v1.134

What's Changed

• Updates to OpenHPC role and source image by @sjpb in #324
• Development quality-of-life improvements by @sjpb in #316
• Add support for freeipa clients by @sjpb in #241

Full Changelog: v1.133...v1.134

Deployment notes

The stackhpc.openhpc role has changed. To update this run:

dev/setup-env.sh

Image Details

New image openhpc-231027-0916-893570de

v1.133

What's Changed

• Update to Open OnDemand v3.0.1 by @thomasbergernz and @sjpb in #314
• Make nvidia-driver install idempotent by @sjpb in #315
• Fix for Slurm CVE 2023 41914 by @sjpb in #320

CI changes

• Use local container image registry for CI to avoid docker.io ratelimits by @sjpb in #318

Deployment notes

The osc.ood role has changed. To update this run:

ansible-galaxy role install --force -r requirements.yml -p ansible/roles

Image Details

New image openhpc-231020-1357-b5d8b056, requiring 10GB root disk.

New Contributors

• @thomasbergernz made their first contribution in #310

Full Changelog: v1.132...v1.133

v1.132

What's Changed

• Fix ssh ControlPath in skeleton by @sjpb in #297
• Fix issues when using GenericCloud image by @sjpb in #313

Full Changelog: v1.131...v1.132

CI changes

• "Fat" image build can now be done either on Arcus (using volume-backed instances -> 10GB virtual disk) or SMS-labs (using non-volume-backed instances - 12GB virtual disk)

Deployment notes

No galaxy-installed roles/collections have changed.

Image Details

Built a new image openhpc-230922-0940-434e190f

Now only requires a 10GB root disk.

v1.131

What's Changed

New features

• Support for CUDA by @sjpb in #253 and #283 - see 253 for full details and configuration

Fixes and Enhancements

• Make etc_hosts role more flexible by @sjpb in #277
• Update prometheus-slurm-exporter version by @m-bull in #280
• Install out of tree openstack builder plugin by @m-bull in #285
• Remove warn parameter for ansible>=2.14 by @mkjpryor in #286
• Fix opensearch grafana plugin at last working version by @sjpb in #292
• Fix query type in the Slurm jobs Grafana dashboard by @mkarpiarz in #293
• Use Python3.9 for jupyter notebook server by @sjpb in #294
• Pin Terraform in CI to MPL licenced version by @sjpb in #302
• Update opensearch to 2.9.0 by @sjpb in #299

CI changes

• Make CI cloud selectable between SMSlabs and Arcus by @sjpb in #288
• Disable EESSI tests in CI and make them debuggable by @sjpb in #295
• Fix SMS ssh by @sjpb in #296
• Use portal-internal network (with normal-mode ports) for Arcus CI by @sjpb in #306

Deployment notes

Galaxy roles/collection versions have been changed so use ansible-galaxy {role,collection} install -f ... after merging to force-update these.

Image Details

• Built a new image openhpc-230811-1548-a49164d1 in #301
• Available from https://object.arcus.openstack.hpc.cam.ac.uk/swift/v1/AUTH_3a06571936a0424bb40bc5c672c4ccb1/openhpc-images/openhpc-230811-1548-a49164d1
• Now only requires a 12GB root disk

New Contributors

• @mkjpryor made their first contribution in #286
• @mkarpiarz made their first contribution in #293

Full Changelog: v1.130...v1.131

v1.130

What's Changed

New functionallity/roles/groups

• Support for EESSI - see roles/eeessi. By @JohnGarbutt in #252

Changes to Packer build functionality

• Allow Packer base images to be specified by either UUID or name by @m-bull in #266
• Support attaching a floating IP to the fatimage builder instance by @m-bull in #267
• Support using volume-backed instances for building and selecting the output image format by @m-bull in #269
• Allow specifying the packer manifest output path by @m-bull in #268
• Allow use of ephemeral SSH keys when building Packer images by @m-bull in #274

Other changes

• Support changing the podman user's uid by @sjpb in #264
• Fix to proxy role: now defaults to including localhost in no_proxy by @sjpb in #270
• Add debug logging options for opensearch & filebeat by @sjpb in #271
• The UCX device to use for hpctests can now be defined per partition by @sjpb in #275
• Always delete resources on deploy failure in CI by @sjpb in #272

Full Changelog: v1.129...v1.130

Deployment notes

• No galaxy reinstalls required since last release.

Image details

• Provides a new image openhpc-230503-0944-bf8c3f63.qcow2 including EESSI support.
• Requires a 40GB root disk.

v1.129

What's Changed

• Return pingpong sbatch output if job fails by @sjpb in #242
• Support configuring nameservers and proxies by @sjpb in #247

Full Changelog: v1.128...v1.129

Deployment notes

• No galaxy reinstalls required

Image details

• No new image. See https://github.com/stackhpc/ansible-slurm-appliance/releases/tag/v1.128.

v1.128

What's Changed

• Environment hooks used when building "fat" image by @sjpb in #255 - allows building site-specific fat images.
• Re-enable NetworkManager control of /etc/resolv.conf after image build by @sjpb in #258 - fixes #257.

Full Changelog: v1.127...v1.128

Deployment Notes

• No galaxy reinstalls required

Image Info

• Provides new image openhpc-230412-1447-e3769af6.qcow2
• Requires a 40GB root disk.

NB: This build uses a RockyLinux 8.6 image plus updates instead of the Rocky-8-GenericCloud-Base-8.7-20221130.0.x86_64.qcow2 image, to avoid an issue with volume mounts during reboot in that image.

v1.127

What's Changed

• Build fat image in appliance by @sjpb in #250:
  • The "fat" image containing binaries for all nodes is now built by this repo, not https://github.com/stackhpc/slurm_image_builder as previously. See section below for latest image details.
  • This removes environment-specific control and login node builds - use the fat image instead (these nodes already required the ansible site.yml playbook running after a reimage).
  • Compute node build is now intended to be performed from the latest fat image, hence removed the yum update *.
  • Various minor fixes, see above PR.

Full Changelog: v1.126...v1.127

Image Info

• Provides image openhpc-230221-1226-f5ba2db7.qcow2.
• Requires a 40GB root disk.

NB: This build uses a RockyLinux 8.6 image plus updates instead of the Rocky-8-GenericCloud-Base-8.7-20221130.0.x86_64.qcow2 image, to avoid an issue with volume mounts during reboot in that image.

v1.126

What's Changed

• Provide container packages by @sjpb in #249: Clusters now have apptainer (for singularity), podman and podman-compose packages by default
• Add script to retrieve CI inventory by @sjpb in #248

NB: This is now based on OpenHPC v2.6.1, removing the workarounds added in v1.123.

Full Changelog: v1.125...v1.126

Deployment Notes

Galaxy roles/collection versions have been changed so use ansible-galaxy {role,collection} install -f ... after merging to force-update these.

v1.125

What's Changed

Replaces OpenDistro with OpenSearch 2.4.0 by @sjpb in #197, including:

• Grafana updated from 8.5.9 to 9.0.3.
• Filebeat-OSS updated from 7.9.3 to 7.12.1.
• Host group and role opendistro replaced with group and role opensearch.
• elasticsearch_* and opendistro_* variables used for opendistro role replaced by opensearch_* variables.
• Changed to host networking for containers.
• Updated CaaS/CI image to openhpc-230110-1629.qcow2.

Full Changelog: v1.124...v1.125

Notes

1. By default opendistro uses self-signed certs with 10 years validity. Certs are automatically updated if necessary when running the appliance.

Deployment

1. Merge this release:

   • Change any environment-specific elasticsearch_* and opendistro_* variables to appropriate opensearch_* equivalents. Note the defaults in the common environment are changed by this release so only non-defaults require manual action.
   • Remove podman_cidr variable as this is unused for container host networking.

2. Update galaxy roles and collections.

3. Run site.yml. OpenDistro data in {{ appliances_state_dir | default('/usr/share') }}/elasticsearch/data will be automatically migrated to OpenSearch . Note the OpenDistro data will not automatically be deleted.