[
https://issues.apache.org/jira/browse/KAFKA-5014?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15956915#comment-15956915
]
ASF GitHub Bot commented on KAFKA-5014:
---------------------------------------
GitHub user ijuma opened a pull request:
https://github.com/apache/kafka/pull/2813
KAFKA-5014: NetworkClient.leastLoadedNode should check if channel is ready
You can merge this pull request into a Git repository by running:
$ git pull https://github.com/ijuma/kafka
kafka-5014-least-loaded-node-should-check-if-channel-is-ready
Alternatively you can review and apply these changes as the patch at:
https://github.com/apache/kafka/pull/2813.patch
To close this pull request, make a commit to your master/trunk branch
with (at least) the following in the commit message:
This closes #2813
----
commit a418b0157eab6740967852ea601e1f94f30cf630
Author: Ismael Juma <ism...@juma.me.uk>
Date: 2017-04-05T14:13:16Z
KAFKA-5014: NetworkClient.leastLoadedNode should check if channel is ready
----
> SSL Channel not ready but tcp is established and the server is hung will not
> sending metadata
> ---------------------------------------------------------------------------------------------
>
> Key: KAFKA-5014
> URL: https://issues.apache.org/jira/browse/KAFKA-5014
> Project: Kafka
> Issue Type: Bug
> Affects Versions: 0.9.0.1, 0.10.2.0
> Reporter: Pengwei
> Priority: Minor
> Fix For: 0.11.0.0
>
>
> In our test env, QA hang one of the connecting broker of the producer, then
> the producer will be stuck in send method, and throw the exception: fail to
> update metadata after request timeout.
> I found the reason as follow: when the producer chose one of the broker to
> send metadata, it connect to the broker, but the broker is hang, the tcp is
> connected and Network client marks this broker is connected, but the SSL
> channel is not ready yet so the channel is not ready.
> Then the Network client chooses the connected node in the leastLoadedNode
> every time to send the metadata, but the node's channel is not ready yet.
> So the producer stuck in getting metadata and will not try another node to
> request metadata. The client should not stuck only one node is hung
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
