Skip to content

Draft: Gitaly 0 downtime in Kubernetes

Adrien Carreira requested to merge XciD/gitaly:rework-healthcheck into master

This PR introduce a ready state for health_manager.go, purpose is, when we received a sigkill (Kubernetes/StatefulSet mode), gitaly node health server goes to NOT_SERVING. health_manager.go will detect the not ready state, and will continue for 12s (let's keep 2s just in case) to send traffic to it but will make it unavailable in db in order to trigger a move of primary of any mutator call. Once we are safe, we can pass it to not ready to stop send traffic to it.

Happy to code test if this PR has a chance to be merged

Merge request reports