Error rate is the number of errors an application produces per unit of time or per unit of traffic, usually expressed as errors per hour or as the percentage of requests that failed.
The denominator decides what the number means. Errors per hour describes load on your on-call rotation. Errors per thousand requests describes quality, and it stays honest when traffic doubles overnight.
Example
errors per hour = 4,812 exceptions / 6 hours = 802/hr
errors per request = 4,812 exceptions / 1,204,000 requests = 0.40%
Same incident, two numbers. The first decides whether to page someone. The second decides whether this release is worse than the one before it.
What Rollbar does with error rates
Rollbar’s Adaptive Alerts watch exception rates rather than individual items, using anomaly detection to learn what normal looks like for each project. Rollbar trains that model on two weeks of your own history in one-hour increments and raises an alert when the current rate is statistically atypical, with a maximum delay of ten minutes. Adaptive Alerts track raw exception types by environment, independently of grouping, so a spike scattered across many new items still registers as one anomaly.
You configure two things. The minimum exception rate, in exceptions per hour, sets a floor below which Rollbar stays quiet. Sensitivity, at low, medium or high, sets how much statistical confidence is required before an alert fires. Adaptive Alerts cover exceptions rather than log messages, and are available on Advanced and Enterprise plans.
Related terms
Read the Adaptive Alerts docs.
Sources
- Rollbar Docs, Adaptive Alerts: https://docs.rollbar.com/docs/adaptive-alerts
