[
https://issues.apache.org/jira/browse/SAMZA-2638?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17306842#comment-17306842
]
Bharath Kumarasubramanian commented on SAMZA-2638:
--------------------------------------------------
We also make changes to how processors that are not part of job model behave.
i.e. they no longer shut themselves down rather stay active and being part of
the quorum. It has following benefits
* Idle processors can be part of the quorum when some active processors goes
down without impacting other active processors
* Smaller debounce window results in rebalances where a processor that joins
after the rebalance start but before completion, ends up getting zk
notifications about the rebalance and terminates itself as its not part of the
job model. The behavior is quite detrimental where we kill the processor that
otherwise should be running.
> Skip container stop/restart for processors that have same work assignment
> across rebalances
> -------------------------------------------------------------------------------------------
>
> Key: SAMZA-2638
> URL: https://issues.apache.org/jira/browse/SAMZA-2638
> Project: Samza
> Issue Type: Improvement
> Reporter: Bharath Kumarasubramanian
> Assignee: Bharath Kumarasubramanian
> Priority: Major
>
> *Problem*:
> As part of rebalance, we always expire the current work assignment and
> proceed to signal consensus and start the container with new work assignment.
> It is inefficient for the processors to do the above when there is no changes
> in the work assignment between the active job model & the proposed job model.
> *Changes*:
> Processors perform onJobModelExpired and onNewJobModel only if there are
> changes to their work assignment across the active and proposed job model.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
