Alert fatigue is the state a team reaches when so few of its alerts are worth acting on that engineers stop reading them, including the ones that matter. It shows up as a muted channel long before it shows up as a missed incident.
Two things cause most of it. Grouping that fingerprints on the message string turns one bug carrying a variable in its text into thousands of separate alerts. Static thresholds fire on ordinary traffic peaks and stay silent through a real anomaly at 3am.
Example
[alert] New item: Invalid session for user 84213
[alert] New item: Invalid session for user 91775
[alert] New item: Invalid session for user 10388
These three alerts, and the nine thousand behind them, all come from one bug, because the user ID is part of the message the tool fingerprints.
What Rollbar does about alert fatigue
Rollbar strips data out of message text before fingerprinting, removing timestamps, email addresses, IP addresses, decimals, integers of two digits or longer and hexadecimals of four digits or longer, so a message differing only by a user ID collapses into one item. Rollbar deliberately preserves values that look like error codes or HTTP status codes, because those distinguish genuinely different failures.
For volume rather than duplication, Rollbar’s Adaptive Alerts replace a static threshold with a baseline learned from two weeks of your own history, and let you set a minimum exception rate below which nothing fires at all. Rollbar also lets you mute an item so it stops producing notifications while staying in the project.
Related terms
See how Rollbar cuts alert noise.
Sources
- Rollbar Docs, Default Grouping Algorithm: https://docs.rollbar.com/docs/grouping-algorithm
- Rollbar Docs, Adaptive Alerts: https://docs.rollbar.com/docs/adaptive-alerts
- Rollbar Docs, Rollbar Terminology: https://docs.rollbar.com/docs/rollbar-terminology
