Mailing-List: contact activemq-users-help@geronimo.apache.org; run by ezmlm
Precedence: bulk
Reply-To: activemq-users@geronimo.apache.org
Received-SPF: neutral (asf.osuosl.org: 207.171.160.38 is neither permitted nor
 denied by domain of jheitmann@gmail.com)
Mime-Version: 1.0 (Apple Message framework v752.2)
In-Reply-To: <5706215.post@talk.nabble.com>
References: <5704490.post@talk.nabble.com>
 <ec6e67fd0608080500u7f66c236gb2398b619341d2d4@mail.gmail.com>
 <5706215.post@talk.nabble.com>
Content-Type: text/plain; charset=US-ASCII; delsp=yes; format=flowed
Message-Id: <59219373-0553-416F-97B6-31BA845E5092@gmail.com>
Content-Transfer-Encoding: 7bit
From: John Heitmann <jheitmann@gmail.com>
Subject: Re: Network with slow link
Date: Tue, 8 Aug 2006 17:22:18 -0700
To: activemq-users@geronimo.apache.org


On Aug 8, 2006, at 6:18 AM, Anders Bengtsson wrote:

> On a related issue, should AMQ detect message loops? Some  
> documentation says
> that it does, by keeping a list of all visited brokers in each  
> message,
> while other documentation (for network TTL) warns about the risk of  
> loops
> with a high network TTL.
> We experienced a few message loops when trying out some more  
> complex broker
> network topologies.

The last I looked into this (late winter, around 4.0-M3 I think)  
there were known issues with some network topologies. True loops  
seemed to be detected ok, but some common non-loop multiple paths  
were not detected. For example if you have a producer to broker A, A  
connected to brokers B1 and B2, B1 and B2 connected only to broker C,  
and you consume from C, then your consumer would see 2 copies of each  
message on a topic.

Since we don't use master/slave and we needed multiple brokers on  
each hop for redundancy we use queues on the paths that would give us  
troubles. Queues ensure that only one path gets a message. The new  
virtual destination functionality will make this kind of setup even  
easier.

John