26.7.0

We are proud to announce the release of:

​✨​ ​✨​ ​✨​ ​✨​ ​✨​ ​✨​ ​✨​ ​✨​ ​✨​

Ryax 26.7.0

​✨​ ​✨​ ​✨​ ​✨​ ​✨​ ​✨​ ​✨​ ​✨​ ​✨​

New features

One Worker chart per site type

The generic ryax-worker chart is replaced by one dedicated chart per site type:

• ryax-worker-k8s for Kubernetes sites
• ryax-worker-slurm-ssh for SLURM HPC sites accessed over SSH

The site type is now determined by the chart you install, which simplifies the configuration and reduces each Worker's footprint to what its site actually needs.

IntelliScale on the main site

IntelliScale now runs on the main site as part of the ryax-engine chart, instead of being deployed within each Worker. It consumes execution metrics and publishes resource recommendations for all sites over the message broker. It is enabled by default and can be disabled with intelliscale.enabled=false.

Energy-aware scheduling

The scheduler now runs energy models directly so that placement decisions can take the energy consumption of an execution into account, alongside performance and cost.

Bug fixes and Improvements

• GPU sharing with NVIDIA MIG works again
• Sites and Node Pools can now be activated and deactivated from the UI
• Passwords handled by workflows are now encrypted at rest, with the encryption key automatically provisioned by the Helm chart
• Log collection now uses Grafana Alloy, replacing the deprecated Promtail
• Deployment management moved from the Runner to the Workers to improve the multi-site architecture
• New MOS v1 placement policy and externalized scheduler to support custom scheduling policies
• Memory warm start (bump-up) logic moved from IntelliScale to the Runner for more robust allocations
• Fix Infrastructure view queries
• Studio API migrated to FastAPI
• Numerous dependency upgrades and security fixes across all services

Upgrade to this version

To upgrade your main cluster, find the values file from your previous install or restore it using:

helm get values -n ryaxns ryax --output yaml > values.yaml

Because we do a major update of the Minio Chart, a breaking change in the deployment selector requires you to run:

kubectl delete deployment ryax-minio -n ryaxns

IntelliScale is now part of the main chart and enabled by default. Also, if your values customize Promtail (promtail section), port that configuration to Grafana Alloy (alloy section).

Then, run the upgrade with:

(note: --take-ownership is required because loki migrate from the worker chart to the ryax one`)

helm upgrade ryax oci://registry.ryax.org/release-charts/ryax-engine:26.7.0 \
  -n ryaxns \
  --take-ownership \
  -f values.yaml

To upgrade a Kubernetes Worker, note that the ryax-worker chart is replaced by ryax-worker-k8s. First, restore your Worker values with:

helm get values -n ryaxns ryax-worker --output yaml > worker.yaml

Then, remove the intelliscale, loki, and promtail sections from this file if present: IntelliScale now runs on the main site, and logs are collected by the main site's Grafana Alloy. Finally, upgrade the release to the new chart:

helm uninstall -n ryaxns ryax-worker
helm install ryax-worker-k8s oci://registry.ryax.org/release-charts/ryax-worker-k8s:26.7.0 \
  -n ryaxns \
  -f worker.yaml

If your Worker was attached to a SLURM cluster (using the hpcOffloading option of the old ryax-worker chart), it must now be installed from the dedicated ryax-worker-slurm-ssh chart. Follow the SLURM_SSH Worker documentation to write the new values file, then replace your old Worker with:

helm uninstall -n ryaxns ryax-worker-hpc
helm install ryax-worker-slurm oci://registry.ryax.org/release-charts/ryax-worker-slurm-ssh:26.7.0 \
  -n ryaxns \
  -f worker.yaml \
  --set-file hpcPrivateKeyFile=./my-ssh-private-key

Concerning the GPU sharing with NVIDIA MIG the GPU nodes must now be pre-partitioned into MIG instances by the cluster administrator, as explained in the new GPU node pools with MIG section of the installation documentation.