Lifecycle hooks for deployment notifications #331
Description
When Keter falls back to the previous application because a new deployment fails its health check, there’s no way for operators to know this happened. The deployment appears to succeed from the outside because the old application is still serving traffic, but the new version never came up.
It would be useful to support user-configurable hooks that Keter runs at key points in the deployment lifecycle — primarily on deployment success and failure, but potentially also before a deployment begins and when a health check times out.
The hook would just be a shell command specified in the Keter config. Keter would run it as a subprocess with relevant context passed as environment variables (app name, failure reason, the SHA it fell back to, etc.). This keeps the implementation simple and lets users wire up whatever notification system they already use without Keter needing to know about any of them.