Nothing is shared between the schedulers, and updates are eventually consistent. Schedulers would subscribe to events for hosts, and the hosts would notify the schedulers on changes. For example, it may simply be on creation and deletion so it just knows what clusters/guests/hosts exist and how to route requests to them. We'll most likely want to store this all in memory, but we can put a simple API in front of it so we could switch in the future. Redis may even be a good choice since it already drops snapshots.
-Eric On Wed, Aug 04, 2010 at 03:19:22PM -0400, Michael Gundlach wrote: > On Wed, Aug 4, 2010 at 2:57 PM, Paul Voccio <[email protected]> > wrote: > > Shouldn*t the layer be pluggable if I wanted a different datastore? > > FYI in case you missed the thread on WSGI, modules, and services: whether > the memory were local or in a memcache or otherwise, the layer would be > pluggable. We'll provide a caching library that is a language binding to > pluggable cache backends -- mock, local, memcache, or otherwise. Auth* > will have the same structure, allowing mock, LDAP, etc. > So while I don't think the cache should be in-memory as Eric has drawn it, > I think pluggability is still assumed in the diagram. > Michael _______________________________________________ Mailing list: https://launchpad.net/~nova Post to : [email protected] Unsubscribe : https://launchpad.net/~nova More help : https://help.launchpad.net/ListHelp
