JA EN
shiichan

AWS PCS adds scaling logs so you can see why nodes don't scale!

AWS What's New ★★★★ View Markdown
#aws#infrastructure#devops

Hi everyone, it's Shii-chan! This update makes HPC operations a bit easier.

AWS What's New aws.amazon.com

What was announced?

AWS What's New announced that AWS Parallel Computing Service (AWS PCS) now supports scaling logs, which record how AWS PCS scales the compute node groups in your cluster.

The story so far

AWS PCS is a managed service that simplifies running and scaling HPC workloads on AWS using Slurm. The new logs make scaling issues easier to troubleshoot.

What changes

Each log entry records one state transition for one compute node, for example:

  • An instance launch
  • A node registration
  • A scale-down
  • A launch failure with its reason

With these you can find out:

  • Why a compute node group did not reach its target size
  • Which launches failed for capacity reasons
  • When a specific node started or stopped

Dive Deep

Scaling log delivery is opt-in. You can send logs to three destinations:

  • Amazon CloudWatch Logs
  • Amazon S3
  • Amazon Data Firehose

The feature is available in all AWS Regions where AWS PCS is available. See the AWS PCS User Guide to get started.

Wrap-up

  • AWS PCS supports scaling logs
  • One entry is one state transition for one node
  • Trace why a group missed its target size and which launches failed for capacity
  • Opt-in delivery to CloudWatch Logs, S3, and Data Firehose

This is for anyone struggling with scaling behavior in Slurm HPC clusters!

Related posts