[
https://issues.apache.org/jira/browse/SLIDER-1166?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15467761#comment-15467761
]
Josh Elser commented on SLIDER-1166:
------------------------------------
I'm not sure if the ZooKeeper heartbeat thread is resilient to the underlying
connection being terminated. One thing you could try is running {{kill -SIGSTOP
<pid>}}, wait for a minute, and then run {{kill -SIGCONT <pid>}}. This would
suspend the entire process, preventing the ZK heartbeat thread from running
(simulating a stop-the-world GC cycle, actually).
Another thing I thought of is the Exceptions that ZK might throw which need to
be handled by the application. Here's a good example from Accumulo:
https://github.com/apache/accumulo/blob/5cb5b9372103761c829403c03007b9f53241400f/fate/src/main/java/org/apache/accumulo/fate/zookeeper/ZooReader.java#L70-L87.
> Every cluster stop operation creates and holds on to a zk session
> -----------------------------------------------------------------
>
> Key: SLIDER-1166
> URL: https://issues.apache.org/jira/browse/SLIDER-1166
> Project: Slider
> Issue Type: Bug
> Components: client
> Affects Versions: Slider 0.91
> Reporter: Billie Rinaldi
> Assignee: Billie Rinaldi
> Priority: Critical
> Fix For: Slider 1.0.0
>
> Attachments: SLIDER-1166.1.patch, SLIDER-1166.2.patch
>
>
> We aren't closing the zookeeper client, leaving it to expire and keep the
> session open much longer than is necessary.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
