The dashboard is clean and easy to navigate
The branching feature is really good idea, it allows you to confidently run migrations with no downtime
The daily backups are also great, no additional steps to set them up
They also have a graph to check query latency which is nice
The only downside is that they dont have foreign keys but I used prisma referential integrity for that so not really an issue for me
So this branching feature is one I still don't understand properly.
Let's imagine my startup is big enough to warrant a PlanetScale database. Hundreds of rows are written every second. When I branch, are those writes applied to the original and the branch? If so, what happens with migrations that mean the incoming writes no longer work? Or is the "confidence" it gives not about full migrations but about knowing that if your migration fails the original DB is still working?
When you merge a branch in PlanetScale, you can still write to the database while the migration is happening, online. This is done through "Online Schema Change", but the TL;DR is that a system in the background creates the new database, reads the mysql binlogs from your original database, including all the writes, and propagates those writes to the new database with the changes applied. It does this continuously, in real time. Once the new database is caught up, the system switches over transparently to make the changes atomically appear live.[1]
The requirement is that the running application server has to be able to handle both the old and new schema at the same time. This is easy if you do something like "add a column that isn't yet used and will be soon", but harder for "dropping a column that is in use". No database alone can solve this however, you have to coordinate your application layer correctly so that it can handle the switch over when you apply the changes. But this means you can have very high uptime with e.g. blue/green deployments and staged rollouts, and you'll never have to take your DB offline for it.
It's also trite but "hundreds of writes a second" is nothing. Vitess can handle 100x that write load (probably much more) and deal with schema changes to petabyte-scale databases completely online. PlanetScale currently doesn't let you control the Vitess keyspace sharding configuration (yet) which is crucial to this but I suspect it will happen eventually. But they'll also let you run Vitess in your own K8S environment while they provide a control plane, so you wouldn't pay list price, anyway.
I do not have any affiliation with PlanetScale but it's a pretty cool product I could see myself using for my low-volume needs, just because of the pay-as-you-go model where I don't pay for e.g. RDS allocation. And I say that as a really big PostgreSQL fan.