Add more information to readiness probes
embano1 opened this issue · comments
Michael Gasch commented
As discussed here sometimes it´s not only necessary to gracefully shutdown (your might also want to mention graceful shutdowns in the doc) but also to call a pre-stop lifecycle hook.
Only this will ensure that the pod is being taken out of the service endpoint list before getting SIGTERM/ SIGKILL (depending on the termination policy).
How to reproduce: Try to hammer the service during scale-in or rolling update with "go-wrk" or "hey" tools to see connections being dropped.
Alexander Klizhentas commented
great idea, will add some examples of this, thanks for the feedback!
Michael Gasch commented
cool, thx!
see here for some code i used to demo:
kubernetes/ingress-nginx#322 (comment)