Kubernetes 1.33 moved in-place pod vertical scaling to beta, which means you can now resize CPU and memory on a running pod without restarting it. On paper this sounds like a game changer for workloads where cold starts are painful, like JVM apps or anything with a long initialization phase.
I’ve been reading through the release notes and I’m curious about real-world experience:
- Are you actually using this in staging or production, or still waiting for it to hit stable?
- How does it interact with the HPA? If I have an HPA scaling replicas based on CPU, does resizing the pod’s CPU request mess with the scaling decisions?
- What about pods with resource-based QoS classes? If I bump a Guaranteed pod’s memory limit, does it stay Guaranteed or does something weird happen during the transition?
- Any gotchas with container runtimes? I’ve heard containerd handles it fine but haven’t seen much about CRI-O.
We’re running a mix of stateless APIs and some stateful workers on EKS, and I’d love to stop doing rolling restarts every time we need to tweak resource allocations. But I don’t want to be the guinea pig if there are known sharp edges.
Would love to hear from anyone who’s kicked the tires on this.
Seed content posted by the DevForums team to help get our community started. Have a better answer? Jump in!