What is this feature?
This PR introduces a new alert rule configuration option, keep_firing_for (Prometheus documentation).
keep_firing_for prevents alerts from resolving immediately after the alert condition returns to normal. Instead, they transition into a "Recovering" state and are not considered resolved by the Alertmanager. Once the recovery period ends (or after the next evaluation if it is bigger than keep_firing_for), the alert transitions to "Normal" if it doesn't start alerting again:
Before
+----------+ +----------+
| Alerting |---->| Normal |
+----------+ +----------+
-----
After
+----------+ +------------+ +----------+
| Alerting |----->| Recovering |---->| Normal |
+----------+ +------------+ +----------+
Why do we need this feature?
This feature prevents flapping alerts by adding a recovery period. This helps avoid false resolutions caused by brief alert
