An endpoint that used to take 80 milliseconds now takes four seconds in production. How would you find the cause?

Measure before guessing. Establish whether it is slow for everyone or one account, then read a trace to see where the time actually sits: application code, database, or an outbound call. Most often it is the database, so pull the slow query log and run EXPLAIN ANALYZE on the suspect query. In parallel, check what changed, since a deploy, a data volume threshold, or a cache that stopped hitting explains most regressions.

First I want the shape of the problem. Is it every request or the ninety ninth percentile, one tenant or all of them, and when exactly did it start. That timestamp usually solves half the case on its own, because it lines up with a deploy, a migration, or a config change. Then I pull a trace for a slow request and look at the waterfall, since guessing which layer is slow wastes the most time. If it is database, I check the slow query log and run EXPLAIN ANALYZE, and the common answers are a plan flip after the table grew, a missing index on a newly filtered column, or an N plus one that only hurts now that accounts have more rows. If it is an outbound call, I check whether that dependency is degraded and whether we have a timeout at all, because a missing timeout turns their bad day into ours. Meanwhile, if users are hurting, I mitigate first: roll back, add a short cache, or shed the expensive path, then finish the diagnosis without the pressure.