Mailing-List: contact user-help@zookeeper.apache.org; run by ezmlm
Precedence: bulk
Reply-To: user@zookeeper.apache.org
Received-SPF: pass (athena.apache.org: domain of ted.dunning@gmail.com
 designates 209.85.220.176 as permitted sender)
DomainKey-Signature: a=rsa-sha1; c=nofws;
        d=gmail.com; s=gamma;
        h=mime-version:in-reply-to:references:from:date:message-id:subject:to
         :cc:content-type;
        b=uiacY7FJpqEfGpAtbCHe90rdelPuTNXYnVkAnwadaipzM4S0c3Yq7HW/4XHVhiQ34t
         bu2qRURRRurpgfv0VzabIi62klMRgpEEMRY3ustzNXgBuJGoTnWoJ1kfvjH3FpcMI5C5
         wAcnMfzQ7zGO+OTAco1Flgg/2amQnmUCTgSVg=
MIME-Version: 1.0
In-Reply-To: <BANLkTimyycKopmpO9=8VkoWRZ-7xrh0eGA@mail.gmail.com>
References: <BANLkTin1Ao+DiGYBPz6MQq12mBnQLNENJQ@mail.gmail.com>
 <BD178E2AD91D544A8CE7F6BE2D9462B3265D30DDB9@SP2-EX07VS05.ds.corp.yahoo.com>
 <BANLkTinePWm9n5VMDTd5c7xUAq191_MwCg@mail.gmail.com>
 <660D007A-C5E2-496B-A73E-CBE2560BB976@me.com>
 <BANLkTi=P7qYpgCO3YsL7MubpubkTSoNxdQ@mail.gmail.com>
 <BANLkTimyycKopmpO9=8VkoWRZ-7xrh0eGA@mail.gmail.com>
From: Ted Dunning <ted.dunning@gmail.com>
Date: Thu, 5 May 2011 10:05:59 -0700
Message-ID: <BANLkTi=s0A+O3F-nNBQf1LCgojNx2YqDPA@mail.gmail.com>
Subject: Re: Growing a cluster
To: user@zookeeper.apache.org
Cc: "zookeeper-user@hadoop.apache.org" <zookeeper-user@hadoop.apache.org>
Content-Type: multipart/alternative; boundary=bcaec501647350e26604a28a6467

--bcaec501647350e26604a28a6467
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: quoted-printable

To be explicit on a (probably) obvious point, you will have a few tens of
seconds of downtime during the restart
of server 1 until the second server comes on-line and participates in the
election.

It there is a difference between 10 seconds of downtime and 2 minutes of
downtime in your application, I recommend
going from 1 to 3 nodes first then to 3+.  It is easier to have a very shor=
t
downtime when only a single additional node
needs to come up.  You might also be able to start server 2 before you
bounce server 1, but you have to be very careful
not to start enough machines to have a quorum before server 1 joins in.

2011/5/5 Patrick Hunt <phunt@apache.org>

> Hi Jared, you can go from 1 to 2+ nodes just fine:
>
> 1) change the configuration of server 1 from standalone to quorum
> based (be sure to list the new members of the ensemble)
> 2) add a myid file for server 1
> 3) setup new servers 2+
> 4) restart server 1
> 5) start servers 2+
>
> At this point the quorum should just come up. I just tried this
> example and it worked fine for me going from 1 to 3 (the data I
> created in standalone was available once I started the ensemble).
>
> Regards,
>
> Patrick
>
> 2011/5/5 Jared Cantwell <jared.cantwell@gmail.com>:
> > Great information guys-- this helps me understand what needs done when
> > expanding from 2 nodes to X nodes.
> >
> > Does anyone have insight on going from 1 node to 3 nodes?
> >
> > ~Jared
> >
> > 2011/5/5 Chang Song <tru64ufs@me.com>
> >
> >>
> >> We are in a bit similar situation.
> >>
> >> 3 node -> 5 node ensemble.
> >>
> >> The only way to do this is the following.
> >>
> >> Assumption is that we have one DNS hostname for three zookeeper ensemb=
le
> >> IP.
> >> Since five node ensemble allows 2 node failure for quorum, we can do
> >>
> >>
> >> 0. First all two new ensemble IPs
> >>
> >> 1. change all three existing node config (zoo.cfg) and add two new nod=
e
> >> information
> >>    Restart all three existing nodes in a sequence.
> >>
> >> 2. Replicate the new existing zoo.cfg to two new ensemble
> >>    Start Zookeeper on two new server
> >>
> >>
> >> You can do this in backward sequence (0 -> 2 -> 1)
> >> In your case, you can do 0 -> 2 -> 1, I think.
> >>
> >> Chang
> >>
> >>
> >>
> >> 2011. 5. 5., =EC=98=A4=ED=9B=84 9:57, Jared Cantwell =EC=9E=91=EC=84=
=B1:
> >>
> >> > It would be acceptable to me to do this non-dynamically and
> non-rolling
> >> as
> >> > well. For example, I can shut down all nodes, make necessary
> >> modifications
> >> > to config files, and then restart all nodes.  If I do this, should
> >> switching
> >> > from standalone mode to multi-node mode work?  Has anyone done this
> >> before?
> >> > Preliminary tests seem to work, but I haven't looked into all the ra=
ce
> >> > conditions and such yet.
> >> >
> >> > ~Jared
> >> >
> >> > On Thu, May 5, 2011 at 12:02 AM, Alexander Shraer <
> shralex@yahoo-inc.com
> >> >wrote:
> >> >
> >> >> Hi Jared,
> >> >>
> >> >> Currently there is no support for adding and removing zookeeper nod=
es
> >> >> dynamically. See:
> >> >> https://issues.apache.org/jira/browse/ZOOKEEPER-107
> >> >>
> >> >> We're currently working to add this feature. However, AFAIK there i=
s
> no
> >> >> intention to support
> >> >> transformation between standalone and multi-node modes, only
> membership
> >> >> changes in multi-node mode.
> >> >>
> >> >> Regards,
> >> >> Alex
> >> >>
> >> >>
> >> >>> -----Original Message-----
> >> >>> From: Jared Cantwell [mailto:jared.cantwell@gmail.com]
> >> >>> Sent: Wednesday, May 04, 2011 7:17 PM
> >> >>> To: zookeeper-user@hadoop.apache.org
> >> >>> Subject: Growing a cluster
> >> >>>
> >> >>> Hello,
> >> >>>
> >> >>> Say I was going to grow a cluster from 1 node to 3 nodes.  Is this
> >> >>> possible,
> >> >>> and what would be the recommended way?
> >> >>>
> >> >>> At first I was thinking that I could go from 1 to 2 and then 2 to =
3,
> >> >>> and it
> >> >>> seems to be working actually.  But I'm not sure if this is
> supported,
> >> >>> mostly
> >> >>> because in standalone mode the on-disk files are different than th=
ey
> >> >>> are in
> >> >>> a multi-node configurations.  Mutli-node configurations embed the
> >> >>> quorum
> >> >>> incarnation into the filename, which standalone does not.  Should =
a
> >> >>> quorum
> >> >>> node be able to startup using snapshots and logs that a standalone
> node
> >> >>> wrote out?  Is there a way around this?
> >> >>>
> >> >>> Thanks,
> >> >>> Jared
> >> >>
> >>
> >>
> >
>

--bcaec501647350e26604a28a6467--