Repository navigation
5.1.3 #2246
mayani
announced in
Announcements
5.1.3
#2246
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
We are happy to announce the release of Pegasus 5.1.3. It is a minor release in the 5.1 branch.
Highlights of the Release
Support for NERSC Superfacility API
This release of Pegasus has support for NERSC Superfacility API, enabling scientists to
submit and manage workflows on the Perlmutter supercomputer directly from their submit host
— no SSH tunnels, no batch system logins required.
Futher details about generating SFAPI tokens and configuring Pegasus can be found in the
documentation
The training notebooks for ACCESS Pegasus
also have been updated to be able to submit to NERSC from the
ACCESS Pegasus endpoint
Hosted Site Catalogs
With 5.1.3 release, users now have an option to get the planner to download the site catalog from a
GitHub repository
that hosts catalog entries for the various compute infrastructures.
To specify the file to download from the GitHub repository, you have to set the following property,
and have Pegasus plan for site named
compute.pegasus.catalog.site.repo.file=<basename of the file to download>
Full details on this can be found here.
Job Tagging
Job tagging
is a new feature that allows users to associate
tagswith their jobs in theinout abstract worklow. The planner uses to look up and overlay additional profiles from the
Site Catalog. This allows you to associate different set of profiles for jobs running at
the same site. For e.g. when running against a HPC cluster, this can be used to set
different profiles depending on whether the job is tagged CPU or GPU.
Profile Expressions
Pegasus supports profile expressions — Python expressions that are evaluated when a
job fails and is about to be retried. These expressions allow you to dynamically
change resource requirements (memory, cores, GPUs, runtime, queue, project, etc.)
based on the actual execution metrics collected from the previous attempt,
enabling smarter retry strategies without manual intervention.
Full details on this can be found here.
New Features and Improvements
Bugs Fixed
All reactions