Docs / Alerts and incidentsBrowse

Incidents

An incident is one outage of one site, from the first failed check to the first good one. It collects what we saw, who was alerted and what your team did.

On this page
  1. How an incident opens and closes
  2. What we saw
  3. Timeline and comments
  4. Escalate now
  5. Resolve by hand
  6. Tell your users, with an AI draft
  7. Write the postmortem, with an AI draft

How an incident opens and closes

One failed check
Recorded, nothing is sent. The site is checked again after 1 minute.
2 failed checks in a row
The incident opens and alerts go out. It starts at the first failed check, so its length is the real length of the outage.
Acknowledged
Someone said they are on it. Escalation stops. See Escalation and on-call.
2 good checks in a row
The incident resolves by itself and the all-clear goes out with how long it lasted.

A site’s When to alert setting can hold the incident back for a few more minutes. During a maintenance window no incident opens.

What we saw

Open an incident from Incidents in the sidebar. At the top you see how long it has been down, who acknowledged it, which alerts went out and the next step. Then What we saw explains the failure in one sentence, with the details behind it:

  • the DNS answer and the address we connected to;
  • the TLS handshake and the certificate;
  • a timing bar: name lookup, connect, TLS, first byte, transfer;
  • the response: status, headers (with credentials and cookies hidden) and the start of the body;
  • for a browser flow, the step that failed.
An incident page with What we saw and the timeline
An incident: the figures at the top, then What we saw and the timeline.

Plan/Pro and up. A screenshot of the page as it failed, for websites

If a deploy was reported up to 30 minutes before, the title says so: “Deployed v1.4.2 (3f9a2c1) 2 min before”. See Deploy markers.

Timeline and comments

The Timeline lists the first failed check, the opening, every alert with whether it was sent, failed or skipped and why, each escalation, the acknowledgement and the end.

Write in the comment box to tell your team what you see and try. Only your team sees comments. Press ⌘ Enter or Ctrl Enter to post. A comment holds up to 2000 characters. You can edit your own; owners and admins can delete any.

Escalate now

When an escalation policy is paging and you cannot take the incident, press Escalate now at the top of the incident. The next step of the policy is alerted at once instead of after its delay, and the steps after it keep their gaps. After the last step, a policy that repeats starts its next round.

A note under the figures then says which step was paged and lists who it reached, by email, push, SMS or call, with whether each one was sent. The timeline names you on that step. The button is there for anyone who looks after the site, until someone acknowledges; the site page and the phone app have it too. See Escalation and on-call.

Resolve by hand

Owners and admins can press Resolve and add a note for the timeline. Checks keep running: if the site still fails, a new incident opens after 2 more failed checks.

NoteMost incidents need no resolving. They close on their own when the checks pass again.

Tell your users, with an AI draft

What visitors read about an outage lives on the status page, not on the incident. If the site is on a status page, owners and admins see it under the incident’s figures, with whether a report tells visitors about it yet.

  1. 1.
    On the incident, press Draft update with AI.
  2. 2.
    The status page’s report form opens, linked to this incident, with the site marked as a major outage and a draft already being written from the incident. Once the report exists, the same button drafts its next update; after the incident is over, it starts on Resolved.
  3. 3.
    Add what you know in a few words, change the stage or the tone and press Try again if you like. Edit the text, then press Publish. Subscribers get it by email unless you untick that.

You can also start from the page: go to Status pages, open it, press New report under Incident reports, link the incident and press Draft with AI. Once a report is resolved, the incident offers a postmortem too: see Incidents.

Plan/Pro and up. AI drafts are on the paid plans

On the free plan the button reads Write update and opens the same form without a draft.

The draft never sees your URLs or raw errors, only what kind of failure it was. Nothing is published until you press Publish. See Status pages.

Write the postmortem, with an AI draft

A postmortem is the write-up visitors read once it is over: what happened, who was affected and for how long, why, and what you will change. It goes on a resolved report and shows in the page’s history.

  1. 1.
    On the incident, once its report is resolved, press Draft postmortem with AI.
  2. 2.
    The postmortem editor opens with a draft already being written, under the headings Summary, Impact, Timeline, Root cause, What we did and What we will do to prevent it.
  3. 3.
    Fill in every gap marked [[ ]], or remove it. The editor counts the ones left, and Publish stays off until none are left. Then press Publish.

The draft is written from what Heedery recorded: the report and its updates, when the incident began and ended, what the checks saw and from which regions, when your team was alerted and acknowledged it, your team’s comments and a deploy just before it. The times in the timeline are the recorded ones. It never sees your URLs or raw errors, and it names no one.

It does not guess. What the data cannot tell, above all the root cause and what you will change, is left as a gap for you. A deploy just before the incident is mentioned only as a gap to confirm or remove. Write the cause and the fix in What caused it, what you changed before you press Generate and the draft uses them.

It is written in the status page’s language when the page has a fixed one, else in English; pick another language beside the tone. A new draft replaces the text in the editor, and Undo draft brings back what was there. Each draft counts towards the same 20 AI drafts an hour as the updates.

Plan/Pro and up. AI postmortem drafts are on the paid plans

On the free plan the button reads Postmortem and opens the editor empty. With a postmortem already published it reads Edit postmortem, and the editor’s Draft with AI writes a new version.

Still stuck? Write to us.