SLOs · 02 of 04 · Three signals

Three signals, three questions

A single "is it red?" hides three different questions: is the budget spent, is it being spent right now, and can we see the SLO at all. Each is its own signal with its own threshold, its own clear level and its own notification.

Three signals, three questions Three horizontal lanes. Budget exhausted: period burn rate on a 0 to 2 times axis, healthy below 0.9, burning above 1.0, and the state held in between. Burning fast: burn rate over 5 minutes and 1 hour on a 0 to 20 times axis, burning only when both exceed 14.4, clearing when either drops below about 13, held in between. No data: fifteen consecutive 120 second checks with no burn-rate series, which fires on the fifteenth, about 30 minutes, and resets on any tick with data. BUDGET EXHAUSTEDPeriod burn rate over the whole SLO windowHEALTHYBURNING00.5×0.9×1.0×1.5×2×HOLDSBURNING FASTBurn rate over 5 minutes AND over 1 hourHEALTHYBURNING05×10×13×14.4×20×HOLDSNO DATAConsecutive checks with no burn-rate series123456789101112131415TICK 1 · EVERY 120 SECONDSTICK 15 · ABOUT 30 MIN · FIRESAny tick with data resets the count.Each signal has its own state, so one SLO can be spent, burning fast and blind at once.LEGENDHealthyHolds previous stateBurningFires

Budget exhausted

Whole-window burn rate above 1× means the window's error budget is spent. It lags: an outage that ended an hour ago can leave it red for days. It clears below 0.9×.

Burning fast

Both windows above 14.4× is 2% of a 30-day budget gone in an hour. Both must agree to fire, so a one-minute spike is ignored; either dropping below 13× clears it.

No data

A failed Prometheus query does not count as a miss, so an outage never reads as every SLO going blind. New SLOs get about 30 minutes while Sloth loads their rules.

Hangar · SLOs