow has been redesigned. Now in addition to jobs in the executable workflow, information about tasks in the DAX is also tracked and can be connected to the corresponding kickstart records.
      Also, pegasus-monitord no longer dies on database related  errors. The statistics and plotting tools have in built checks that will notify a user if a DB was not populated fully for a workflow run.
      pegasus-monitord now logs timestamps in monitord.log and monitord.done files to reflect the time monitord finishes processing specific  sub-workflows.
      Information about the updated database schema can be found here.
  3. Updated pegasus-statistics and plots
    • pegasus-statistics and pegasus-plots have been updated to retrive all information from the runtime stampede database. pegasus plots now generates plots using protoviz and generates charts showing invocation breakdown, workflow gantt chart, host over time chart that shows how jobs ran on various hosts and a Time chart shows job instance/invocation count and runtime of the workflow run over time
      More information about updated statistics and tools can be found here
  4. Updated pegasus-status tool
    • The pegasus-status tool has been reimplemented for this  release. The new tool shows the current state of a Condor Q and  formats it better. For hierarichal workflows, the tool now displays jobs correctly grouped by sub workflows.
      More information can be found here
  5. Improved support for S3
    • With 3.1 release, there is a pegasus-s3 client that uses the amazon api to create buckets, put and retrieve files from buckets. This  client has further been incorporated into pegasus-transfer . The  pegasus-s3 looks up a configuration file to look up connection  parameters and authentication tokens. The S3 config file is  automatically transferred to the cloud with jobs when a   workflow is configured to run in the S3 mode.
      In the S3 mode, jobs will run in the cloud without requiring a shared filesystem.
      More information about S3 mode in Pegasus can be found here
  6. Tools now have a –conf option 
    •  Most of the command line tools now have a –conf option that can be   used to pass a properties file to the tools. Properties can no longer be passed to a tool using
  7. Improved rescue dag semantics for hierarchal workflows
    • In earlier releases, a rescue dag submission of a hierarchal workflow lead to re-planning of the sub workflows even though rescue dags were submitted for the sub workflows. This could create problems as the re-planning resulted in the braindump files being  over-written and monitord attempting to load information into the  stampede database with a new workflow uuid.
      In this release, this issue has been addressed. By default for sub workflows  rescue dags are always submitted unless a –force-replan option is provided to pegasus-plan. In case of replanning, now a new submit directory is created for the sub workflow. The submit directories for sub workflows are now symlinks that point to the current submit directory for a sub workflow. This ensures that there are no race conditions between monitord and the workflow while populating to the database.
  8.  Default categories for certain types of jobs.
    •  subdax, subdag , cleanup and registration jobs now have default DAGMan categories associated with them.
      dax subwf
      dag subwf
      cleanup cleanup
      registration registration
      This allows for a user to control maxjobs for these categories   easily in properties by specifying
         dagman.[CATEGORY NAME].maxjobs property
    • If a file based replica catalog is used, then maxjobs for registration jobs is set to 1. This is to ensure, that multiple    registration jobs are not run at the same time.
  9.  Automatic loading of DAXParser based on schema number
    • Earlier the users needed to specify the pegasus.schemea.dax property to point to the corresponding DAX schema definition file to get Pegasus to load DAX’es with version < 3.2 and plan it.
      Pegasus now inspects the version number in the adag element to  determine what parser will be loaded.
  10. pegasus-tc-client
    • pegasus-tc-client now displays output in the new multi line Text format, rather than the old File format.
      The support for the File format for the Transformation Catalog will be removed in the upcoming releases.
  11. Removed the requirements for specifying grid gateways in Site Catalog
    • Grid Gateways are associated with a site in a Site Catalog to  designate the jobmanagers associated with a grid site. However, in  the case where jobs were submitted in a pure condor enviornment or  on a local sites ( where jobs are not submitted via jobmanagers),  Pegasus still required users to associate dummy grid gateways with the site.  This is no longer required . The Grid Gateways need to be specified only for grid sites now.
  12. Workflow metrics file in the submit directory
    • A workflow metrics file is created by the planner in the submit  directory that gives a breakdown of various jobs in the executable workflow by type.
  13. pegasus-plan is always niced
    • Starting with this release, pegasus-plan always nice’s the corresponding java invocation that launches the planner. This is helpful in keeping the load on the submit host in check.
  14. Dropped support for VORS and MyOSG backends
    • pegasus-sc-client now relies only on one backend ( OSGMM ) to generate a site catalog for OSG. VORS and MyOSG are no longer suppored by OSG.