[
https://issues.apache.org/jira/browse/FLINK-2722?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14902240#comment-14902240
]
ASF GitHub Bot commented on FLINK-2722:
---------------------------------------
Github user tillrohrmann commented on a diff in the pull request:
https://github.com/apache/flink/pull/1159#discussion_r40064104
--- Diff:
flink-runtime/src/main/java/org/apache/flink/runtime/net/NetUtils.java ---
@@ -276,6 +289,13 @@ private static InetAddress
findAddressUsingStrategy(AddressDetectionState strate
InetAddress interfaceAddress = ee.nextElement();
switch (strategy) {
+ case LOCAL_HOST:
--- End diff --
Why do we execute the `LOCAL_HOST` strategy for every interface? I thought
we use `InetAddress.getLocalHost` as the outbound interface.
> Use InetAddress.getLocalHost() first when detecting TaskManager IP address
> --------------------------------------------------------------------------
>
> Key: FLINK-2722
> URL: https://issues.apache.org/jira/browse/FLINK-2722
> Project: Flink
> Issue Type: Bug
> Components: Distributed Runtime, TaskManager
> Affects Versions: 0.9, 0.10
> Reporter: Robert Metzger
> Assignee: Robert Metzger
> Fix For: 0.9.2
>
>
> A user reported a connection issue with Netty being unable to connect to a
> TaskManager to subscribe to an intermediate result.
> The problem occurred when the TaskManager and JobManager were running on the
> same host (something that can easily happen on YARN).
> In that case, the TaskManager was reporting a host-local ip address to the
> JobManager when connecting.
> To avoid the issue in the future, the TaskManager first tries to use the
> hostname returned by InetAddress.getLocalHost(). In a properly set-up
> environment, this will return a connection which is accessible by all
> machines in a cluster.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
