[
https://issues.apache.org/jira/browse/HBASE-19772?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16323609#comment-16323609
]
stack commented on HBASE-19772:
-------------------------------
I tried it on one of my wonky machines. It did a hard hang. I can't jstack it.
W/o the patch, zk tests mostly pass but I'll then get an occasional test fail
usually in TestReadOnlyZKClient.
I've started builds on a GCE instead. Its much smoother. Interesting though is
I saw what I think is the hang here:
https://issues.apache.org/jira/browse/HBASE-19753?focusedCommentId=16321416&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-16321416
from the parent where Master gets stuck reading clusterid.
Ran the hbase-zookeeper module tests w/ this patch on GCE and its fine there.
All run promptly as opposed to slow and hanging on my wonky machines.
> Do not close connection to zk when there are still pending request in
> ReadOnlyZKClient
> --------------------------------------------------------------------------------------
>
> Key: HBASE-19772
> URL: https://issues.apache.org/jira/browse/HBASE-19772
> Project: HBase
> Issue Type: Sub-task
> Components: Zookeeper
> Reporter: stack
> Assignee: Duo Zhang
> Fix For: 2.0.0-beta-1
>
> Attachments: HBASE-19772-v2.patch, HBASE-19772.master.001.patch
>
>
> Here is [~Apache9] 's patch from the parent so it applies on top of what was
> committed in the parent.
> Patch makes it so we we don't close out zk if available Tasks to run and
> nicer logging.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
