5 alerts that wake you up for nothing, and what to page on instead

Checklist: what to do about it

• CPU above 80% → page on what users feel: latency and errors, not a busy server.
• Disk 70% full → page on time-to-full. If it fills in under a day, wake someone; if not, make a ticket.
• One failed health check → one blip is noise. Page on N failures in a row.
• Every single 5xx → page on the error rate against your SLO (error budget burn), not single errors.
• Anything nobody acts on → if the page ends in "ignore", delete it or turn it into a ticket.
• Review last month's pages and tag each one: acted on, or ignored.
• Every remaining page should link to a runbook step.