[
https://issues.apache.org/jira/browse/YARN-7617?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16292034#comment-16292034
]
Arun Suresh commented on YARN-7617:
-----------------------------------
[~cheersyang], to clarify the the container update workflow is as follows:
# The AM asks the RM to update the contianer
# The AM receives an updated container token from the RM.
# The NM is notified to update the running container
There are currently 2 ways by which step 3 can happen:
# The AM can explicitly use the NMClient to tell the NM to update the container
via {{NMClientAsync#updateContainerResourceAsync}} /
{{NMClientImpl#updateContainerResource}}
# if {{yarn.resourcemanager.auto-update.containers}} is set to true in the RM
conf, Then the AM does not have to do anything. The RM will instruct the NM to
update the container in the next heartbeat.
Ideally, Option 1 should be used to only to update a containers resources - but
as per the ContainerManagement protocol, there is basically no difference
between update container resource and execution type, since it is uniquely
determined by the updated container Token. (I had initially thought the NM
rejects those requests :) - which is why I assumed you have
auto-update.containers turned on)
Anyway. Long story short:
* Your patch actually looks fine. +1 pending the changes to the doc.
* I think we should also just add an updateContainer() method in the NMClient
just for completeness. (we can do this in another JIRA)
> Add a flag in distributed shell to automatically PROMOTE opportunistic
> containers to guaranteed once they are started
> ---------------------------------------------------------------------------------------------------------------------
>
> Key: YARN-7617
> URL: https://issues.apache.org/jira/browse/YARN-7617
> Project: Hadoop YARN
> Issue Type: Sub-task
> Components: applications/distributed-shell
> Affects Versions: 2.9.0
> Reporter: Weiwei Yang
> Assignee: Weiwei Yang
> Priority: Minor
> Attachments: YARN-7617.001.patch, promote_containers.jpg
>
>
> Per discussion in YARN-7610, it would be good to add such a flag, e.g
> {{promote_opportunistic_after_start}}. This is for the purpose of
> demonstrating how container promotion works.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
