A Kubernetes controller for make-before-break rolling restarts of single-replica deployments when their node is disrupted.
Single-replica deployments don’t get rescheduled until a disrupted node is fully drained — causing unnecessary downtime. This controller detects node disruption signals (cordoning or taints) and triggers a rolling restart, spinning up the replacement pod on a healthy node before the old one is terminated. Multi-replica deployments are excluded since Kubernetes handles those natively.
More info and source on GitHub.