[
https://issues.apache.org/jira/browse/MESOS-1584?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14060915#comment-14060915
]
brian wickman commented on MESOS-1584:
--------------------------------------
I don't think this is quite what I had in mind. It's my understanding that
mesos-resolve does a single resolve and then exits.
Compare to the mesos-slave which does realtime master detection including
failover detection, in addition to proxying framework messages on behalf of
executors to whomever is leading at the time. It seems like we should have the
analogue for schedulers as well.
> allow slave to proxy scheduler registration in service to master detection
> --------------------------------------------------------------------------
>
> Key: MESOS-1584
> URL: https://issues.apache.org/jira/browse/MESOS-1584
> Project: Mesos
> Issue Type: Improvement
> Components: framework
> Reporter: brian wickman
>
> This is just an idea -- right now each pure language binding will need to
> implement a master detector (which in most cases means your language will
> need zookeeper bindings.)
> Instead of each binding needing to implement a master detector, perhaps the
> slave should support a 'proxy' mode whereby it acts as master detector on
> behalf of a dumber pure language binding. This could also help facilitate
> the launching of replica schedulers. Alternately this could be a separate
> binary but could just as easily live in the slave process.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
