Hi,
yes, your case is the exception. In usual deployments kafka has to be
there 100% all the time.
So as the name rolling restart suggest, you usually upgrade / do
maitenance on boxes (a few at a time) depending how your
topics are laied our across brokers.
On 09.05.2018 12:13, M. Manna wrote:
Thanks Jan. We have 9 broker-zookeeper setup in production and during
monthly maintenance we need to shut it down gracefully (or reasonably) to
do our work.
Are you saying that it's okay not to shut down the entire cluster?
Also, will this hold true even when we are trying to do rolling upgrade to
1.0x as prescribed here - https://kafka.apache.org/documentation/#upgrade ?
Regards,
On 9 May 2018 at 11:06, Jan Filipiak <jan.filip...@trivago.com> wrote:
Hi,
this is expected. A gracefully shutdown means the broker is only
shutting down when it is not the leader of any partition.
Therefore you should not be able to gracefully shut down your entire
cluster.
Hope that helps
Best Jan
On 09.05.2018 12:02, M. Manna wrote:
Hello,
I have followed the graceful shutdown process by using the following (in
addition to the default controlled.shutdown.enable)
controlled.shutdown.max.retries=10
controlled.shutdown.retry.backoff.ms=3000
I am always having issues where not all the brokers are shutting
gracefully. And it's always Kafka, not zookeeper.
Has anyone experienced this discrepancy ? If so, could you please let me
know how to get around this issue?
Regards,
