[
https://issues.apache.org/jira/browse/YARN-5576?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15457446#comment-15457446
]
Jian He edited comment on YARN-5576 at 9/2/16 4:08 AM:
-------------------------------------------------------
Thanks for the review, Arun !
bq. We need to override the ContainerManagerImpl::localize() method in the
QueuingContainerManagerImpl. Re-localization should not be allowed if the
container is currently queued (not yet running)
It is not allowed, the method in ContainerManagerImpl only allows localization
while running
bq. I only see entries added to ResourceSet::resourcesFailedToBeLocalized set.
Shouldnt we remove these once the AM is notified of the failure ? Also,
Shouldn't these be notified back to the AM ? or we are just relying on the
diagnostic string sent to the AM in the GetContainerStatus response to notify
the AM ?
The status part is not yet implemented as mentioned in the parent jira. It'll
will be done once the requirement is clear. Earlier I was thinking these will
be sent as part of container status.
bq. wondering if we should have another RE_LOCALIZE_CONTAINER_RESOURCE event in
the ResourceLocalizationService to distinguish from the localization needed for
container initialization and correspondingly send different events to the
Container. Or maybe for the timebeing, we should just rename
INIT_CONTAINER_RESOURCE to LOCALIZE_CONTAINER_RESOURCE.
I don't think adding new events type for doing the same thing is necessary at
this point. This will also add additional complexity as you need to
conditionally sends different types of events. The goal is to reuse existing
code. I can rename it.
bq. : spurious change in the imports of ContainerImpl and BaseAMRMProxyTest
That's done by IDE auto fixing some unused imports, I edited it manually.
was (Author: jianhe):
bq. We need to override the ContainerManagerImpl::localize() method in the
QueuingContainerManagerImpl. Re-localization should not be allowed if the
container is currently queued (not yet running)
It is not allowed, the method in ContainerManagerImpl only allows localization
while running
bq. I only see entries added to ResourceSet::resourcesFailedToBeLocalized set.
Shouldnt we remove these once the AM is notified of the failure ? Also,
Shouldn't these be notified back to the AM ? or we are just relying on the
diagnostic string sent to the AM in the GetContainerStatus response to notify
the AM ?
The status part is not yet implemented as mentioned in the parent jira. It'll
will be done once the requirement is clear. Earlier I was thinking these will
be sent as part of container status.
bq. wondering if we should have another RE_LOCALIZE_CONTAINER_RESOURCE event in
the ResourceLocalizationService to distinguish from the localization needed for
container initialization and correspondingly send different events to the
Container. Or maybe for the timebeing, we should just rename
INIT_CONTAINER_RESOURCE to LOCALIZE_CONTAINER_RESOURCE.
I don't think adding new events type for doing the same thing is necessary at
this point. This will also add additional complexity as you need to
conditionally sends different types of events. The goal is to reuse existing
code. I can rename it.
bq. : spurious change in the imports of ContainerImpl and BaseAMRMProxyTest
That's done by IDE auto fixing some unused imports, I edited it manually.
> Core change to localize resource while container is running
> -----------------------------------------------------------
>
> Key: YARN-5576
> URL: https://issues.apache.org/jira/browse/YARN-5576
> Project: Hadoop YARN
> Issue Type: Sub-task
> Reporter: Jian He
> Assignee: Jian He
> Attachments: YARN-5576.1.patch, YARN-5576.2.patch, YARN-5576.3.patch,
> YARN-5576.4.patch
>
>
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org
