The slave databases are supposed to be locked via configuration to be read-only, so that it's not possible to write to them except via replication. I've recently discovered that some of them aren't, and this can cause problems with applications that misbehave and write to the wrong DB handle, causing slaves to be out-of-sync, and we'd never know there was a problem because no error would be generated until some day maybe a write to the master hits a duplicate key or something.
I think this is on for tonight - start time, downtime estimate?
7pm pacific, no downtime expected, it's a runtime change. Just announcing it so people know to file bugs if they notice anything funny after it's been done.