zzutpralik
Log inSign up

Alerts and who hears them

An outage that nobody is told about is a log entry. This is how a message gets from a failed check to a person — and what the product does when it cannot.

On this page
  1. Channels
  2. What generates a message
  3. How the noise is kept down
  4. What happens when a delivery fails
  5. Your own notifications

Channels#

A channel is somewhere a message can go. Add them on the Alerts page, one per destination, and switch them on or off without deleting them.

E-mail
An address. Needs SMTP credentials on the server.
Telegram
A chat id. Needs a bot token on the server.
Webhook
An HTTPS endpoint that receives a JSON body. Works with no server configuration at all, which today makes it the one that actually delivers.
Nothing is ever recorded as sent when it was not. With no SMTP and no bot token, a message to those channels is written to the delivery log with the status skipped and the reason. The alternative — a green tick for a message nobody received — is the failure this product exists to prevent, so it is not available even as a convenience.

What generates a message#

An incident opening
A site failed a check and is now down.
An incident closing
It recovered, with how long it was out.
An expiry warning
A certificate or a domain approaching its date, days in advance.
A monthly report
The client-facing PDF, on a schedule — see Reports.

How the noise is kept down#

Grouping
When several sites fail together, they arrive as one message rather than twenty. Twenty messages at 03:00 are the same information delivered as panic.
Escalation
An incident that stays open past a threshold escalates to another channel. Off by default: escalating to somebody who has not agreed to be woken is a good way to lose a colleague.
Maintenance windows
A planned window suppresses alerting for the sites it covers. The checks still run and the results are still recorded — the minutes are simply not counted as unplanned downtime, and nobody is woken for work you scheduled.

What happens when a delivery fails#

A failed delivery is retried, but never past the point where it is worth having. An incident alert has a shelf life of an hour: a message about an outage that is two hours stale tells somebody about a problem they have already lived through, and trains them to distrust the next one. An expiry warning keeps for a day, a test message for five minutes, and a monthly report is not retried at all — it is regenerated.

Every attempt is in the delivery log, with what happened and why.

Your own notifications#

The channels above belong to the workspace. Each person also has their own, on their profile — so a colleague can be told about incidents on their phone without every client’s alerts landing there too.