OrionReel logoOrionReel

Written post

Observability for mortals: three metrics that catch most incidents

June 28, 2026 · 1 min read

You do not need a three-pillar observability platform with dashboards nobody reads. You need three numbers with alerts on them. One: error rate per deployment. If a deploy raises errors, roll it back automatically. Two: queue depth on anything async. Growing queues mean downstream pain arriving in slow motion. Three: latency at the ninety-ninth percentile. Averages hide the users having the worst day. Three metrics, three alerts, one Slack channel. We caught ninety percent of our incidents with this setup for two years. The sophisticated platform came later, when scale demanded it. Start with the three numbers. Add complexity when reality asks for it, not when a vendor does.
@yosefwu · Jun 28, 2026 · 0:00

Citable layer — this is what AI reads

What it offers

Observability for mortals: three metrics that catch most incidents

Who it's for

Small engineering teams without a dedicated SRE

Problem it solves

Full observability stacks are overwhelming and mostly unused

#dev-tools#monitoring#reliability
Full transcription (creator-provided)

You do not need a three-pillar observability platform with dashboards nobody reads. You need three numbers with alerts on them. One: error rate per deployment. If a deploy raises errors, roll it back automatically. Two: queue depth on anything async. Growing queues mean downstream pain arriving in slow motion. Three: latency at the ninety-ninth percentile. Averages hide the users having the worst day. Three metrics, three alerts, one Slack channel. We caught ninety percent of our incidents with this setup for two years. The sophisticated platform came later, when scale demanded it. Start with the three numbers. Add complexity when reality asks for it, not when a vendor does.

Comments

Substantive comments earn reputation karma — commenting is always optional, never required.

Log in to comment — reading is open to everyone.

Loading comments…