Roll back ramp-ups from Datadog alerts
When a Datadog monitor alerts, a webhook can roll back or pause the GrowthBook ramp-ups that match filters you choose, such as a feature tag. You set the webhook up once, and it also covers ramp-ups that start after you create it.
How matching works
The REST API has two bulk actions for ramp schedules. Each one finds ramp-ups by the features they run on:
/api/v1/ramp-schedules/actions/rollbackrolls back running or paused ramp-ups that match. It needs at least one filter./api/v1/ramp-schedules/actions/pausestops further progression and keeps current exposure. With no filters, it pauses every running ramp-up.- The filters are
environment,project,tag,feature, andcustomFields[<key>]. Each takes comma-separated values, and different filters must all match.
The per-schedule actions need a schedule ID, so a webhook built on them has to change with each release. A filter such as tag=checkout keeps working as you tag new features.
What to expect
- A rollback restores the rule's saved base state, which isn't necessarily zero exposure.
- Completed ramp-ups aren't matched, so the webhook can't undo a release after its ramp finishes.
- The response lists each match as
appliedorskippedwith a reason, such as an API key that can't publish that feature. A successful response can still contain skipped entries. - Nothing resumes when the monitor recovers. After the incident, someone resumes or restarts the schedule.
Getting started
Ramp-ups need a paid plan that includes Ramp Schedules. Create a Secret Key under Settings → API Keys with a role that can publish features in the affected environments. Use a key that doesn't have Require confirmation turned on. In Datadog, create a webhook that points to the rollback action with your filters. Send Authorization: Bearer <key> as a custom header, and include a reason in its payload. Then mention the webhook inside {{#is_alert}} in each monitor that should trigger it. Datadog's Test Notifications sends a real request, so try it on a disposable ramp-up in staging first. The Datadog integration guide covers the full setup, plus a flag-toggle option for features without a ramp-up.
