[ 
https://issues.apache.org/jira/browse/MESOS-6419?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15676992#comment-15676992
 ] 

Neil Conway commented on MESOS-6419:
------------------------------------

[~adam-mesos] -- I've been working on this for a while; should have RRs up 
today. However, the approach we decided to take involved a pretty significant 
change to how we represent "recovered" frameworks that haven't re-registered 
after master failover (current diff: "24 files changed, 1278 insertions(+), 584 
deletions(-)")

If we want to backpatch to 1.1.1, it would be possible to develop a minimal fix 
to just enable teardown to work for unregistered frameworks.

> The 'master/teardown' endpoint should support tearing down 
> 'unregistered_frameworks'.
> -------------------------------------------------------------------------------------
>
>                 Key: MESOS-6419
>                 URL: https://issues.apache.org/jira/browse/MESOS-6419
>             Project: Mesos
>          Issue Type: Bug
>          Components: master
>    Affects Versions: 0.26.2, 0.27.3, 0.28.2, 1.0.1
>            Reporter: Gilbert Song
>            Assignee: Neil Conway
>            Priority: Critical
>              Labels: endpoint, master
>
> This issue is exposed from 
> [MESOS-6400](https://issues.apache.org/jira/browse/MESOS-6400). When a user 
> is trying to tear down an 'unregistered_framework' from the 'master/teardown' 
> endpoint, a bad request will be returned: `No framework found with specified 
> ID`.
> Ideally, we should support tearing down an unregistered framework, since 
> those frameworks may occur due to network partition, then all the orphan 
> tasks still occupy the resources. It would be a nightmare if a user has to 
> wait until the unregistered framework to get those resources back.
> This may be the initial implementation: 
> https://github.com/apache/mesos/commit/bb8375975e92ee722befb478ddc3b2541d1ccaa9



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

Reply via email to