Design Doc · SRE Team · In Review
Adopt SLOs and an Error Budget for the Public API
Press Enter to flag this paragraph.
Problem
The public API has 380 threshold alerts and paged on-call 140 times last month. A review of last quarter's pages found that 71% needed no action. Availability over the last 90 days was 99.93%.
Indicators
Availability is the share of requests that return a non-5xx response, measured from the API servers' access logs.
Latency is measured as average response time, with a target under 300 ms.
Targets
Every endpoint gets the same target: 99.99% availability over a rolling 30 days.
Alerting
On-call is paged whenever the error rate over the last five minutes exceeds 0.01%.
All 380 threshold alerts are deleted the day the SLO alerts go live.
Error Budget Policy
When the error budget is exhausted, all deploys to the API stop until the budget recovers.
The SRE team enforces the policy.
Reporting
SLO status is published on a Grafana dashboard.