memcached failover

[Jacob Lundberg]

Hi all,

On Wed, 2016-08-03 at 13:29 +0200, Peter Schober wrote:
> Jfyi, note that SimpleSAMLphp implemements session distribution and
> replication "among several memcache servers

What happens if (for example in a two-node cluster, but it generalizes)
one server is unreachable for a while and then the other server is
unreachable for a while (a common maintenance scenario)?  The problem
with client-based solutions to this scenario in memcached clusters is
old data can be revived in the cluster.  I'm not certain what the
implications are for an IdP; perhaps reversion of data is no big deal.
Probably people might find that SSO breaks and they must sign in a
second time.  Possibly logouts could be reverted as well?  Somebody more
familiar with the data model than I am would have to answer that.

However, this is what internal replication products like repcached or
Couchbase are intended to solve.  If you use Couchbase, you can present
it to each IdP node as if it were memcached by running a moxi proxy on
the IdP node.

-Jacob

-- 

Jacob Lundberg
Director, IT Architecture
jacob at collegenet.com
503.290.0100 (voice)
503.973.5252 (fax)
503.901.8343 (cell)