[
https://issues.apache.org/jira/browse/STORM-520?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15137660#comment-15137660
]
Grant Traynor commented on STORM-520:
-------------------------------------
Confirm the same issue and workaround on Storm 0.9.4. Our containers are
deployed via ansible onto multiple hosts. Using the "hostname" ansible setting
and not setting "storm.local.hostname" at all allowed storm to function
correctly.
The observed behaviour from my perspective was:
1. Topologies with a single worker would deploy and run fine
2. Topologies with multiple workers would hang trying to establish Netty
connections.
> Using storm.local.hostname causes bolts to not communicate with each other
> --------------------------------------------------------------------------
>
> Key: STORM-520
> URL: https://issues.apache.org/jira/browse/STORM-520
> Project: Apache Storm
> Issue Type: Bug
> Components: storm-core
> Affects Versions: 0.9.2-incubating
> Environment: Docker container (Docker > 1.1.2)
> Reporter: Daniel Schonfeld
>
> In our setup we run different supervisors as docker container on top of
> CoreOS.
> If I use docker's -h (host) and give the container the same hostname as the
> host machine, the bolts are able to register with nimbus correctly, talk to
> each other as well as transfer data inside the same container between
> different workers located on the same machine.
> Utilizing 'storm.local.hostname' instead of using docker's -h option results
> in the bolts having a hard time (read: unable to) communicating between
> workers on the same machine/container (we only run one container per physical
> machine which contains 4 workers + supervisor) and having sporadic success
> (50/50) communicating with bolts in different machines.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
