Prompt
Grafana alerting is confusing
Latest observation
Nov 5, 2026
OpenAI APIWeb search: off
Grafana alerting can definitely be confusing. The main reason is that it mixes a few different concepts:
- Query: get the data
- Expression / reduce / math: turn data into a single value
- Condition: decide when to alert
- Rule group / evaluation interval: how often it checks
- Labels / annotations: metadata for routing and messages
A simple way to think about it:
-
Fetch data
Example: CPU usage for a server. -
Transform it into one number
Example: “latest CPU value” or “average over 5 minutes”. -
Set a threshold
Example: alert if CPU > 80%. -
Route the alert
Example: send to Slack, email, PagerDuty, etc.
Common gotchas:
- No data vs. zero: Grafana treats these differently.
- Instant vs. range queries: some alerts need a time window, not just the latest point.
- Multiple series: one alert can fire per time series unless you aggregate first.
- Pending duration / for: condition must stay true for a while before firing.
- Classic condition vs. expressions: newer alerting uses expressions more heavily.
If you want, I can explain Grafana alerting in one of these ways:
- Very simple mental model
- Step-by-step example
- How to configure an alert in the UI
- Troubleshooting why an alert won’t fire
- Differences between legacy and unified alerting
If you share what part is confusing, I can break that piece down plainly.