The Kubernetes project has released version 1.33, code-named “Octarine,” introducing 64 enhancements, including 18 stable features, 20 in beta, and 24 in alpha. Two features have been deprecated or removed.

This release continues Kubernetes’ tradition of delivering regular, high-quality updates, driven by a global community of contributors. Key deprecations and removals are included, and users of older versions are encouraged to review them.

Named after the magical color from Terry Pratchett’s Discworld, v1.33 celebrates the open source spirit and collaboration that power Kubernetes. As the platform enters its second decade, this release honors the maintainers and contributors who make its ongoing evolution possible.

What’s New and Stable in Kubernetes v1.33: A Quick Rundown

Kubernetes v1.33 is here, and with it comes a fresh batch of enhancements that have officially graduated to stable. This release focuses on making workloads smarter, networking faster, and security tighter. Here’s a quick tour of the highlights:

Backoff limits per index for indexed Jobs

You can now set retry limits per index for Indexed Jobs, giving you more control over how individual tasks fail or succeed without impacting the whole Job. Plus, a new Job success policy lets you define what “success” actually means—whether it’s all pods succeeding, just a few, or specific ones.

Stronger Security for ServiceAccounts

Tokens used by ServiceAccounts now include unique IDs and node-specific restrictions. This adds layers of auditing and enforcement, helping ensure tokens can’t be misused across the cluster.

kubectl Gets Subresource Powers

The --subresource flag is now generally available, letting you use commands like kubectl get or kubectl patch on subresources. Handy for more precise resource management straight from the CLI.

Scalable Networking and Faster Proxying

With support for multiple Service CIDRs, clusters can dynamically expand IP ranges. Meanwhile, the nftables backend for kube-proxy is now stable, delivering better performance (especially for large clusters) on Linux systems.

Topology-Aware Routing (PreferClose)

Traffic can now be steered to endpoints in the same zone using the PreferClose option. This reduces latency and cost by keeping traffic local when possible.

Improved CPU and Pod Placement Control

The CPU Manager can now enforce SMT-aligned workloads, ensuring better isolation. Meanwhile, affinity rules and taints are considered in PodTopologySpread, helping pods get scheduled exactly where they should be.

Volume Populator Hits GA

Volume populators let you pre-fill PersistentVolumes from a wide variety of sources—not just snapshots or clones. This is now generally available, making data initialization much more flexible.

PV Reclaim Policy Fix

A long-standing issue is finally resolved: Persistent Volumes now honor their reclaim policy no matter the deletion order of the PVC and PV. That means no more ghost volumes clogging up your storage.

Each of these features represents thoughtful improvements shaped by SIGs across the Kubernetes ecosystem—from SIG Node and SIG Apps to SIG Storage, SIG Network, and beyond.