[
https://issues.apache.org/jira/browse/YARN-415?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14099792#comment-14099792
]
Eric Payne commented on YARN-415:
---------------------------------
[~jianhe] and [~kkambatl]
Thank you both for your comments.
[~jianhe] wrote:
{quote}
Because of this, for consistency, I think we better use getCurrentAttempt to
charge finished containers against current attempt also for work-presrving am
restart?
{quote}
If I understand correctly, is the suggestion that all finished containers be
charged against the current attempt? That would be tricky, since even in a
normal use cases, an attempt can be in the complete state before all of its
containers are finished. Also, if the first attempt dies after some of its
containers are finished, then would the metrics for the finished containers
need to be transferred to the new attempt? I think that, since the metrics are
reported at the app level, charging the running containers to the current app
until the containers finish will be seemless to the end user. One thing that
could be done is to have RMAppAttemptMetrics#getRMAppMetrics get a copy of the
liveContainers and report only on the ones applicable to that attempt. That
seems like more overhead that may not be necessary.
[~kkambatl] wrote:
{quote}
Just took a look at the patch. The major concern I have is the use of
RMStateStore to store app resource usage information. If we add more resources
and more other statistics, storing all of them to the RM state store could be
placing too much overhead on the store, particularly if it is ZKRMStateStore.
Would it make more sense to store this information in the History/Timeline
store?
{quote}
Can you please help me to understand in more detail how this would be
accomplished?
> Capture memory utilization at the app-level for chargeback
> ----------------------------------------------------------
>
> Key: YARN-415
> URL: https://issues.apache.org/jira/browse/YARN-415
> Project: Hadoop YARN
> Issue Type: New Feature
> Components: resourcemanager
> Affects Versions: 0.23.6
> Reporter: Kendall Thrapp
> Assignee: Andrey Klochkov
> Attachments: YARN-415--n10.patch, YARN-415--n2.patch,
> YARN-415--n3.patch, YARN-415--n4.patch, YARN-415--n5.patch,
> YARN-415--n6.patch, YARN-415--n7.patch, YARN-415--n8.patch,
> YARN-415--n9.patch, YARN-415.201405311749.txt, YARN-415.201406031616.txt,
> YARN-415.201406262136.txt, YARN-415.201407042037.txt,
> YARN-415.201407071542.txt, YARN-415.201407171553.txt,
> YARN-415.201407172144.txt, YARN-415.201407232237.txt,
> YARN-415.201407242148.txt, YARN-415.201407281816.txt,
> YARN-415.201408062232.txt, YARN-415.201408080204.txt,
> YARN-415.201408092006.txt, YARN-415.201408132109.txt,
> YARN-415.201408150030.txt, YARN-415.patch
>
>
> For the purpose of chargeback, I'd like to be able to compute the cost of an
> application in terms of cluster resource usage. To start out, I'd like to
> get the memory utilization of an application. The unit should be MB-seconds
> or something similar and, from a chargeback perspective, the memory amount
> should be the memory reserved for the application, as even if the app didn't
> use all that memory, no one else was able to use it.
> (reserved ram for container 1 * lifetime of container 1) + (reserved ram for
> container 2 * lifetime of container 2) + ... + (reserved ram for container n
> * lifetime of container n)
> It'd be nice to have this at the app level instead of the job level because:
> 1. We'd still be able to get memory usage for jobs that crashed (and wouldn't
> appear on the job history server).
> 2. We'd be able to get memory usage for future non-MR jobs (e.g. Storm).
> This new metric should be available both through the RM UI and RM Web
> Services REST API.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
