Evicted! All the Ways Kubernetes Kills Your Pods and How To Avoid Them

Explore the critical aspects of pod lifecycle management in Kubernetes through this 28-minute conference talk that examines the various mechanisms by which Kubernetes can terminate your pods and strategies to prevent unexpected evictions. Delve into the lesser-known abrupt pod eviction modes triggered by core Kubernetes components including kubelet, scheduler, and controller-manager, gaining deep insights into the internal workings of these termination processes. Learn about the specific guarantees you can expect from each eviction mechanism and discover how these terminations can cause applications to fall below their Pod Disruption Budgets, potentially leading to service unavailability. Understand the risks associated with running stateful systems on Kubernetes and debunk common misconceptions such as the belief that kubelet restarts are always safe. Gain practical knowledge about maintaining predictable pod lifecycles in large-scale production environments and receive actionable guidance through a comprehensive cheatsheet covering all eviction modes in Kubernetes, enabling you to better reason about and prevent unexpected pod terminations in your cloud native deployments.