links: site | repo | project page
ApiMocker’s website and health endpoint were returning 200, but the public dataset wasn’t what it was supposed to be. Records were missing and the collection counts had drifted. The service was running. The daily reset wasn’t.
The production configuration had the reset scheduler set to external.
That disables the scheduled job inside the API, which is fine if something
else actually runs the reset. Nothing did. There was no external scheduler
configured, and the automatic reset had been missing for roughly two months.
I switched back to the existing in-process scheduler. For the current single-replica API, that was the smallest fix. I also ran a reset to restore the shared mock dataset, replacing the temporary changes people had made through the public API.
Afterward, I checked the expected totals: 10 users and 100 posts, plus 200 todos and 500 comments. I checked individual records and relationship routes, then verified the search filter that had exposed part of the problem.
The important check came after the next midnight UTC. The counts were still correct, and fresh seed timestamps confirmed the scheduled reset had run without me triggering it.
A green health check told me the process was alive. It didn’t tell me whether someone following a tutorial could still fetch the records they expected. That’s the part I needed to verify.