One ANALYZE command resulted in a 100x query speedup for one person. There was no need to restart the server, reconfigure the network, or receive an alert at 3am.
That's the Julia Evans post about running SQLite in production that linked in my brain with a chat I had with Richard Schneeman five years ago and I couldn't shake it loose. It's one of those times where you read something, or have a conversation, and it feels like a missing half you didn't know you needed has finally slipped into place.
The stack we all pretend we need
Many teams opt to use managed Postgres even before experiencing load from a single user. After that, they implement a connection pooler, followed by a read replica, and then a caching layer to mask the introduced latency of the other stuff they bolted on to go fast.
This represents a full stack you would normally have to create, maintain, and monitor on your own.
