Kubernetes v1.35 Introduces In-Place Pod Restart for Enhanced Efficiency
The release of Kubernetes version 1.35 introduces a significant new feature called Restart All Containers, currently in alpha stage. This functionality enables an efficient, in-place restart of all containers within a Pod, addressing limitations of previous restart policies that only handled individual container crashes or required costly full Pod recreation. By preserving critical resources such as the Pod’s UID, IP address, network namespace, and attached volumes, this feature significantly reduces overhead during failure recovery. It is particularly beneficial for large-scale AI and machine learning workloads, where synchronizing thousands of Pods after a failure previously caused massive bottlenecks and estimated resource waste of up to $100,000 per month. The feature is activated via the RestartAllContainersOnContainerExits feature gate and extends the Container Restart Rules, which have graduated to beta. This update allows developers to offload complex failure-handling mechanisms to Kubernetes configuration, improving system robustness and enabling optimizations like node-level caching tied to specific Pod identities.
Wire timeline
Kubernetes v1.35 Introduces In-Place Pod Restart for Enhanced Efficiency
The release of Kubernetes version 1.35 introduces a significant new feature called Restart All Containers, currently in alpha stage. This functionality enables an efficient, in-place restart of all containers within a Pod, addressing limitations of previous restart policies that only handled individual container crashes or required costly full Pod recreation. By preserving critical resources such as the Pod’s UID, IP address, network namespace, and attached volumes, this feature significantly reduces overhead during failure recovery. It is particularly beneficial for large-scale AI and machine learning workloads, where synchronizing thousands of Pods after a failure previously caused massive bottlenecks and estimated resource waste of up to $100,000 per month. The feature is activated via the RestartAllContainersOnContainerExits feature gate and extends the Container Restart Rules, which have graduated to beta. This update allows developers to offload complex failure-handling mechanisms to Kubernetes configuration, improving system robustness and enabling optimizations like node-level caching tied to specific Pod identities.
Kubernetes Blog