Documentation of the database platform behind the services you use, including how it fails and how it is protected.

What this covers

Almost every Think service keeps its data in a database, and until now nothing described how that layer actually works. It is now documented end to end: how a database is created, how it survives a server failure, how it is upgraded, and — the part that matters most — exactly what can and cannot be recovered when something goes wrong.

Why it is worth writing down

Several of the incidents behind this documentation were invisible while they were happening. Everything reported healthy while a service quietly ran on the wrong database for eleven days, and while two others filled their disks. The documentation records each of those honestly, including the ones still open, so the next person to look does not have to rediscover them.

It also states plainly where protection stops. Recovery today goes back to the most recent snapshot, not to an arbitrary moment — which is fine for hardware failure and weak against a mistake.