[rabbitmq-discuss] Bitcask Message Store Index

[Matthew Sackman]

Hi Alvaro,

On Thu, Oct 13, 2011 at 01:18:40PM +0200, Alvaro Videla wrote:
> On Oct 13, 2011, at 10:46 AM, Matthias Radestock wrote:
> > I doubt bitcask would perform any better than the default msg_store,
> > since the latter has been highly tuned for queue-like access
> > patterns. One area where bitcask might excel is recovery time,
> > particularly after an unclean shutdown / crash.
> 
> I totally agree regarding the implementation of the msg_store. I think
> that putting another database(MySQL, Redis, Riak) behind a queue could
> bring a severe performance problem. Having said that what do you think
> of putting a layer on top of the rabbit_msg_store that could add
> replication/clustering a la riak_core? So the idea here is to use the
> same storage backend that the variable queue uses but coordinating
> read/writes using riak_core. (Keep in mind this is not riak_kv the
> database known as Riak).

Well, you could do that. But then you'd a) pick up only the messages
which are sent to disk anyway, and b) whilst you'd pick up the messages
themselves, you'd have no idea which queues they belong to or what order
they appear in those queues.

What properties are you after which are not provided by Rabbit's new HA
features that appeared in 2.6.0? - http://www.rabbitmq.com/ha.html

Matthew