Shared databases are the most comfortable bad decision in software. They're fast on day one, invisible on day ninety, and lethal on the Sunday your disk fills up.
This one had six writers. Two of them were cron jobs nobody had documented. One was a legacy PHP admin panel that the ops team believed had been decommissioned in 2023.
An unowned system doesn't fail slowly. It fails all at once, on the day you're least available.
The immediate cause was an audit table with no retention policy. The real cause was that no single team could tell you what the database was for, so no single team felt responsible for its health.
If you can't name the owner, you don't have a system. You have a liability.
We bought time with a partition drop, then spent six weeks assigning owners — not migrating, just naming. Ownership alone eliminated four of the six writers, because two of them turned out to be nobody's requirement at all.
