Hi there. I've run into an debugging problem with the flink operator When my task managers encounter an error during the sink, they get killed immediately and replaced with new instances. I'm trying to get to the logs of the previously run pods, but it looks like k8s is also deleting the logs for those failed pods. Is there a way to 1. Keep the logs from being deleted? 2. Have an additional delay in the pods so that I have time to view logs/poke around instead of being reaped instantly?

Some sort of log collection seems to be the most robust solution. There's many options available. We collect logs using the Gelf log4j appender which sends logs of to Graylog.

Thanks. That's what I think I've ended up settling on as there didn't seem to be a way to control this in the flink job level or operator level.