Rob Sanheim wrote:
There is also the case where you have federated data, so you'd have:
db1: users 1 - 100k
db2: users 101k - 200k
db3: users 201k - 300k
hmm federated data will need special treatment if cross-node relations are allowed, i may be naive but it seems that things can get tricky soon.
I do absolutely agree though with what you say below, that is should be at least possible for the developers to address a specific connection.
I don't think you can possibly handle all the different cluster setups
out there. Just as long as the developer can manually pick the
connection in the code if need be, it will be clean for those with
simple setups (rw to master, read-only slaves), and a uglier but
manageable for the massive, extremely complicated setups.
yes it should be trivial for simple setups and doable for data-partitioned clusters