Hi all,
I have an external web service that I need to log into at a specific
part of my web app,. The problem is that I only have one login to
this web service, and if another login attempt is made while an
existing login session is still active, the existing session gets
terminated.
How can I best serialise access across my mongrel cluster to this
external API?
I thought of creating an external drb service that all cluster nodes
connect to, but that could also be re-introducing a single point of
failure that I eliminated by using a cluster.
I could do something with db transactions and a “locks” table… any
other ideas?
Thanks
Jon
On Mon, 2006-06-26 at 14:25 +0100, Jon Evans wrote:
I thought of creating an external drb service that all cluster nodes
connect to, but that could also be re-introducing a single point of
failure that I eliminated by using a cluster.
A single point of failure that you know about and can protect is better
than a lot of brittle locations that always fail. Many times an SPF can
be a good thing since it funnels potential trouble to one location that
is easy to manage.
But, you can also use tools to eliminate the SPF. You can typically
mirror the protocol or state between two servers or balance between them
(much simpler) so that when one goes down the other picks up.
So, I would say do the DRb server, and then scale it up as you need.
This is much easier than trying to cram some sophisticated locking
system into all of the Mongrel servers, which adds a larger potential
failure to all of the processes.
Also, DRb is cake, you could probably have this done in an hour to a
day. With all your free time you could then learn about Rinda or how to
do TCP balancing with pen or balance.
–
Zed A. Shaw
http://mongrel.rubyforge.org/
Hi Zed,
On 27 Jun 2006, at 00:35, Zed S. wrote:
So, I would say do the DRb server, and then scale it up as you need.
This is much easier than trying to cram some sophisticated locking
system into all of the Mongrel servers, which adds a larger potential
failure to all of the processes.
Also, DRb is cake, you could probably have this done in an hour to a
day. With all your free time you could then learn about Rinda or
how to
do TCP balancing with pen or balance.
Very useful advice, thanks very much. I hadn’t even heard of Rinda,
pen or balance, but I’ve bookmarked their respective sites to look at
after I’ve thrown the DRb server together. You’re right, it shouldn’t
take very long at all - I already have all the code written inside a
class in my rails lib directory, just need to wrap it in a bit of DRb
glue.
Jon