IDP Cluster going forward (JDK 7)

[Cantor, Scott]

On 5/28/13 9:13 AM, "Marc Thornton" <marc.thornton at gmail.com> wrote:

>I would be interested in having access to Paul's Infinispan plugin and/or
>any other clustering-related plugin sources.

https://wiki.shibboleth.net/confluence/display/SHIB2/Contributions

> My understanding with Terracotta was that there really wasn't any
>Terracotta-specific code in the Shibboleth IDP, therefore any work that
>may have been done to identify the integration points for an alternate
>clustering approach would be more than welcome!

There's basically one API you implement. If you want to try and implement
that for Terracotta's "next" big thing, you're certainly welcome to.
That's why it's not a migration path from DSO, it's just a totally
different approach.

My issue with clustering is that if it doesn't solve the maintenance
problem, I'm not very interested in it. By which I mean that if I can't
take down any of my active IdP nodes for maintenance with no service
interruption, I don't understand what it's buying me. I might as well use
active-passive, in which case I can avoid clustering entirely.

>From a general point of view, memcache is a reasonable substitute for
Terracotta for the time being, in that it adds a process that has to be
running or the system won't work. But that breaks the maintenance property
since I can't patch the memcache host without breaking the IdP.

The downside of the stateless approach is that you can't implement logout.
Since I believe logout doesn't work anyway, I don't see that as being much
of a tradeoff.

But V3 needs to provide both alternatives in a maintainable way.

-- Scott