This isn't (currently) replicated
If it were, this would be a non-issue. In the future, we plan to have multiple slaves. Then we just drop the master from the load balancer and do the update. Drop one slave and kick off replication. Put that slave back in, drop and replicate the next. Repeat as necessary.
This is a mostly read-only DB. We need to do bulk insert/update operations. Currently we do the update in a DEV environment, then move the changes up to PROD.
What I'm looking at now are the issues around simply copying the files and doing a detach/attach on PROD. If that's a fast operation, with a small chance of failing, and high confidence in rolling back to the old version if it fails, that might be the way to go.
--
Drew
