이 페이지는 아직 한국어로 제공되지 않으며 번역 작업 중입니다. 번역에 관한 질문이나 의견이 있으시면 언제든지 저희에게 연락해 주십시오.
A frequent issue or pain point can be alert fatigue, or when alerts ‘flap’ (rapidly switching from an ‘ok’ to an ‘alert’ status).
There is functionality within Datadog that often leads to less noisy, more meaningful alerts.
Re-Evaluate the Alert Threshold Value
- The easiest way to reduce flapping when the alert <-> ok or state changes are frequent could be to increase/decrease the threshold condition.
Use the min threshold
- This triggers the alert only when all data points for the metric in the timeframe violate the threshold
Reframe the query using Functions- rates, moving averages, or time-shift differentials
- This means, you can compare the difference between a metric stream’s values with the values from a week ago and set alert conditions based off the difference
- A time-shift differential allows you to combine functions and can give a historical view as well. For example:
abs(system.cpu.system{*} - week_before(system.cpu.system{*})) - If your metric frequently spikes, and those spikes are not inherently indicative of issues, applying a rate or average to it will allow you to set a more meaningful threshold.
Consider the states of other monitors using Composite alerts
- Composite Monitor allows you to combine two or more previously created alerts.
For example: if CPU is high AND disk is high on a host, trigger the alert.
Use some built-in analysis modules with Anomaly or Outlier
- Anomaly Detection uses some seasonality analysis to issue an alert when a data stream behaves in a historically inconsistent way.
- Outlier Detection uses other data streams of the same context to issue an alert when a stream behaves in a way different compared with its peers
- Both can also be used in conjunction with Composite alerts.
If the issue is alert routing, template variables and the separation of warning or alert states with conditional variables will be of interest!
