Mailing-List: contact zookeeper-user-help@hadoop.apache.org; run by ezmlm
Precedence: bulk
Reply-To: zookeeper-user@hadoop.apache.org
Received-SPF: neutral (athena.apache.org: local policy)
DomainKey-Signature: a=rsa-sha1; s=serpent; d=yahoo-inc.com; c=nofws; q=dns;
	h=received:user-agent:date:subject:from:to:message-id:
	thread-topic:thread-index:in-reply-to:mime-version:content-type:
	content-transfer-encoding:return-path:x-originalarrivaltime;
	b=j9DTPEgTGbKZFbfaXNOsqEgWlKQfgkkQyCLtFp5/Q0NQa10SXBFsnlhqD1PhV4ev
User-Agent: Microsoft-Entourage/12.24.0.100205
Date: Thu, 29 Apr 2010 13:24:45 -0700
Subject: Re: Partially partitioned connectivity
From: Mahadev Konar <mahadev@yahoo-inc.com>
To: <zookeeper-user@hadoop.apache.org>
Message-ID: <C7FF391D.34CBB%mahadev@yahoo-inc.com>
Thread-Topic: Partially partitioned connectivity
Thread-Index: Acrn2gGl9VLVw9A/606x0EEcTI7IxA==
In-Reply-To: <20100429112205.7ca282eb@kwebb.ucsd.edu>
Mime-version: 1.0
Content-type: text/plain;
	charset="US-ASCII"
Content-transfer-encoding: 7bit

Hi Kevin,
  I had the response set up but didn't hit send. Ted already answered your
question, but to give you a more technical background assuming that you know
a little bit more about transaction ids in ZooKeeper and server ids:

If B and C are partitioned from each other, and A is the leader, there would
not be a problem since A can talk to B and C and the cluster would continue
functioning.

But if B/C is the Leader then the other would get disconnected and stop
functioning unless the partition heals. The clients connected to this server
will get disconnected with connected to the working quorum of A and B/C.


If a quorum election happens in this situation, there are two possibilities

1) A is the leader
2) A is not the leader

If A is not Leader either B or C is the leader and the quorum functions with
the disconnected server not functioning.

If A is the leader then since it can talk to both B and C, all three servers
will function and will work all fine.

Hope that answers your question.

Thanks
mahadev

In the first case A will not be the lea
On 4/29/10 11:22 AM, "Kevin Webb" <kcwebb@cs.ucsd.edu> wrote:

> Suppose I have three zookeeper servers (A, B, and C).  A can
> communicate with both B and C, but B and C are partitioned from one
> another.
> 
> Is the system behavior under such conditions documented anywhere?  If
> not, can someone explain what will happen at the servers and their
> clients?
> 
> Thanks!
> 
> -Kevin