how to graceful shutdown the kafka cluster ?

[ahu0605]

I know kafka cluster had graceful shutdown process, leave clutser and sychronize data,election leader? when strimzi kafka will shutdown,how to kill pod? kubenetes will asychronized resource? can you tell me the code how to implements? I have no idea

[scholzj]

Maybe I don't understand what re you trying to do. But as long as the termination grace period is long enough, the pod just shutdowns cleanly when you delete it (which basically sends it SIGTERM signal).

[ahu0605]

Sorry, I don't really get what you are asking for or what point are you trying to make. controlled.shutdown.enable is enabled by default and the termination grace period defines how much time it gets to shutdown cleanly. There is no pre-stop hook needed for this.

when broker node is busy , I think kafka shutdown need long time ,the pod will be killed ,maybe not clean shutdown

[scholzj]

Well, there is a trade-off. You can set the grace period to something very high to protect against situations like this. But then the termination might take a long time even when the pod is dead and you would want to roll it. In any case, the containers need to be able to terminate sooner or later.

[ahu0605]

our test man，discovery when the kafka node be killed un graceful， when the node start agai， the topic leader not rebalance！ but when you exec shudown.sh wait until gracefule shutdown， the kafka will election the leader for balance， maybe kafka need the kubernetes wait long time...

[scholzj]

When you kill Kafka in ungraceful way (e.g. by force deleting the pod), you kill it in ungraceful way. There is not much anyone can do with it. And it can happen anytime when for example your worker node dies. For a clean shutdown, it is about the time you give it. And that is where - as I said - you can use and adjust the termination grace period.

[ahu0605]

When you kill Kafka in ungraceful way (e.g. by force deleting the pod), you kill it in ungraceful way. There is not much anyone can do with it. And it can happen anytime when for example your worker node dies. For a clean shutdown, it is about the time you give it. And that is where - as I said - you can use and adjust the termination grace period.

yes，I know，if busy node check unclean shutdown process, will recovery log segement and index file, it is spent long time,the kubernetes will health check and failed .