You can set a throttle_period long enough to ensure that the anomalies are resolved.
If the current anomaly (i.e. what makes the condition come true) is resolved within the throttle period:
- the next watch execution won't have it's
conditionmet so... - no
actionswill be executed and - the alert will be "rearmed", ready for the next anomaly.
A similar question has been asked here recently, you can get more information there: How to stop sending duplicate Slack notifications for the same error?