Interesting issue with crytpographic nameID and memcached on an SP

[Roy, Nicholas S]

All,

I thought I'd post this as it was an interesting issue that might potentially impact some of you if you decide to use the new "stateless clustering" approach on your IdP.

We put our new IdP servers into production today, configured as follows:
The whole thing is fronted by an F5 BigIP LTM, which uses cookie persistence on port 443 (SSO) and terminates SSL at the F5 so it can manipulate the cookies to support persistence.  On port 8443 (web services), we use least-load load balancing, and terminate SSL on the IdP servers.  We don't list in metadata and have disabled all relying-party support for the artifact resolution profile.  We use the "DataSealer" bean with a shared symmetric key to do the cryptographic nameID stuff.

The crypto nameID is significantly longer than the previous nameID format.  This is where the problem arises.  If an SP uses memcached, the xmltooling memcached extension seems to put the nameID value in a key, which is limited to 250 characters in length.  The xmltooling library fails to be able to persist the nameID value to memcache and you get a Shibboleth SP error page.

We worked around this issue by preventing the crypto transient ID from being released to this SP, and instead, release a different, shorter ID to the SP.

Nick

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://shibboleth.net/pipermail/users/attachments/20110805/a993a76d/attachment.html