[rabbitmq-discuss] Network partitions

Felipe Gutierrez Cruz fgutierrezcruz at gopivotal.com
Thu Oct 3 15:15:05 BST 2013


Thanks Simon, Alvaro,

I don't think they are using big messages at least less that 2k.
I'll start looking at the logs, to see what I can find.

Thanks again

Felipe Gutierrez
Sr. Consultant
SpringSource/GoPivotal

On Oct 3, 2013, at 3:21 AM, Simon MacMullen <simon at rabbitmq.com> wrote:

> On 03/10/2013 2:27AM, Alvaro Videla wrote:
>> While I don't know about your exact problem keep in mind that a GC pause
>> could appear as a network partition.
> 
> Err, what? Where did you get that idea?
> 
>> Also on some situation very big messages can cause network partitions
>> since Erlang nodes won't be able to ping each other.
> 
> This is sadly still true. I had some idea that this was fixed in recent Erlang versions but I just tested with R16B01 and it's not.
> 
> It's pretty hard to run into though - you need to be sending huge enough messages that each message takes a multiple of net_ticktime (i.e. by default 60 seconds) to transfer.
> 
> I was only able to replicate it by building a cluster across a poor-quality wifi link. Which is obviously not recommended.
> 
> Still, even the theoretical possibility is disquieting, so I think a future RabbitMQ release will chunk large messages across clusters.
> 
> Cheers, Simon
> 
> -- 
> Simon MacMullen
> RabbitMQ, Pivotal

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.rabbitmq.com/pipermail/rabbitmq-discuss/attachments/20131003/421ad977/attachment.htm>


More information about the rabbitmq-discuss mailing list