[ https://issues.apache.org/jira/browse/SPARK-23155?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16758102#comment-16758102
]
Gera Shegalov commented on SPARK-23155:
---------------------------------------
[~kabhwan], [~vanzin] I would still be interested to be able to use the new mechanism with
the old logs. [https://github.com/apache/spark/pull/23720] is a quick draft to demo how we
could achieve this flexibly with named capture groups.
> YARN-aggregated executor/driver logs appear unavailable when NM is down
> -----------------------------------------------------------------------
>
> Key: SPARK-23155
> URL: https://issues.apache.org/jira/browse/SPARK-23155
> Project: Spark
> Issue Type: Improvement
> Components: Deploy
> Affects Versions: 2.2.1
> Reporter: Gera Shegalov
> Assignee: Jungtaek Lim
> Priority: Major
> Fix For: 3.0.0
>
>
> Unlike MapReduce JobHistory Server, Spark history server isn't rewriting container log
URL's to point to the aggregated yarn.log.server.url location and relies on the NodeManager
webUI to trigger a redirect. This fails when the NM is down. Note that NM may be down permanently
after decommissioning in traditional environments or when used in a cloud environment such
as AWS EMR where either worker nodes are taken away with autoscale, the whole cluster is used
to run a single job.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org
|