Kubernetes

Palark Blog

<p>Kubernetes v1.37 is scheduled for release on August 26, 2026. Here’s a rundown of the notable changes, from deprecations to new alpha features.</p>
<h2>Deprecations and removals</h2>
<p><strong><code>kubectl run --filename</code></strong> is being deprecated. If you use it, start planning the migration now.</p>
<p><strong>kube-proxy’s ipvs mode</strong> is deprecated, with full removal expected around v1.43. The nftables backend is taking over - more on that below.</p>
<p><strong>Static pods referencing Secrets or ConfigMaps</strong> are also on the deprecation list.</p>
<p><strong>cgroup v1</strong> support continues its phase-out. In v1.37, kubelet will fail to initialize on cgroup v1 nodes by default. If you’re still running cgroup v1, this is the release that forces the issue.</p>
<h2>Breaking changes</h2>
<p><strong>SELinuxMount graduates to GA.</strong> This one deserves attention: pods that share volumes with different SELinux labels may start failing. If you’re running workloads with SELinux and shared volumes, test this before upgrading.</p>
<h2>Stable graduations</h2>
<p><strong><a href="http://metrics.k8s.io" target="_blank" rel="noopener nofollow">metrics.k8s.io</a> API</strong> finally reaches stable after nine years in beta. Nine years. It’s a long time coming, but it’s there.</p>
<h2>Beta promotions</h2>
<p><strong>Kubelet Rootless Mode (UserNS)</strong> moves to beta, making it easier to run containers without root privileges at the node level.</p>
<h2>New alpha features (22 total)</h2>
<h3>Pod-level checkpoint and restore</h3>
<p>Using CRIU, you can now snapshot entire pods - not just individual containers. This opens up live migration and faster restart scenarios that weren’t practical before.</p>
<h3>DRA enhancements</h3>
<p>Several Dynamic Resource Allocation improvements land in alpha:</p>
<ul>
<li><strong>Optional node preparation</strong> - devices don’t always need node-level setup before scheduling</li>
<li><strong>Device compatibility groups</strong> - express which devices can work together</li>
<li><strong>Standardized <code>numaNode</code> attribute</strong> - consistent NUMA topology reporting across drivers</li>
<li><strong>Derived attributes</strong> - more flexible hardware binding without hardcoding device properties</li>
</ul>
<h3>CompositePodGroup API</h3>
<p>A new API for hierarchical, multi-level gang scheduling. Useful when you need groups of pods to schedule together across multiple dimensions - think distributed training jobs with complex placement requirements.</p>
<h3>nftables as default kube-proxy backend</h3>
<p>With ipvs deprecated, nftables is stepping up as the default. This has been in the works for a while; v1.37 makes it the default in alpha.</p>
<h3>Scheduler preemption for in-place pod resize</h3>
<p>Pods requesting in-place resource changes can now trigger preemption if needed. Previously, resize requests could stall indefinitely if resources weren’t available.</p>
<h3>TLS support for gRPC probes</h3>
<p>gRPC liveness and readiness probes now support TLS. If your services require encrypted connections even internally, this removes a workaround.</p>
<h3>HTTP/2 cleartext (h2c) container probes</h3>
<p>HTTP probes can now use HTTP/2 without TLS. Useful for services that speak HTTP/2 natively.</p>
<h3>Bind mount options for volumes</h3>
<p>You can now set <code>noexec</code>, <code>nosuid</code>, and <code>nodev</code> on volume mounts. These are standard Unix mount security flags that were previously unavailable at the Kubernetes level.</p>
<h3>Topology support for VolumeSnapshots</h3>
<p>VolumeSnapshots can now express topology constraints, which matters for multi-zone clusters where snapshot availability is zone-specific.</p>
<h3>Volume Health Monitor (revamped)</h3>
<p>Four new CSI RPCs provide machine-readable storage health reporting. Previously, storage health information was mostly opaque to the control plane; this gives operators and controllers something concrete to act on.</p>
<h3>Recreate update strategy for StatefulSets</h3>
<p>StatefulSets gain a <code>Recreate</code> update strategy, similar to what Deployments have had for years. Useful for stateful workloads that can’t tolerate running two versions simultaneously.</p>
<h3>Internal API type elimination</h3>
<p>An internal refactor that removes redundant API type conversions. The practical effect is a meaningful performance improvement in the API server, though you won’t see a new feature flag for it.</p>
<h2>Summary</h2>
<p>v1.37 is a fairly dense release. The cgroup v1 change and SELinuxMount GA are the two most likely to cause upgrade friction. On the new features side, pod-level checkpoint/restore and the DRA improvements are worth watching if you’re running GPU or specialized hardware workloads.</p>


Collections

Kubernetes v1.37, scheduled for August 26, 2026, brings significant changes including the deprecation of kube-proxy's ipvs mode (replaced by nftables), cgroup v1 support breaking by default, and SELinuxMount graduating to GA (a potential breaking change for pods sharing volumes with different SELinux labels). The metrics.k8s.io API finally reaches stable after nine years in beta. Notable new alpha features include pod-level checkpoint/restore using CRIU for live migration, multiple Dynamic Resource Allocation (DRA) enhancements for GPU/specialized hardware, a CompositePodGroup API for hierarchical gang scheduling, TLS support for gRPC probes, bind mount security options (noexec, nosuid, nodev), and a Recreate update strategy for StatefulSets. The release also includes an internal API server refactor for performance gains.

What's coming in Kubernetes v1.37