Cloud-native community has long desired the ability to resize pods in-place because pod restarts are expensive for long-running applications and disruptive to services. To resize pods, we commonly rely on Vertical Pod Autoscaler (VPA) to observe usage, recommend, and reactively enact resource updates.
An alternative approach is to resize pods based on deterministic events. For example, you can capture events with eBPF to detect when a CPU-intensive command is going to be executed, and proactively resize the pod CPU accordingly. In this talk, Pablo will show an interesting use case where remote development environments run inside pods. These pods need minimal resources when a developer is writing code, but need significantly higher CPU & memory when a developer issues a “build” command or runs a battery of tests. In-place resize is mandatory in this scenario, otherwise, the development experience would be broken on every pod restart.
Vinay will then talk about the current in-place pod resize feature design, which is soon landing as alpha in Kubernetes. He will go over the CRI changes, discuss the design rationale & trade-offs. He will then lay out the next steps and discuss what the community can do to help to drive this feature to a rock-solid GA over the next year.