The failure mode that actually gets you
Backup failures are rarely loud. The password rotated, the disk filled, the container stopped being scheduled — and nothing anywhere alerts, because the thing that would alert is the thing that stopped running. Every veteran has a story about the cron job that had been dead for eight months.
Invert it: silence = alarm
A dead-man switch expects a ping after every successful run and alerts when pings STOP:
pg_dump ... && curl -fsS https://hc-ping.com/<uuid>
Configure the expected period (daily + grace). Now the monitoring doesn't depend on your job being alive — the one dependency you can't trust.
Monitor the whole pipeline, not just the dump
Ping per stage: scheduler ran, backup stored, verification completed, pruning ran. A pipeline that takes backups but stopped verifying them is quietly worthless. Firedrill pings four independent Healthchecks for exactly these stages — the 'never trust the pipeline to report its own failure' principle, applied to ourselves.