maxUnavailable

maxUnavailable is a parameter used in Kubernetes rollout strategies to limit disruption during updates. It specifies the maximum number of pods that can be unavailable during a rolling update, expressed as an absolute number or a percentage of the desired pod count.

In a Deployment or DaemonSet, maxUnavailable is configured under the RollingUpdate settings: spec.strategy.type must be RollingUpdate, and spec.strategy.rollingUpdate.maxUnavailable can be an integer (for example 2) or a percentage string (for example "20%"). The value controls how many pods can be terminating or not yet ready at any given time during the rollout.

Effectively, during the update, Kubernetes will create new pods and terminate old ones within the limit set by maxUnavailable. If maxUnavailable is 0, no pod may be unavailable during the rollout, so the process waits for each replacement pod to become ready before proceeding. If a related setting like maxSurge is configured (in deployments), the system may also run extra pods above the desired count to facilitate the update.

Scope and considerations: StatefulSets use a different update mechanism (partition-based) and typically do not rely on maxUnavailable. DaemonSets can also leverage rolling updates with maxUnavailable, but the exact behavior may vary across Kubernetes versions and workload types. When choosing a value, operators balance update speed against application readiness, traffic patterns, and the cost of temporarily serving fewer pods.

Examples: With 10 replicas and maxUnavailable: 2, at most two pods can be down at once. With 5 replicas and maxUnavailable: 0, updates will pause until each replacement pod is ready before proceeding.