Description
Confmon does not properly handle updates to memcached reliably. Because the configuration is now pull-based rather than push-based, we don't actually directly refresh if there is a cluster topology change. The solution is that upon detection of a memcached bucket, perform the following:
(1) Disable CCCP
(2) Bump the http 'disconn timeout' to a very high number
(3) Never have the HTTP provider call 'failed' but rather always cycle the nodes.