Currently, when upgrading from one Dgraph version to the next, there is no way to do this without some downtime in the cluster. For high availability, it would be awesome to be able to upgrade a Dgraph cluster with zero downtime.
One possible approach was suggested by @MichelDiz :
Ideally, the zero downtime would be for both reads and writes.
I’m aware that currently tablets are blocked for writes while they are being moved between groups, but that there’s an update that’s been included in Master (which should be included in 21.07) that will mostly make writes available during tablet moves, but not entirely. This is discussed here, and there may be some thoughts discussed that are helpful to this feature request also.
Thanks for your consideration of this request, and keep up the awesome work!