Updated our slave database servers to 10.3 a couple of weeks ago... we roll out phases since we've got so many damn servers. It's always a nightmare. If no issues for the next couple of weeks or additional patches, we'll roll out to primary database servers (they're all sets of master-master, the slaves are for search primarily, but also are useful for a live test if everything passes staging, and they're a hot backup too if any issues arise) and then web servers. We never go down so it's a painfully slow process.