The AWS documentation on Neptune describes scenarios in which a cluster in need of maintenance will be taken down for some loosely-defined interval of time, saying:
Updates are applied to all instances in a DB cluster simultaneously. An update requires a database restart on those instances, so you experience downtime ranging from 20 or 30 seconds to several minutes, after which you can resume using the DB cluster.
First, I'm wondering whether I'm reading this correctly. I find this surprising because so much of AWS is built with zero downtime in mind (distribute your cluster across multiple availability zones!) but then along comes a required update like this and you're basically guaranteed downtime.
I'm wondering how people deal with these in practice, for those who don't want to incur that downtime. I imagine that some sort of variation on blue/green deployment scenario is possible where you spin up a second cluster to take over while the first is undergoing updates, using Neptune streams to keep the two in sync up to the switchover point. And actually, after that you can toss out the old cluster because that's easier than switching back over to it. That's the only thing I can think of. Are there some alternatives that I'm missing?