Escalation & on-call

When a monitor opens an incident, an escalation policy decides who hears about it and when. A policy is a ladder of steps: page the first target now, widen after a few minutes if no one acknowledges, repeat until someone does.

​

Build a policy

1. Go to Escalation → New policy.
2. Add steps. Each step has a delay (measured from the previous step) and one or more targets.
3. A target is a channel (Slack, Discord, a webhook), a person, or an on-call schedule.

Set how the policy repeats while an incident stays unacknowledged. Point a monitor at the policy, or set one as the workspace default.

​

Add a runbook

Give a policy a runbook: a checklist or playbook for whoever it pages. Write it in the policy editor under Runbook. Markdown works. The runbook shows on the incident and on the acknowledge link, so a responder sees it as soon as they’re paged. Use it for the first things to check, who to call, and where the dashboards live.

​

Screen noisy alerts

Not every alert should wake someone. Mark a policy silent to record its incidents on your dashboard without paging anyone, then triage them by hand.

1. In the policy editor, turn on Silent.
2. Point a noisy monitor at the silent policy (or set it as the workspace default).

A silent policy has no steps and sends nothing. The incident still shows on your dashboard and status page. When one turns out to be real, open the incident and use Escalate to policy to put it on a paging policy. It pages from the first step right then, and keeps escalating until someone acknowledges.

​

Compose policies

A step can page another policy instead of copying its responders. Pick “Page another policy” on a step and everyone in the chosen policy is paged at that step. Use it for a shared group: keep one “Payments on-call” policy and reference it from every service that depends on payments. Change the group once and every policy that references it picks up the change. References resolve recursively, and a policy can’t page itself, directly or through a chain.

​

Routing a step

A step can carry conditions, so one policy handles different cases: Give two steps complementary time windows to notify a channel in business hours and page on-call out of hours.

​

Acknowledge

Acknowledging an incident stops the escalation. Acknowledge from the dashboard, or from the signed link in the alert (it opens a page; a button does the ack, so a link preview can’t acknowledge for you). A Member can’t acknowledge; that takes Responder access or higher.

​

How people are paged

Each person sets their own contact details in their account. By default an escalation reaches them on every method they have set up. They can define a personal ladder instead (“email now, text after 5 minutes”). SMS and calls go only to a phone they’ve verified; email always works, so a page is never dropped for a missing verification.

​

On-call schedules

A schedule rotates who’s on call. Configure the rotation cadence and the responders; cover a gap with a one-off override. A step that targets a schedule pages whoever is on call at that moment. If a schedule has no one on call, the step pages the workspace owners and admins. You can opt in to an email before your shift starts. It’s off by default. Holiday mode pauses your own paging while you’re away. The rest of your team is still paged.

​

Calendars

Subscribe to a schedule from Google, Apple, or Outlook through a signed iCal feed, or connect Google Calendar for a two-way mirror.