Cluster Toolkit v1.107.0 upgrades Slurm to 26.05.4, adds random_name_suffix variable

Cluster Toolkit v1.107.0 is available. This release adds the random_name_suffix variable to the resource-policy module so that you can optionally omit the random hexadecimal suffix from the names of resource policies.

This release also upgrades Slurm to version 26.05.4 and updates image families to include CUDA 13.2.2 and NVIDIA driver version 595 for H4D, A3 Ultra, A4, and A4X blueprints.

Additionally, this release enforces context timeouts to prevent unresponsive commands in the gcluster CLI, and pins kubectl calls and the dynamic client to the configured kubeconfig context for operations on GKE.

This update also resolves slice shorthands for Cloud TPU, unifies placement policies across GPU and TPU machine types when you submit jobs, and waits for Pods to start before the gcluster job logs command fetches logs.

For more information, see the release announcement on GitHub.