Clustering the IDP using Terracotta?

[Ryan Larscheidt]

The memcached storage service plugin in the contributions (https://wiki.shibboleth.net/confluence/display/SHIB2/Memcached+StorageService) is working well for us, after a few tweaks (changing timeout to 100ms and not running the filter on the CSS/images/JS paths).

It keeps a copy of the session data in the memcached cluster AND the IdP's internal store, so it can survive memcached node deaths without issues (I tested with partial and complete cluster failure).  A complete memcached cluster failure doesn't cause an outage; the IdPs just behave as though you don't have clustering at all.  Also, each session that is "touched" after a memcached cluster recovery is copied from the IdP internal store into the memcached cluster.

So you get a lot of the clustering benefits without any of the "single point of failure" downsides.

Thanks,
Ryan

On Jan 24, 2013, at 10:29 , Peter Schober <peter.schober at univie.ac.at> wrote:

> * David Langenberg <davel at uchicago.edu> [2013-01-24 16:49]:
>> We're seeing good results using the Memcache storage plug-in to achieve
>> clustering in a terracotta free way.  Maybe give that a shot?
> 
> How does using memcache provide "clustering"?
> Unless you add a replication/redundancy layer on top of memcache all
> you do is move the single point of failure to a shared memcached?
> -peter
> --
> To unsubscribe from this list send an email to users-unsubscribe at shibboleth.net

-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 7327 bytes
Desc: not available
Url : http://shibboleth.net/pipermail/users/attachments/20130124/ab0f04c2/attachment.bin