[atombender]

That sounds complicated. Presumably you could use watches to quickly get notified on catalog changes, but it suggests that any node could potentially have outdated metadata leading to weird and subtle consistency issues.

With some things like statistics, maybe not a big problem, but with things like dropping columns from tables, you could end up in a situation where a query sees the old table definition.

[fabianlindfors]

Certainly complicated! This is one of those places where I don't think Postgres compatibility is achievable or even desirable, schema changes in a distributed databases are a whole different beast (and very complicated in large Postgres deployments as well).

It will probably end up being something similar to online schema changes in Cockroach where the changes that can be made are limited for safety and run as a background job that can be tracked: https://www.cockroachlabs.com/docs/stable/online-schema-chan...

I've got a bit of experience with schema changes from building Reshape: https://github.com/fabianlindfors/reshape. I'm hoping to transfer over some concepts from there into pgfdb eventually!