OK thanks for the suggestion. I think we’ll do the following:
- use a version identifier exposed through the device inventory to separate devices and plan deployments properly. On this subject, this would be easier to track with “smart” device groups.
- use a ArtifactInstall_Enter script as failsafe to block erroneous (e.g. artifact which has RPi bootfiles upgrade disabled but actually requires it) or uselessly dangerous (e.g. artifacts that migrate the Mender conf files when that’s already been done)
An idea for later occurrences of such migrations, which are bound to happen : let the mender client handle the migration automatically, and instead provide an opt-out flag for users who want more control. I cannot speak for all users but I reckon this would offer a better tradeoff for a lot of us.