FeedBack on MySQL replication based solution asked

[Arnoud Vermeer]

Dear users,

My name is Arnoud Vermeer, and I've been running a MogileFS cluster for some
time now. The only problem that I found was that MySQL was a single point of
failure. In the mailing list often is referred to MySQL Cluster, but I've
noticed that that solution just isn't that perfect. This is one of the
reasons I've decided to work on a solution that removes the single point of
failure. This post is not exactly on its place on this mailing list, but
since it is of interest to most ' heavy mogilefs users ', I decided to post
it.

The current setup is a tier 2 setup. The application and the database. My
idea is to update it to a tier 3 setup: The application, the query director
and the database. This all is MySQL Replication based, but the query
director will do some vital task such as provide high availability. The main
idea is to ' load balance ' the queries via an active-backup query director
setup. This open-source program will balance all SELECT queries to the
SLAVES (in combination with memcached) of the mysql replication park, and
all the WRITES (UPDATE, DELETE, INSERT) to the master.

A detailed version of the plan can be read on
http://funzoneq.livejournal.com/19503.html. I've also included graphics to
explain the situation.

I would like to receive some feedback on my plans so I can incorporate them
into the final design, before I start programming.

I hope to hear from you soon.

Yours sincerely,

Arnoud Vermeer
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.danga.com/pipermail/mogilefs/attachments/20060803/e4ae5e29/attachment.html