[
https://issues.apache.org/jira/browse/YARN-4389?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15101824#comment-15101824
]
Rohith Sharma K S commented on YARN-4389:
-----------------------------------------
Hi [~sunilg]/[~djp] , Overall patch looks good to me.
I think I found one issue(was there earlier also), we can fix in this JIRA if
agreed else we can raise new JIRA. I am fine with either of the ways. Please
provide your thoughts.
If cluster configured with global {{blacklistDisableFailureThreshold<0}} OR
{{blacklistDisableFailureThreshold > 1}} by mistake, then also we need to reset
to MINIMUM_THRESHOLD_VALUE or MAXIMUM_THRESHOLD_VALUE.
Consider case where {{blacklistDisableFailureThreshold > 1}} then
{{SimpleBlacklistManager#getBlacklistUpdates}} will always go for blacklist
addition. It never removes the nodes even though all the nodes in the cluster
are blacklisted.
> "yarn.am.blacklisting.enabled" and
> "yarn.am.blacklisting.disable-failure-threshold" should be app specific
> rather than a setting for whole YARN cluster
> -------------------------------------------------------------------------------------------------------------------------------------------------------
>
> Key: YARN-4389
> URL: https://issues.apache.org/jira/browse/YARN-4389
> Project: Hadoop YARN
> Issue Type: Bug
> Components: applications
> Reporter: Junping Du
> Assignee: Sunil G
> Priority: Critical
> Attachments: 0001-YARN-4389.patch, 0002-YARN-4389.patch,
> 0003-YARN-4389.patch, 0004-YARN-4389.patch, 0005-YARN-4389.patch,
> 0006-YARN-4389.patch, 0007-YARN-4389.patch, 0008-YARN-4389.patch,
> 0009-YARN-4389.patch
>
>
> "yarn.am.blacklisting.enabled" and
> "yarn.am.blacklisting.disable-failure-threshold" should be application
> specific rather than a setting in cluster level, or we should't maintain
> amBlacklistingEnabled and blacklistDisableThreshold in per rmApp level. We
> should allow each am to override this config, i.e. via submissionContext.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
