|
| 1 | +--- |
| 2 | +layout: blog |
| 3 | +title: "Kubernetes v1.34: Pod Replacement Policy for Jobs Goes GA" |
| 4 | +date: 2025-0X-XX |
| 5 | +draft: true |
| 6 | +slug: kubernetes-v1-34-pod-replacement-policy-for-jobs-goes-ga |
| 7 | +author: > |
| 8 | + [Dejan Zele Pejchev](https://github.com/dejanzele) (G-Research) |
| 9 | +--- |
| 10 | + |
| 11 | +In Kubernetes v1.34, the _Pod Replacement Policy_ feature reaches general availability (GA). |
| 12 | +This blog post describes the Pod Replacement Policy feature and how to use it in your Jobs. |
| 13 | + |
| 14 | +## About Pod Replacement Policy |
| 15 | + |
| 16 | +By default, the Job controller immediately recreates Pods as soon as they fail or begin terminating (when they have a deletion timestamp). |
| 17 | + |
| 18 | +As a result, while some Pods are terminating, the total number of running Pods for a Job can temporarily exceed the specified parallelism. |
| 19 | +For Indexed Jobs, this can even mean multiple Pods running for the same index at the same time. |
| 20 | + |
| 21 | +This behavior works fine for many workloads, but it can cause problems in certain cases. |
| 22 | + |
| 23 | +For example, popular machine learning frameworks like TensorFlow and |
| 24 | +[JAX](https://jax.readthedocs.io/en/latest/) expect exactly one Pod per worker index. |
| 25 | +If two Pods run at the same time, you might encounter errors such as: |
| 26 | +``` |
| 27 | +/job:worker/task:4: Duplicate task registration with task_name=/job:worker/replica:0/task:4 |
| 28 | +``` |
| 29 | + |
| 30 | +Additionally, starting replacement Pods before the old ones fully terminate can lead to: |
| 31 | +- Scheduling delays by kube-scheduler as the nodes remain occupied. |
| 32 | +- Unnecessary cluster scale-ups to accommodate the replacement Pods. |
| 33 | +- Temporary bypassing of quota checks by workload orchestrators like [Kueue](https://kueue.sigs.k8s.io/). |
| 34 | + |
| 35 | +The _Pod Replacement Policy_ feature gives you control over when Kubernetes replaces terminating Pods, helping you avoid these issues. |
| 36 | + |
| 37 | +## How Pod Replacement Policy works |
| 38 | + |
| 39 | +The feature introduces a new Job-level field, `podReplacementPolicy`, which controls when Kubernetes replaces terminating Pods. |
| 40 | +You can choose one of two policies: |
| 41 | +- TerminatingOrFailed (default): Replaces Pods as soon as they start terminating. |
| 42 | +- Failed: Replaces Pods only after they fully terminate and transition to the `Failed` phase. |
| 43 | + |
| 44 | +Setting the policy to `Failed` ensures that a new Pod is only created after the previous one has completely terminated. |
| 45 | + |
| 46 | +For Jobs with a Pod Failure Policy, the default `podReplacementPolicy` is `Failed`, and no other value is allowed. |
| 47 | +See [Pod Failure Policy](/docs/concepts/workloads/controllers/job/#pod-failure-policy) to learn more about Pod Failure Policies for Jobs. |
| 48 | + |
| 49 | +You can check how many Pods are currently terminating by inspecting the Job’s `.status.terminating` field: |
| 50 | + |
| 51 | +```sh |
| 52 | +kubectl get job myjob -o=jsonpath='{.status.terminating}' |
| 53 | +``` |
| 54 | + |
| 55 | +## Example |
| 56 | + |
| 57 | +Here’s a simple Job spec that ensures Pods are replaced only after they terminate completely: |
| 58 | + |
| 59 | +```yaml |
| 60 | +apiVersion: batch/v1 |
| 61 | +kind: Job |
| 62 | +metadata: |
| 63 | + name: example-job |
| 64 | +spec: |
| 65 | + podReplacementPolicy: Failed |
| 66 | + template: |
| 67 | + spec: |
| 68 | + restartPolicy: Never |
| 69 | + containers: |
| 70 | + - name: worker |
| 71 | + image: your-image |
| 72 | +``` |
| 73 | +
|
| 74 | +With this setting, Kubernetes won’t launch a replacement Pod while the previous Pod is still terminating. |
| 75 | +
|
| 76 | +## How can you learn more? |
| 77 | +
|
| 78 | +- Read the user-facing documentation for [Pod Replacement Policy](/docs/concepts/workloads/controllers/job/#pod-replacement-policy), |
| 79 | + [Backoff Limit per Index](/docs/concepts/workloads/controllers/job/#backoff-limit-per-index), and |
| 80 | + [Pod Failure Policy](/docs/concepts/workloads/controllers/job/#pod-failure-policy). |
| 81 | +- Read the KEPs for [Pod Replacement Policy](https://github.com/kubernetes/enhancements/tree/master/keps/sig-apps/3939-allow-replacement-when-fully-terminated), |
| 82 | + [Backoff Limit per Index](https://github.com/kubernetes/enhancements/tree/master/keps/sig-apps/3850-backoff-limits-per-index-for-indexed-jobs), and |
| 83 | + [Pod Failure Policy](https://github.com/kubernetes/enhancements/tree/master/keps/sig-apps/3329-retriable-and-non-retriable-failures). |
| 84 | +
|
| 85 | +
|
| 86 | +## Acknowledgments |
| 87 | +
|
| 88 | +As with any Kubernetes feature, multiple people contributed to getting this |
| 89 | +done, from testing and filing bugs to reviewing code. |
| 90 | +
|
| 91 | +As this feature moves to stable after 2 years, we would like to thank the following people: |
| 92 | +* [Kevin Hannon](https://github.com/kannon92) - for writing the KEP and the initial implementation. |
| 93 | +* [Michał Woźniak](https://github.com/mimowo) - for guidance, mentorship, and reviews. |
| 94 | +* [Aldo Culquicondor](https://github.com/alculquicondor) - for guidance, mentorship, and reviews. |
| 95 | +* [Maciej Szulik](https://github.com/soltysh) - for guidance, mentorship, and reviews. |
| 96 | +* [Dejan Zele Pejchev](https://github.com/dejanzele) - for taking over the feature and promoting it from Alpha through Beta to GA. |
| 97 | +
|
| 98 | +## Get involved |
| 99 | +
|
| 100 | +This work was sponsored by the Kubernetes |
| 101 | +[batch working group](https://github.com/kubernetes/community/tree/master/wg-batch) |
| 102 | +in close collaboration with the |
| 103 | +[SIG Apps](https://github.com/kubernetes/community/tree/master/sig-apps) community. |
| 104 | +
|
| 105 | +If you are interested in working on new features in the space we recommend |
| 106 | +subscribing to our [Slack](https://kubernetes.slack.com/messages/wg-batch) |
| 107 | +channel and attending the regular community meetings. |
0 commit comments