With fluentbit you might think you could work around this by configuring it all to use a pvc buffer and checking modify stamps on it.  This assumes that you'll always have legit events for all your replicas over that time. However, the default fluentbit container images contain no shells, so the livenessProbe command will always fail for that reason, so you would need to run the "debug" fluentbit images in production if you want valid health checks :cwy: :welp:.