Do I have to rebuild my monitoring?

No. One HTTP call gets a different address and a different key. Whoever talks to Pushover today only changes the address: the route /1/messages.json takes the same field names.

What happens during an alert storm?

Messages sharing a dedup_key are one incident rather than one ring per repetition, and the incident stays open until it is resolved. On top of that a per-source limit caps how much a single connection can set off at all.

Can several people receive the same alerts?

A rule can escalate to a second device or to a mail address after a period without acknowledgement. That is not a rota handing the duty on automatically, and this page does not claim it is.

How loud is it before I have written the first rule?

Without rules the sender's suggestion decides: info becomes low, warning becomes normal, error becomes high, critical becomes critical. That is a usable start and exactly the default you overwrite in the first week.

PagerDeck for IT operations

Ten connected systems, one phone, and a rule chain that decides which of them is allowed to ring at night.

This page describes the case PagerDeck was built for: being on call for systems nobody notices as long as they run. The middle of it is four rule chains, and every one of them is evaluated exactly as it is written here.

What it looks like after the third month

At the start one source is connected and everything is calm. The problem appears later, once ten systems are sending and each of them brings its own idea of what may pull a person out of bed at four in the morning.

Every system arrives with its own notion of critical, and none of them knows who is on call.
Changing how loud something is means changing it where the alert is raised, often in a configuration that belongs to somebody else.
What is left is the choice between constant noise and switching the notifications off, which is the decision you regret at the next outage.

Four chains from a real rotation

None of them needs a capability that does not exist. The conditions are source, tags, text pattern, urgency and time window; the result is one of the five levels.

Disk on a web server at 94 percent
Conditions (when)
source: prometheus-prodtags.any: disk22:00-07:00 Europe/Berlin
Low
It will not fill up in the next six hours. Delivered quietly at night, and in the morning it sits at the top of the history. The same message at midday misses the time window and stays at normal.
UPS reports it is running on battery
Conditions (when)
source: ups-dc1match.title: /on battery/i
Actions (then)
stop: true
High
Time critical, at any hour, and without a later rule softening it afterwards. That is what stop is for: the chain ends here.
The overnight backup failed
Conditions (when)
source: backupseverity_at_least: error23:00-06:00 Europe/Berlin
Silent
The sender reports an error, and an error it is. Only a failed backup wakes nobody: at six it stands in the history without anybody having got up.
The monitoring itself stopped answering
Conditions (when)
source: watchdog
Actions (then)
escalate_after: 5mthen_notify: device:backup-phone
Critical
The one alert after which no further alert can arrive. Left unacknowledged for five minutes it additionally goes to the second device. Without Apple's approval PagerDeck delivers it as high, visible in the history.

The conditions are written the way they stand in the rule editor. Evaluation runs top to bottom, and the first matching rule carrying stop ends the chain.

What usually sends

An HTTP call with the source's ingest key, or a mail to the source's own address. None of it has to be installed inside the monitoring system.

Prometheus Alertmanager

A receiver that can shape the call itself, such as a webhook relay or the route through the source's mail address. Whatever you send in the tags field becomes a condition in the rule chain.

Zabbix, Checkmk, Icinga

All three can run an HTTP call as an action. What changes is the address and the key.

Uptime and certificate checks

Expiry warnings are the classic case for silent: they belong in the history, not in the night.

Backup and job systems

Whatever only speaks mail gets its own address per source from Business, allowlist for senders included.

Your own scripts and cron jobs

A curl at the end of the script. No SDK, no library, no dependency that has to be maintained alongside.

CI and deployments

A failed nightly run is rarely urgent. As a source of its own it can be quietened in a single line.

Where PagerDeck is the wrong tool

Three cases in which something else is the better answer. They stand here because the alternative is finding them out after connecting everything.

Android. There is an app for iPhone and iPad, and one for Windows on the desktop that holds its connection open itself and cannot wake a machine that is switched off; on Android only the history in a browser remains.
Rotas with a rotation. PagerDeck knows rules, devices and escalation targets, but not a calendar that hands the on-call duty on every week.
A single system that you own yourself. Then the priority sits exactly right at the sender, and a rule chain would be a layer with no job.

Frequently asked from operations

Do I have to rebuild my monitoring?: No. One HTTP call gets a different address and a different key. Whoever talks to Pushover today only changes the address: the route /1/messages.json takes the same field names.
What happens during an alert storm?: Messages sharing a dedup_key are one incident rather than one ring per repetition, and the incident stays open until it is resolved. On top of that a per-source limit caps how much a single connection can set off at all.
Can several people receive the same alerts?: A rule can escalate to a second device or to a mail address after a period without acknowledgement. That is not a rota handing the duty on automatically, and this page does not claim it is.
How loud is it before I have written the first rule?: Without rules the sender's suggestion decides: info becomes low, warning becomes normal, error becomes high, critical becomes critical. That is a usable start and exactly the default you overwrite in the first week.

Fourteen days, full feature set, no payment method

Connect one source, write three rules, wait one night. After that you know whether the chain is right.

Product and company names mentioned are trademarks of their respective owners. PagerDeck is not affiliated with, sponsored or endorsed by them. Details in the imprint.