v1.43.1

• Added per-pool global quotas
• Accounting for GPU tasks assigned to nodes in the current matching cycle
• Force processing when state scanning in k8s

v1.43.0

• Added GPU job support in k8s

v1.42.3

Updates changelog and version for v1.42.3

v1.42.2

• Don't set CPU limit on sidecar if not setting on main container
• Upgrades k8s client library to 7.0.0
• Log pod metadata
• Skips match when there are no considerable jobs
• Do not use :missing state for preempted pod

v1.42.1

• Added check for k8s node preemption using preemption pod label
• Resolved ClassNotFoundException for Mesos task-launching

v1.42.0

• Added Killed by user reason code
• Added fallback to k8s checkpointing disabled when max attempts exceeded
• Added logging of k8s pod events
• Added safe-to-evict annotation to k8s synthetic pods
• Made matches go to launch-tasks in bulk

v1.41.0

bump version

v1.40.4

• Added improved k8s autoscaling metrics
• Allow removing CPU limit in k8s
• Show rate limited users in HTTP log
• Change checkpointing volume from init container to scratch space

v1.40.3

• Added main container environment variables to init container in k8s
• Added writable scratch space separate from the k8s sandbox
• Added experimental API and schema support for checkpointing in k8s
• Added fast fail for k8s pods with un-initialized containers
• Made k8s sidecar readiness probe optional
• Made k8s pod watch initialization process each pod only once
• Removed node anti-affinity for blocklist labels from k8s synthetic pods

v1.40.2

• Added node anti-affinity for block-list labels to synthetic pods