Replacing a $3000/mo Heroku bill with a $55/mo server

It’s all fine and dandy, but I wonder why so little discussion around this (mainly high-level comments “DBs are hard”?

> disco provides a "good enough" Postgres addon.

> This addon is a great way to quickly setup a database when Postgres is not mission critical to your system. If you need any non-basic features, like replication, automatic failover, monitoring, automatic backups and restore, etc. you should consider using a managed Postgres provider, such as Neon or Supabase.

How come automatic backups is considered an “advanced” feature?

Also I can’t think of a single application since 2012 that I have worked on that did not have a secondary/follower instance deployed. Also suggesting Neon and friends is fine, but I wonder what is your average latency, Hetzner does not have direct connection to the DCs these databases are hosted.

Backups are only advanced in the context of our Postgres being "Good" enough (maybe our built-in Posgres could be called "Barely enough" but that sounds a bit lame) :-)

I fully agree with you though, it's table stakes (unintended pun!) for any prod deployment, just as read-only followers, etc. Our biggest, most important point, is that folks should be using real dbs hosted by people who know what they're doing. The risk/reward ratio is out of whack in terms of doing it yourself.

And finally, re Hetzner and cross-DC latency, that's unfortunately a very good issue that we had to plan for in another case - specifically, a customer using Supabase (which is AWS-based). The solution was to simply use an EC2 machine in the same region. Thankfully, some db providers end up being explicit about which AWS region they run in - and obviously, using AWS RDS is also an option! It's definitely a consideration.