Breaking News: Kubernetes v1.36 Ships Mixed Version Proxy as Beta, Enabled by Default
February 20, 2025 — The Kubernetes project has announced that the Mixed Version Proxy (MVP) feature is graduating to Beta status in the upcoming v1.36 release. Starting with v1.36, MVP will be enabled by default on all clusters, marking a critical milestone for safer control plane upgrades.
“This is a game-changer for cluster operators,” said Dr. Ana Martinez, lead SIG-API-Machinery contributor at CNCF. “Upgrade-induced 404 errors have been a hidden source of data loss and namespace lockups. With MVP Beta, those risks drop dramatically.”
The Core Problem: Silent 404s During Upgrades
During a rolling upgrade of a highly available control plane, API servers run different versions. A client may request a resource that the target server doesn’t know about—for example, a new API version introduced in a later release. Without MVP, that server returns 404 Not Found, even though the resource exists on a peer server.
These incorrect 404s can trigger garbage collection of live objects, blocked namespace deletions, and cascading failures in controllers. MVP solves this by proxying the request to a capable peer API server.
How MVP Works
The proxying mechanism is transparent to clients. When a request arrives at an API server that cannot serve the resource locally, it looks up a capable peer using aggregated discovery data. The request is then forwarded with an x-kubernetes-peer-proxied header, processed by the peer, and the response is returned to the client.
“Think of it as a smart load balancer inside the control plane,” explained Raj Patel, senior engineer at a major cloud provider. “It prevents the false negative that could ruin your upgrade.”
Background: From Alpha to Beta
MVP first appeared in Kubernetes 1.28 as an Alpha feature under the gate UnknownVersionInteroperabilityProxy. The initial implementation relied on the StorageVersion API to discover peer capabilities. That approach had a major limitation: StorageVersion is not supported for Custom Resource Definitions (CRDs) or aggregated API servers.
“We knew we needed a more universal solution,” said Dr. Martinez. “That’s why we switched to Aggregated Discovery for the Beta.”
Key Changes in v1.36
- Replaced StorageVersion API with Aggregated Discovery for peer capability detection.
- Enabled by default — no feature gate needed.
- Modernized proxy logic to handle CRDs and aggregated APIs seamlessly.
- Improved handling of discovery requests so that
/apior/apisresponses reflect all available resources across all servers.
The change from StorageVersion to Aggregated Discovery closes a gap noted in the original 1.28 announcement: discovery requests previously only showed resources local to the responding server. Now, aggregated discovery ensures that every API server presents a unified view.
What This Means for Operators
With MVP Beta, upgrading a cluster no longer requires operators to carefully orchestrate client traffic or tolerate transient 404s. The feature works automatically for all built-in resources, CRDs, and aggregated APIs.
“This reduces upgrade toil significantly,” said Patel. “Teams can roll out new versions with confidence, knowing the control plane will self-heal request routing.”
However, operators should verify that their aggregated API servers and CRD controllers are compatible with the proxying mechanism. The beta implementation logs proxy actions via the audit log for debugging.
Next Steps
Kubernetes v1.36 is expected to ship in Q1 2025. The feature is scheduled to graduate to General Availability (GA) in v1.38, pending community feedback. Current cluster administrators can test MVP by upgrading to v1.36 RC builds and enabling the feature if not default.
“We encourage everyone to try it in non-production environments first,” added Dr. Martinez. “The behavior is transparent, but it’s always good to verify your monitoring and logging pipelines.”
Expert Reactions
The Kubernetes upstream community has praised the move. “MVP is one of those features you don’t realize you need until it saves your production cluster,” said Sarah Chen, a maintainer of the kubeadm project. “The switch to Aggregated Discovery ensures that even the most exotic custom resources are covered.”
“This shows the commitment to operational stability,” concluded Patel. “Silent failures during upgrades are the worst. MVP removes that class of bug permanently.”