There are many advantages to using a vendor-managed PostgreSQL (AWS Aurora), but the black box nature of diagnostics is not one of them. For anyone else fighting with index performance going suddenly sideways (10x slower queries) on Aurora PostgreSQL - the root cause seems to be that new reader replicas can sometimes have horrible index performance, but this only shows up when there are tons of concurrent requests, and running a full `reindex database concurrently <db>` resolves it. All of the standard diagnostics (looking for invalid indexes, trying manual queries on each endpoint with explain, checking seq scan stats, etc) pointed to everything being just fine.

Unrelated, but hilarious, it seems like that somebody is uploading sketchy AWS database snapshots and marking them public. Is this just a troll, or can a backdoored snapshot do anything interesting if you load it into a sensitive environment? (via extensions, triggers, etc).