simple manual fail over for IDP

[Peter Schober]

* jehan procaccia <jehan.procaccia at it-sudparis.eu> [2012-04-30 15:36]:
> I did read 
> https://wiki.shibboleth.net/confluence/display/SHIB2/IdPClusterIntro
> and before getting into deep high availability or load balancing
> I just want to enable a single entry point to my shibboleth SSO service 
> so that
> I can  shutdown one IDP server for maintenance/upgrade and manually 
> promote the other one to serve requests and vice versa.
> Example:
>                sso.domain.fr
>                          |
>                      /      \
>                     /        \
>              IDP1        IDP2
> 
> most of the time , sso.domain.fr redirects to IDP1, when a maintenance 
> occurs on IDP1, sso.domain.fr redirects to IDP2.
> does the IDP[1,2] must have the same metadata, share a common entityIDs 
> and bindings Locations ?
> Is this simple scenary already documented somewhere ?

Well, unless there's real cluster there's nothing to document as far
as the IdP is concerned: Both IdPs should be configured exactly
identical.
As Scott said, changing DNS alone won't do. But you could move the IP
address (enabling/disabling an alias on an ethernet device, for
example) or use any of the software-only open source load
balancers/content switches. Or use a hardware loadbalancer and point
it to only one of the IdPs (with failover to the second and possibly
back to the first if #2 fails).

Of course all "SSO" sessions users have with the IdP will be lost this
way, which may be acceptable in some situations. You can also leave
the servlet container running on #1 (the IdP to be replaced) so that
if anything goes wrong with #2 you can quickly switch back to #1 and
all sessions will still be there (except for those unlucky few who
accessed #2 until you found out it's not working as expected),
-peter