DEV Community

Hive80-lab
Hive80-lab

Posted on

7 Monitoring Rules That Pay for Themselves in One Incident

Monitoring bills creep because teams monitor everything. Senior teams monitor the few things that page correctly and drop the rest. Here are 7 alert rules that earn their keep — each one mapped to a real incident class.

1. The Checkout Canary (synthetic, not CPU)

A synthetic transaction that logs in and completes checkout every 60s. CPU at 90% with checkout fine = green. Checkout failing with CPU at 20% = page. You get paid when you catch revenue-level failure before a customer email does.

2. The Silent Death Detector

Services that die and stay dead cost more than flappy ones (nobody notices). Alert on: up == 0 for 2 minutes plus a rate-of-change guard so restart storms don't mask a true outage.

3. Error Budget Burn Rate

Not "5xx > 1%" — that fires all the time. Alert when you burn 2% of your monthly error budget in 1 hour. Google's SRE playbook calls this multi-window burn-rate alerting; it cuts noisy pages by an order of magnitude.

4. The Queue That Never Drains

Any queue (jobs, emails, webhooks): alert if oldest message age > 10 minutes. Queue depth is a lagging indicator; age is the leading one.

5. Disk Space With a Headroom Rule

90% full = warning, 95% = page. But the rule that actually saves you: alert if disk grew >5% in the last hour — that's a log fire, and you'll be full by morning otherwise.

6. The Expired Certificate Alarm

Still the #1 self-inflicted outage. Alert 30 days out. One line of config; one embarrassing incident avoided.

7. The Restore-Path Tripwire

Once a week, a cron job restores your latest backup into staging and runs a checksum. If the restore fails, you get an alert — not a surprise during a real incident. A backup you've never restored is Schrödinger's backup.

The Meta-Rule

For every alert that exists, someone must be able to answer: what do I do when this fires at 3AM? If the answer is "read the source code", delete the alert and write a runbook instead.


If you want these 7 rules as ready-to-paste configs (Prometheus/UptimeRobot/friendly cron variants) plus the runbook and incident-checklist templates, we packaged them as the Ops Starter Kit.

Which of the 7 are you missing today? Say it in the comments — I'll reply with the cheapest way to add it.

Top comments (0)