Kubernetes Contributor Summit North America 2024

PSI (pressure stall information) are metrics the kernel optionally emits that gives finer grained timeseries information on resources uses on the node. Other projects like Facebook's oomd have used these metrics to preempt workloads from being OOM killed and instead give them a more graceful lifecycle. runc 1.2.0 will eventually expose these metrics, and KEP 4205 will add them to kubernetes, but this is just the first step. Join Peter and Dixita in a roundtable discussion on various ways PSI metrics can be utilized in kubernetes. Improved OOM killing? Better informed eviction? Variable swap allocations? The sky is the limit!