using dns in conjunction with memcache

[Jed Reynolds]

I was discussing using round-robin mysql replicas today and how an 
internal DNS server is often used to manage the "spreading" of database 
connections across a sizable replication array. It occurred to me that 
the management of memcache services might be handled similarly.

I don't know if this has been discussed previously, but it occured to me 
that a service similar to DNS could be appropriate for managing keys 
across a lot of servers. First, it offers a nice separation of concerns 
between the management of your applications and your caches. Second, it 
could be a process that manages redundancy or migration of keys across 
caches.

Two things that I'm certainly considering are how to update my "bucket 
map" so that my memcache clients don't map to rebooting or missing 
servers. Right now I've got a hand-coded array of ip addresses in my 
clients, but with only two or three memcache servers at the moment, I'm 
not to concerned. However, come the day when I upgrade my single web 
server to a cluster, I don't want to have to edit the client code on 
multiple machines to reflect transitions in my memcache pool.

I'm sure someone's already figured this out, no? And wouldn't it also be 
super handy if this were something I could incorporate into a LVS 
load-director?

Jed