Monitoring glossary

Dead man's switch

A mechanism that fires when an expected signal stops arriving, rather than when something reports an error. In monitoring, a job checks in on each run; if a check-in is late or never arrives, the switch raises the alert. Mortemain is a dead man's switch for scheduled jobs: silence is the alarm.

Heartbeat monitoring (check-in monitoring)

Watching for a periodic "I ran / I'm alive" signal from a job or service. The monitored thing pings the monitor from the inside on each run, so a missing beat is a first-person report that something is wrong, not an outside guess.

Passive check

Monitoring where the job reports its own status to the monitor, as opposed to an active check where the monitor probes from outside. With a passive check, the absence of an expected check-in is itself the signal.

Uptime monitoring

Checking that a service is reachable, usually by probing it from the outside. It answers "is it up from here", which can miss a job that silently failed while the host still responds to the probe.

Silent failure

A failure that produces no error and no alert: a job that never started, a backup that never finished, a sync that quietly stalled. Error-only monitoring can't see it; absence-based monitoring can.

Grace period

The extra time a monitor waits after a job's expected check-in before declaring it late. It absorbs normal jitter (a backup that sometimes runs a few minutes over) without firing false alarms.

Flapping

Rapid alternation between up and down states, which floods you with noisy alerts and trains you to ignore them. Grace periods and de-bouncing keep a flaky check from crying wolf.

Ping URL (check-in URL)

The unique, unguessable URL a job requests to signal that it ran. In Mortemain, requesting it marks the check up; requesting the same URL with /fail on the end marks it down immediately.

Cron job

A command scheduled to run automatically at fixed times or intervals on Unix-like systems, defined in a crontab. The workhorse behind countless backups, cleanups and data syncs, and a prime candidate for a dead man's switch.

Cron expression

The five-field syntax (minute, hour, day-of-month, month, day-of-week) that tells cron when to run a job. For example 0 3 * * * means 3:00am every day.

Scheduled Task (Windows)

The Windows equivalent of cron, managed by Task Scheduler, which runs programs or scripts on a trigger or schedule. It can call a check-in URL directly, or via a one-line PowerShell command.

systemd timer

A systemd unit that starts another unit on a schedule or after an event, the modern Linux alternative to cron for many services. Timers pair with a service unit that does the work.

Kubernetes CronJob

A Kubernetes object that runs a Job on a cron schedule inside a cluster, for periodic batch work like report generation or database maintenance.

Alert fatigue

Desensitisation caused by too many alerts, especially false positives, until real incidents get ignored. Fewer, meaningful alerts, the kind that only fire when something truly stopped, are the cure.

False negative / false positive

A false negative is a real problem with no alert; a false positive is an alert with no real problem. Silent failures are the classic false negative, and they are exactly what absence-based monitoring is built to catch.

On-call

The rotation of being responsible for responding to alerts, often outside normal working hours. Good monitoring respects on-call by only paging for things that matter.

SLA, SLO, SLI

A Service Level Agreement is a promised level of service; a Service Level Objective is the internal target; a Service Level Indicator is the measured value, such as uptime percentage. Your monitoring produces the SLIs you report against.

Backup monitoring

Verifying that scheduled backups actually ran and completed, not merely that the backup tool is installed. A missed or half-finished backup is a textbook silent failure, and you want to know before you need the restore.

Exit status (exit code)

The number a process returns when it finishes: 0 for success, non-zero for failure. A script can send it to the monitor so a check distinguishes "ran" from "ran and failed".

Webhook

An HTTP callback an app fires when an event happens. Pointed at a check-in URL, it lets no-code tools (Zapier, Make, Power Automate, n8n) report a job done without any script at all.

See the integration guides to wire any of these into a Mortemain check.