Database Operations

A database is the one component in most systems that cannot simply be redeployed when something goes wrong — it holds state that no other system has a copy of. That changes the math on every operational decision: a connection leak that would be a shrug in a stateless service can take down every service sharing that database, and a "quick" index build can lock a table your whole product depends on.

The job is not keeping the database running — it mostly runs itself. The job is making sure every change to it, and every failure mode around it, has already been thought through before it happens under pressure.

Treat the database as the one thing you cannot roll back to a clean slate — every operation against it should assume that.

1. Treat the connection pool as a shared, finite resource