[jira] [Created] (SAMZA-770) JobCoordinator -> Util.getLocalHost selects wrong network interface when multiple are available

["Michael Martin (JIRA)" <ji...@apache.org>]

Michael Martin created SAMZA-770:
------------------------------------

             Summary: JobCoordinator -> Util.getLocalHost selects wrong network interface when multiple are available
                 Key: SAMZA-770
                 URL: https://issues.apache.org/jira/browse/SAMZA-770
             Project: Samza
          Issue Type: Bug
            Reporter: Michael Martin
            Priority: Blocker


Here's the scene:

I am running my Samza (0.9.0) job in a YARN (2.7.1) cluster that is running in docker (1.7.1) and networked together using Weave (1.1.0)

When the YARN AM starts up the first Samza container on the master node it starts the Job Coordinator's HttpServer using the IP of the 'eth0' network interface because that is the first valid interface that it found.

When the second Samza container is started on the slave node, it tries to fetch configuration from the Coordinator URL, but we get "Connection Refused" because it cannot communicate with the master node on the IP assigned to the master's 'eth0' network interface.

Essentially we need the Coordinator URL to use the hostname, which will resolve to the correct IP on the master's 'ethwe' network interface from any machine on the Weave network.

Any thoughts?



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)