[
https://issues.apache.org/jira/browse/HBASE-19753?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16321450#comment-16321450
]
Duo Zhang commented on HBASE-19753:
-----------------------------------
The problem is that, the keepalive time is too small, that it will close the
connection to zk while we are still trying to connect to zk.
The design philosophy here is that the keep alive time is usually much greater
than the operation time. The default value is 1 minute. Here we want to test
the keep alive mechanism so we set the keep alive time to 3 seconds, which is
too small for your environment. You can try up it to see if it helps.
And it is really strange that, why a zk connecting can take even more than 3
seconds? There must be something wrong. What's your OS sir? Linux or mac?
Thanks.
> Miscellany of fixes for hbase-zookeeper tests to make them more robust
> ----------------------------------------------------------------------
>
> Key: HBASE-19753
> URL: https://issues.apache.org/jira/browse/HBASE-19753
> Project: HBase
> Issue Type: Bug
> Reporter: stack
> Assignee: stack
> Attachments: HBASE-19753.branch-2.001.patch
>
>
> On my cluster which slows zk, tests hbase-zookeeper rarely all pass.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
