[
https://issues.apache.org/jira/browse/YARN-1041?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13867184#comment-13867184
]
Vinod Kumar Vavilapalli commented on YARN-1041:
-----------------------------------------------
Knowing well this is an early patch, quick review comments
- Document RegisterApplicationMasterResponse.getRunningContainers() with its
semantics. Particularly about allocated, acquired, reserved containers from
previous AM being killed.
- yarn_service_protos.proto: running_containers ->
containers_from_previous_attempt ? Running or not may(or may not) change
depending on the implementation. If we do this, we should change the name
everywhere.
- YarnScheduler.getAllRunningContainers(): Rename and also deal with an
app-attempt instead of application? Because that is how scheduler looks at
containers - per attempt.
- Per [YARN-1490
comment|https://issues.apache.org/jira/browse/YARN-1490?focusedCommentId=13866267&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-13866267],
we need to take care of NMTokens too.
The code duplication in schedulers is a bigger problem that needs to be
addressed. But it has to start somewhere. I am +1 for starting with a
AbstractYarnScheduler.
> Protocol changes for RM to bind and notify a restarted AM of existing
> containers
> --------------------------------------------------------------------------------
>
> Key: YARN-1041
> URL: https://issues.apache.org/jira/browse/YARN-1041
> Project: Hadoop YARN
> Issue Type: Sub-task
> Components: resourcemanager
> Affects Versions: 3.0.0
> Reporter: Steve Loughran
> Assignee: Jian He
> Attachments: YARN-1041.1.patch
>
>
> For long lived containers we don't want the AM to be a SPOF.
> When the RM restarts a (failed) AM, it should be given the list of containers
> it had already been allocated. the AM should then be able to contact the NMs
> to get details on them. NMs would also need to do any binding of the
> containers needed to handle a moved/restarted AM.
--
This message was sent by Atlassian JIRA
(v6.1.5#6160)
