[
https://issues.apache.org/jira/browse/HADOOP-16256?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16821265#comment-16821265
]
Jon Bringhurst edited comment on HADOOP-16256 at 4/18/19 4:15 PM:
------------------------------------------------------------------
We've had this issue on 2.9.2, but it appears to affect 2.9.0+, 3.0.0-alpha4+.
Our current workaround is to have an external process periodically kill off
"stuck" ContainerLocalizer processes.
was (Author: jonbringhurst):
We've had this issue on 2.9.2, but it appears to affect 2.9.0+, 3.0.0-alpha4+.
> HTTPFileSystem disallows users from configuring the connectionTimeout and
> readTimeout of HttpClient.
> ----------------------------------------------------------------------------------------------------
>
> Key: HADOOP-16256
> URL: https://issues.apache.org/jira/browse/HADOOP-16256
> Project: Hadoop Common
> Issue Type: Improvement
> Reporter: Shanthoosh Venkataraman
> Priority: Major
>
> HADOOP-14383 added a Hadoop filesystem API implementation to localize
> resources from HTTP server for container localization.
> Currently HttpFileSystem uses the default HttpClient connectionTimeout and
> readTimeout, which is infinity in most JVM.
> When the HTTP server is slow, this prolongs the execution time of the
> container localization phase and in some cases we've observed that it runs
> for hours.
> It would be better to choose reasonable defaults for connectionTimeout and
> readTimeout of HttpClient and allow users to configure them.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
