Hi all,

We had the similar need with ZOOKEEPER-2437. During leader elections and
other high load situations, we see a lot of disconnects. We wanted to
understand whether

   -

   the disconnects were expected client behavior and not caused by any
   server action


   -

   the servers were just too overloaded so the clients would disconnect and
   try another host
   -

   the server was disconnecting the client due to some error
   -

   some action (or inaction) on the ZK host causes the client to disconnect
   without doing any useful work.

In general, we just didn't have the visibility into the causes of the
disconnects so it was harder for us to direct our investigations in the
right direction. The stats are just as much about ruling out possible
issues as identifying an actual root cause. We want to understand how often
we have unexpected disconnects and what are the causes for those unexpected
disconnects. It might be useful to have that support on the open source
code base since it has valuable server/client behavior insights on the
quality of the zookeeper operation.

Regards,
Seda

On Tue, May 31, 2016 at 3:39 PM, Dan Benediktson <dbenedikt...@twitter.com>
wrote:

>
> ---------- Forwarded message ----------
> From: Flavio Junqueira (JIRA) <j...@apache.org>
> Date: Tue, May 31, 2016 at 12:48 PM
> Subject: [jira] [Commented] (ZOOKEEPER-2437) Need detailed diagnostics for
> Zookeeper Connection Issues
> To: dev@zookeeper.apache.org
>
>
>
>     [
> https://issues.apache.org/jira/browse/ZOOKEEPER-2437?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15308461#comment-15308461
> ]
>
> Flavio Junqueira commented on ZOOKEEPER-2437:
> ---------------------------------------------
>
> Thanks for reporting this issue, but I'm not sure what the ask is and if
> it needs to be done by zookeeper as opposed to solr. What additional
> diagnostics is the description referring to?
>
> > Need detailed diagnostics for Zookeeper Connection Issues
> > ---------------------------------------------------------
> >
> >                 Key: ZOOKEEPER-2437
> >                 URL:
> https://issues.apache.org/jira/browse/ZOOKEEPER-2437
> >             Project: ZooKeeper
> >          Issue Type: Improvement
> >          Components: java client
> >    Affects Versions: 3.5.1
> >            Reporter: harcor
> >            Priority: Critical
> >
> > Using a zookeeper ensemble with Apache Solr, the client connection
> (socket) can be disconnected by either Solr or Zookeeper.  If the
> connection fails on the Zookeeper and we are in DEBUG mode then additional
> diagnostics should be written to the log withe connection exception.
>
>
>
> --
> This message was sent by Atlassian JIRA
> (v6.3.4#6332)
>
>

Reply via email to