[
https://issues.apache.org/jira/browse/KAFKA-1790?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14234496#comment-14234496
]
James Oliver commented on KAFKA-1790:
-------------------------------------
Thanks for the input, Joel. We'll keep that in mind as a possible solution.
Jun, we are using Apache Mesos which uses resource isolation instead of VMs to
schedule distributed workloads (Kafka in this case). I'm not sure if it's
considered the exception anymore - it is a different way of doing things. For
example, Google uses isolation instead of VMs for its datacenter computing. In
this context, there is no VM but simply a sandbox (path on some mesos-slave).
To bring down an instance, the entire process tree is sent a signal to
shutdown. However, the shutdown timeouts in our scheduling framework are mostly
out of our control - there's no good way to guarantee controlled shutdown
occurs before the process is forced to terminate. In order to make sure it
happens, we instead thought we might send a request directly to the broker to
perform the controlled shutdown and, once everything is jolly, send the kill
signal (which causes a restart). Since the remote controlled shutdown went
away, we now think we might need to defer that responsibility to a wrapper
script with a simple remote interface. Thoughts?
> Remote controlled shutdown was removed
> --------------------------------------
>
> Key: KAFKA-1790
> URL: https://issues.apache.org/jira/browse/KAFKA-1790
> Project: Kafka
> Issue Type: Bug
> Components: core
> Affects Versions: 0.8.2
> Reporter: James Oliver
> Assignee: James Oliver
> Priority: Blocker
> Fix For: 0.8.2
>
>
> In core:
> kafka.admin.ShutdownBroker was removed, rendering remote controlled shutdowns
> impossible.
> A Kafka administrator needs to be able to perform a controlled shutdown
> without issuing a SIGTERM/SIGKILL.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
