zookeeper-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From chao chu <chuchao...@gmail.com>
Subject Re: Multiple disconnected events received within one millisecond (using zookeeper 3.3.5)
Date Tue, 02 Jul 2013 16:38:00 GMT
sorry for the confusion, by looking into our own logs very carefully, it
turned out that we ran into an edge case in our own code that we mistakenly
created multiple sessions during the same period. I think that might
explain why we saw some repeat disconnected/connected events in quite a
short time. however, from what i can tell, all of the sessions got expired
at almost the same time (when the connection to zk ensemble re-established?
Since I have no access to the zk server side logs, I have no way to
verify). Does it make sense?

On Fri, Jun 28, 2013 at 10:35 PM, chao chu <chuchao333@gmail.com> wrote:

> Hi,
> we observed some very strange behavior, from our server's log, we saw tens
> of disconnected events received within one millisecond (we basically logged
> every zk events), which doesn't make much sense to me.
> some details about this:
> due to some flaw of our own code, which caused continuous major GC pauses
> (the longest one is about 30s), we saw some repeated
> "connected->disconnected->expired" events and we got tens of stale znodes
> (we have created some ephemeral-sequential znodes for leader election).
> I have two questions on this:
> 1. should it be guaranteed that the ephemeral nodes will be deleted when
> session got expired?
> 2. why there are so many continuous 'Disconnected' events received? I saw
> there was a quite old discussions related to this back to 2009 (
> http://www.mail-archive.com/zookeeper-user@hadoop.apache.org/msg00184.html),
> does this remain the same?
> Thanks & Regards,
> --
> ChuChao


  • Unnamed multipart/alternative (inline, None, 0 bytes)
View raw message