Clustering the SP

[Cantor, Scott]

On 1/21/14, 7:54 AM, "Martin Haase" <Martin.Haase at DAASI.de> wrote:
>
>I'd like your thoughts regarding SP clustering.

I'd start by questioning how you cluster the application itself.

>* A) ODBC/MySQL clustering. Fails under high load:
>https://issues.shibboleth.net/jira/browse/SSPCPP-602

Have you reported the bug to Ubuntu? Presumably with a stack trace they
should be able to look into it. I also noted that you should verify you're
not using prefork.

>* C) Shared shibd. Problem: single point of failure. What about
>inserting a proxy before the TCPListener, did anybody implement a
>failover shibd with active-passive, and how?

memcache is also a single point of failure, but the major problem with
that is performance. I know for a fact it doesn't scale, nor was it meant
to.

>* D) Our SLO is front-channel. What about forgetting the SP session and
>relying on session stickyness by the LB?

Most applications don't cluster well, and require sticky sessions. In such
cases, it doesn't matter much if the SP also does. But I would say that
your main issue is a broken ODBC driver.

-- Scott