AWS PCS Can Now Upgrade Slurm Major Versions Without Stopping!
Hey there, it's me, Shii-chan! Today I found a neat little update that HPC folks will love, so let me share it.
AWS What's New
aws.amazon.com
What was announced?
Today's news comes from AWS What's New. AWS Parallel Computing Service (PCS) now supports in-place Slurm major version upgrades. You can jump up to three major versions ahead at once, with no disruption to running jobs. PCS is a managed service that lets you build elastic HPC environments using Slurm.
The story so far
Upgrading the major version of a job scheduler like Slurm used to be a nerve-wracking task, since it also touches the controller and the accounting layer, and often meant rebuilding the cluster or stopping jobs. Nobody wants to interrupt a long-running computation, right?
What changes
From now on you just update the cluster configuration through the AWS Management Console, the CLI, or the UpdateCluster API. PCS takes care of upgrading the controller, the accounting database, and the REST API for you.
The best part is that running jobs keep going. Queued jobs resume after the running jobs finish, and your accounting data is preserved. Being able to keep up with versions without stopping your compute is a big deal.
Dive Deep
Compute node updates can happen later, whenever it's convenient for you, after the main upgrade completes. It's available in all AWS Regions where PCS is available. The step-by-step details are in the PCS User Guide, so give it a read before you upgrade.
Wrap-up
- PCS can now upgrade Slurm major versions in place, up to three versions ahead
- Just update the config via Console / CLI /
UpdateClusterAPI; PCS handles the controller, accounting DB, and REST API - Running jobs keep going, queued jobs resume after they finish, and accounting data is preserved
- Compute nodes can be updated later at your convenience, available in every Region where PCS runs
This is an update that quietly pays off for anyone running HPC clusters with long-running jobs!
