I really like the idea because, "at scale", database migrations tend to get bifurcated in to two parts: 1) schema growth, deprecation, and pruning and 2) live data migrations. The former seems like a really good fit for Pulumi. The one part that seems tricky is that there may need to be a more fine-grained notion of "protect: true". That is, updating a table resource that adds columns should be safe, but removing columns should require they go through a deprecation phase.