Scalr/agent-helm agent-job-0.6.4 on GitHub

Added task.job.backoffLimit (default 1) to bound retries for task pods that fail before the workload starts, such as launch and infrastructure failures.
Added a kubeVersion: ">=1.35.0-0" constraint so Helm rejects installation and upgrades on clusters older than Kubernetes 1.35.
Added agent.providerCache.dropKernelPages (default true) to control dropping Linux kernel page cache entries for the provider cache directory, exported as SCALR_AGENT_PROVIDER_CACHE_DROP_KERNEL_PAGES.

Task Jobs now use a podFailurePolicy so a pod that already started the task is no longer retried. A container that exits non-zero fails the Job immediately, while a pod disrupted before it starts (eviction, preemption, node shutdown) is retried without consuming the backoff budget. Previously backoffLimit was fixed at 0 and never retried.