eviction
Removing a running pod from its node, either through the API during maintenance or by the kubelet when the node runs short of resources.
In more detail
API-initiated eviction respects PodDisruptionBudgets and is what kubectl drain uses. Node-pressure eviction is a kubelet decision to protect the node, and chooses pods partly by how far they exceed their requests. Evicted pods of a controller are recreated elsewhere; bare pods are not.
Related terms
- disposability — The property of a process that can be started or stopped at a moment’s notice without harm.
- scheduler — A component that decides which piece of pending work runs next and when, based on priorities.
- PodDisruptionBudget — An object that limits how many pods of an application may be voluntarily taken down at once, such as during a node drain.
- node pressure — A node condition reporting that it is running short of memory, disk or process IDs, which leads to taints and evictions.
- drain — Cordoning a node and then evicting its pods through the eviction API so it can be safely maintained or removed.
- kubelet — The agent on every Kubernetes node that starts and watches the node's containers and reports back to the API server.
Taught in
- Disposability, Dev Prod Parity And Logs — DevOps & Site Reliability Engineering
- Node Affinity, Pod Affinity And Topology Spread — Kubernetes: Production-Grade Container Orchestration
- Taints, Tolerations And Disruption Budgets — Kubernetes: Production-Grade Container Orchestration
- Troubleshooting: Pending Pod Triage — Kubernetes: Production-Grade Container Orchestration
- Troubleshooting: OOMKilled, Throttling And Evictions — Kubernetes: Production-Grade Container Orchestration
- Drain, Upgrade, Version Skew And Certificate Maintenance — Kubernetes: Production-Grade Container Orchestration