swarmsible
swarmsible copied to clipboard
Improve upgrade mechanisms to keep service as healthy as possible
Currently we only wait until the node is drained. We should investigate whether it is feasible to wait for all stacks to finish being moved over. Wait for all services to stop scheduling new things during cluster upgrade?
Maybe we need to take a snapshot of all services and the replica counts before the upgrade and we then wait until the same replica counts are back?