I'm trying to get used to helm, and I'm having a problem while using helm upgrade. The question is, is it possible to have 0-downtime?
The things that I noticed is that helm/k8s removes the old pod before the new one is ready/live. Is it possible to say to remove the old one only when the new is up and running? it seems a very logical case, but I can't figure out how to do it.
NB: right now I'm forcing pod recreation with helm upgrade --recreate-pods notes notes/, can maybe be this?
Your applications should have Livness and Readiness probes, thus it will shut down the old ones only when Readiness probe on the new ones passed. Also you can take a look at Deployment Strategy.
Liveness and Readiness probes is a must when architecting applications for Kubernetes, as described in this article.