[Cosmo-dev] Deployment of Cosmo persistence load distribution

[Jared Rhine]

Jared Rhine wrote:
>> ...multi-database distribution...

Brian Moseley wrote:
> not that i can think of. theoretically we could design a feature where
> cosmo has to perform a transaction across multiple databases, but i
> haven't imagined that feature yet.

Here's one feature I think we need that needs to be done across databases
either without a lock or with a read-only lock: the admin's list of users or
the /cmp/users operation.

In addition to the "given a username, tell me which database to go to"
feature that the service locator layer will need to provide, does that
service also then need to support getting a list of all databases in use?
And then Cosmo would itself iterate over all databases to get the complete
list of users in the cluster?

I was wondering if I couldn't do a MySQL view which does connections to each
database on the backend.  But I don't like relying on views much because of
the implementation specifics between databases.

-- Jared