[
https://issues.apache.org/jira/browse/YARN-4284?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14966288#comment-14966288
]
Sunil G commented on YARN-4284:
-------------------------------
Thank you [~sjlee0] for the comments.
Yes, I understood your point and got the idea from the patch also. I was having
an assumption that, we are looking into a general blacklisting for all apps by
seeing a failure for one app attempt in a node. Thank you for clarifying the
same.
This change seems almost fine for me. But as you told, the solution is slightly
aggressive in marking a node as blacklisted per app. Also I am worried about
cases like preemption from RM ({{ContainerExitStatus.PREEMPTED}} or
{{KILLED_BY_RESOURCEMANAGER}}). Due to some queue over usage, RM may select AM
container to preempt (again this is very unlikely to happen with YARN-1496, but
its possible). And if application mark this node as blacklisted due to
preemption or some similar cases, its not so correct I think. How do you feel?
> condition for AM blacklisting is too narrow
> -------------------------------------------
>
> Key: YARN-4284
> URL: https://issues.apache.org/jira/browse/YARN-4284
> Project: Hadoop YARN
> Issue Type: Bug
> Components: resourcemanager
> Affects Versions: 2.8.0
> Reporter: Sangjin Lee
> Assignee: Sangjin Lee
> Attachments: YARN-4284.001.patch
>
>
> Per YARN-2005, there is now a way to blacklist nodes for AM purposes so the
> next app attempt can be assigned to a different node.
> However, currently the condition under which the node gets blacklisted is
> limited to {{DISKS_FAILED}}. There are a whole host of other issues that may
> cause the failure, for which we want to locate the AM elsewhere; e.g. disks
> full, JVM crashes, memory issues, etc.
> Since the AM blacklisting is per-app, there is little practical downside in
> blacklisting the nodes on *any failure* (although it might lead to
> blacklisting the node more aggressively than necessary). I would propose
> locating the next app attempt to a different node on any failure.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
