Kubernetes 1.31 introduces the Pod Failure Policy feature to General Availability (GA), significantly enhancing Job management. This update allows users to better differentiate between retriable and non-retriable Pod failures, ultimately optimizing resource usage.
The benefits of the new Pod Failure Policy include:
- Reduction of unnecessary Pod restarts, which can lead to lower operational costs.
- Ability to define specific rules for handling failures within Job specifications.
- Improved efficiency by differentiating failure types based on container exit codes.
Previously, Kubernetes Jobs relied on the backoffLimit field, which could lead to increased expenses, especially in environments with numerous long-running Pods. The new policy allows for more nuanced responses to failures:
- Ignore certain failures.
- Fail the entire Job.
- Fail only the index related to the failed Pod.
Kubernetes 1.31 also introduces the DisruptionTarget Pod condition, which identifies Pods affected by Kubernetes-initiated disruptions, ensuring that not all disruptions count against the Job’s failure limits. This addition enhances overall operational efficiency.
To implement the Pod Failure Policy:
- Set the Job’s Pod template with a restartPolicy of “Never” to avoid race conditions during failure counting.
With these advancements, users can expect significant improvements in Job handling, granting better control over workload execution and cost management. For more information, visit the official Kubernetes documentation.
