[
https://issues.apache.org/jira/browse/MESOS-715?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13783376#comment-13783376
]
Kevin Sweeney commented on MESOS-715:
-------------------------------------
We've observed very long failover times O(5-10min) in large production Aurora
clusters that could be significantly mitigated with this feature. While we've
been able to make data-structure optimizations to reduce this and can still do
more work in this area, there's still a near-linear processing delay during
scheduler startup and this work could be performed. As long as followers can
see log entries in order we can tolerate being arbitrarily far behind; this
feature request is only a performance optimization.
> Allow framework followers to read log
> -------------------------------------
>
> Key: MESOS-715
> URL: https://issues.apache.org/jira/browse/MESOS-715
> Project: Mesos
> Issue Type: Improvement
> Reporter: Joe Smith
>
> If a framework has a leader election, the newly elected leader needs to
> re-play the log of task transitions to build up the state of the world. It
> would help frameworks failover much faster if followers were able to read the
> log to keep themselves primed to take over.
--
This message was sent by Atlassian JIRA
(v6.1#6144)
