You are viewing a plain text version of this content. The canonical link for it is here.
Posted to yarn-issues@hadoop.apache.org by "Karthik Kambatla (JIRA)" <ji...@apache.org> on 2016/12/14 22:47:58 UTC

[jira] [Commented] (YARN-5600) Allow setting yarn.nodemanager.delete.debug-delay-sec on a per-application basis

    [ https://issues.apache.org/jira/browse/YARN-5600?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15749754#comment-15749754 ] 

Karthik Kambatla commented on YARN-5600:
----------------------------------------

This is quite handy. Thanks for working on this Miklos. 

Comments:
# ApplicationConstants: typo in the javadoc. 
# YarnConfiguration: max-per-application-debug-delay-sec
## "per-application" is redundant and makes the config long. Can we just have {{max-debug-delay-sec}} or {{debug-delay-sec.max}}.
## I don't see the need to have a separate constant for the corresponding default value. Can we use the value of {{delete.debug-delay-sec}} or a multiple of it (say, 2) as default instead?
## Also, looks like max-debug-delay-sec could be smaller than debug-delay-sec. That is confusing. Can we make sure max is actually max?
# Move the javadoc from ApplicationImpl to Application. Also, can we linkify {{Application.NEVER}}.
# If the admin sets debug-delay-sec to a positive value (say, 5 mins) and the user explicitly sets it to a smaller value (even 0), shouldn't the user's setting override for all containers of that app? 
# It appears we allow setting this value on a per-container basis through ContainerLaunchContext, but consider only the largest value. Is there value in allowing specifying this on a per-container basis. Why not specify it on a per-app basis? Say, through ApplicationSubmissionContext?
# Any specific reason for capturing the deletion time in Date instead of long? Isn't Date more expensive? 
# yarn-default: s/mumber/number
# DeletionServe
## scheduleFileDeletionTask takes a delay, and then adjusts it based on configuration. These seems error-prone. Callers of this method would assume the file will be deleted at the provided time. Can the caller take care of the adjustment?
## getEffectiveDelay: if the limitedDelay > maxDebugDelay, one could set limitedDelay to maxDebugDelay without computing Math.min. 

I haven't looked at the tests yet. Will take a look in my next round. 

> Allow setting yarn.nodemanager.delete.debug-delay-sec on a per-application basis
> --------------------------------------------------------------------------------
>
>                 Key: YARN-5600
>                 URL: https://issues.apache.org/jira/browse/YARN-5600
>             Project: Hadoop YARN
>          Issue Type: Improvement
>          Components: nodemanager
>    Affects Versions: 3.0.0-alpha1
>            Reporter: Daniel Templeton
>            Assignee: Miklos Szegedi
>              Labels: oct16-medium
>         Attachments: YARN-5600.000.patch, YARN-5600.001.patch, YARN-5600.002.patch, YARN-5600.003.patch, YARN-5600.004.patch, YARN-5600.005.patch, YARN-5600.006.patch, YARN-5600.007.patch, YARN-5600.008.patch, YARN-5600.009.patch, YARN-5600.010.patch, YARN-5600.011.patch, YARN-5600.012.patch, YARN-5600.013.patch, YARN-5600.014.patch, YARN-5600.015.patch, YARN-5600.016.patch
>
>
> To make debugging application launch failures simpler, I'd like to add a parameter to the CLC to allow an application owner to request delayed deletion of the application's launch artifacts.
> This JIRA solves largely the same problem as YARN-5599, but for cases where ATS is not in use, e.g. branch-2.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)

---------------------------------------------------------------------
To unsubscribe, e-mail: yarn-issues-unsubscribe@hadoop.apache.org
For additional commands, e-mail: yarn-issues-help@hadoop.apache.org