Fixing the depooling is easy. The complicating part is deciding on how to run schema changes across the fleet.
A schema change can go in one of three ways in the secondary dc:
- Status quo: Run it on master of secondary dc with replication enabled. This can be done possibly in some cases in the future as well or when a dc in depooled
- Run on each replica and run it on master at the end. Similar to the most of schema changes we do on eqiad.
- Run on all replicas. Do a switchover of master and run it on the depooled old master. This will be needed in schema changes that will take hours. Thankfully switchover on a read-only dc should be fairly straightforward. We might even be able to delegate it to auto_schema.