recall

← recall

rolling deployment pattern

Roll out new code by replacing a small batch of instances at a time, watching health checks, then moving to the next batch. The standard k8s deploy strategy. Cheaper than blue-green (no doubled capacity) but rollback is slower.

Roll out new code by replacing a small batch of instances at a time, watching health checks, then moving to the next batch. The standard k8s deploy strategy. Cheaper than blue-green (no doubled capacity) but rollback is slower.

symptoms

  • need zero-downtime deploys without doubled capacity
  • gradual rollout for safety

causes

  • blue-green is expensive
  • don't need atomic cutover

fixes

  • k8s rolling update strategy with maxSurge / maxUnavailable
  • health checks gate batch progression
  • PodDisruptionBudgets for safety

you might say

  • rolling deploy
  • rolling restart
  • maxUnavailable

related

aliases: rolling update, rolling restart

topics: deployment, operations