[
https://issues.apache.org/jira/browse/MESOS-2308?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14612714#comment-14612714
]
Benjamin Mahler commented on MESOS-2308:
----------------------------------------
{quote}
Currently, if you call reconcileTasks with an empty list, I do not know of any
way to know when the reconciliation has finished.
This would be a really nice feature since a framework would not have to persist
task state anymore because it can recover that state from Mesos on startup.
{quote}
This is unrelated, feel free to file a ticket for this. We have discussed this
before but didn't have any strong reasons for adding it, so worth a discussion.
:)
{quote}
Without this feature, a framework like Marathon might try to scale up an
application unnecessarily because it has not yet received information about all
tasks.
{quote}
Hm.. I don't see how this follows, but would love to understand this better.
Mind starting a discussion either in the ticket you file for the above request,
or on the dev list?
> Task reconciliation API should support data partitioning
> --------------------------------------------------------
>
> Key: MESOS-2308
> URL: https://issues.apache.org/jira/browse/MESOS-2308
> Project: Mesos
> Issue Type: Story
> Reporter: Bill Farner
>
> The {{reconcileTasks}} API call requires the caller to specify a collection
> of {{TaskStatus}}es, with the option to provide an empty collection to
> retrieve the master's entire state. Retrieving the entire state is the only
> mechanism for the scheduler to learn that there are tasks running it does not
> know about, however this call does not allow incremental querying. The
> result would be that the master may need to send many thousands of status
> updates, and the scheduler would have to handle them. It would be ideal if
> the scheduler had a means to partition these requests so it can control the
> pace of these status updates.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
