[
https://issues.apache.org/jira/browse/MESOS-1221?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13974431#comment-13974431
]
Ian Downes edited comment on MESOS-1221 at 4/18/14 8:16 PM:
------------------------------------------------------------
Confirmed. This is indeed a bug as described by Vinod.
The resources aren't known at recover() so they cannot be added then. Instead,
the check at the start of update should be the promises hashmap.
Separately, there are too many different variables tracking container state in
MesosContainerizer. These should be unified.
Furthermore, this wasn't caught earlier because the slave doesn't check return
of Containerizer::update().
was (Author: idownes):
Confirmed. This is indeed a bug as described by Vinod.
The resources aren't known at recover() so they cannot be added then. Instead,
the check at the start of update should be the promises hashmap.
Separately, we there are too many different variables tracking container state
in MesosContainerizer. These should be unified.
> Slave should update the containerizers with executor resources after recovery
> -----------------------------------------------------------------------------
>
> Key: MESOS-1221
> URL: https://issues.apache.org/jira/browse/MESOS-1221
> Project: Mesos
> Issue Type: Bug
> Affects Versions: 0.18.0
> Reporter: Ian Downes
> Assignee: Vinod Kone
> Fix For: 0.18.1
>
>
> Executor resources are not included during recovery so containerizers don't
> know resource (cpu and memory) limits and won't include these in the resource
> statistics unless the resources are updated.
> See comment on 719 in mesos_containerizer.cpp.
> We should change the slave so it always 'updates' with the current resources.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
