zookeeper-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From tobe <tobeg3oo...@gmail.com>
Subject Re: Why sessionTimeout is restricted by tickTime
Date Thu, 28 Aug 2014 08:00:31 GMT
Thanks @Deng. I think I should set maxSessionTimeout rather than increasing
the tickTime. The default value of max session timeout is confusing and it
took us some time to notice this limitation. I will update the log to print
negotiationSessionTimeout rather than the sessionTimeout from client.


On Thu, Aug 28, 2014 at 12:35 PM, Hongchao Deng <fengjingchao@hotmail.com>
wrote:

> My understanding is that large session timeout limits scalability of
> client numbers. Theoretically speaking, it is possible to increase the
> maximum session timeout in your case.
>
> - Hongchao Deng
>
>
> > Date: Thu, 28 Aug 2014 12:22:52 +0800
> > Subject: Re: Why sessionTimeout is restricted by tickTime
> > From: tobeg3oogle@gmail.com
> > To: user@zookeeper.apache.org
> > CC: dev@zookeeper.apache.org
>
> >
> > By the way, it comes down to a conflict if one ZooKeeper client set
> timeout
> > as 10 and one client set timeout as 110. The tickTime can't satisfy all.
> >
> >
> > On Thu, Aug 28, 2014 at 11:53 AM, tobe <tobeg3oogle@gmail.com> wrote:
> >
> > > Refer to the official guide, we know that the negotiationTimeout must
> be
> > > greater than twice tickTime and smaller than 20 times of tickTime.
> > >
> > > I know the tickTime should not be too large or even larger than
> > > sessionTimeout, or the session will time out all the time. This
> limitation
> > > is reasonable, but why the sessionTimeout must be smaller than 20
> times of
> > > tickTime?
> > >
> > > The problem happens in our production environment. We share a ZooKeeper
> > > cluster for a few HBase cluster. By default, the tickTime is 2s and the
> > > sessionTimeout is 30s. It's ok. But now one of our HBase clusters need
> to
> > > increase the sessionTimeout to 90s because of long time gc. After
> updating
> > > the sessionTimout in HBase, we have to update the tickTime in
> ZooKeeper.
> > > And this setting may have an effect on other HBase clusters.
> > >
> > > So I wonder why ZooKeeper has this limitation. Can anyone explain to
> me?
> > >
>

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