This file is used to list changes made in each version of the aws-parallelcluster-node package.
CHANGES
nodewatcher
: sge - improve logic to detect if a compute node has running jobssqswatcher
: remove invalid messages from SQS queue in order to process remaining messagessqswatcher
: add number of slots to the log of torque schedulersqswatcher
: add retries in case aws request limits are reached
BUG FIXES
sqswatcher
: keep processing compute node termination until all scheduled jobs are terminated/cancelled. This allows to automatically remove dead nodes from the scheduler once all jobs are terminated.jobwatcher
: better handling of error conditions and usage of fallback valuesnodewatcher
: enable daemon when cluster status isUPDATE_ROLLBACK_COMPLETE
TOOLING
- Add a script to simplify node package upload when using
custom_node_package
option
- China Regions, cn-north-1 and cn-northwest-1 support
Bug Fixes:
- Don't schedule jobs on compute nodes that are terminating
- Align version to main ParallelCluster package
- Rename package to AWS ParallelCluster
Bug fixes/minor improvements:
- Changed scaling functionality to scale up and scale down faster.
Bug fixes/minor improvements:
- Upgraded Boto2 to Boto3 package.
Bug fixes/minor improvements:
- Fixed Slurm behavior to add CPU slots so multiple jobs can be scheduled on a single node, this also sets CPU as a consumable resource
Bug fixes/minor improvements:
- Fixed Torque behavior when scaling up from an empty cluster
- Avoid Torque server restart when adding and removing compute nodes