Glossary

Agent — accelerated, the binary installed on a server. Dials out; never listens. See Fleet.

Alert — a statement that a condition is currently true, with a severity, a state and a fingerprint. See How alerting works.

Baseline column — a column of the same row that a condition compares against, instead of a fixed number. See Conditions.

Canary — the first machine in a staged agent rollout. Deliberately an unsupervised box, so a bad version is found where it costs least.

Channel — a delivery destination for notifications. See Channels.

Cooldown — the pause after a resolve before the same condition may fire again.

Coverage — whether anything is actually watching a thing that is running. See Coverage.

Direction column — a text column whose value says whether a rise or a fall is the problem, per row. See Directions.

DLQ — dead-letter queue. Notifications that exhausted every automatic retry. Shown as Gave up.

Environment — a deployment tier as a first-class object. ResolveEnvironment is the only answer to "is this production?". See Environments.

Error budget — how much unreliability an SLO allows before you stop shipping. See SLOs.

Fingerprint — the identity of an alert. The same condition firing again updates the existing alert rather than creating a second one.

For — how long a condition must hold before anyone is told.

Grant — time-limited access to a server through the agent, carrying an OS identity. See Access and grants.

Group key / subject — the column that identifies a row in a per-row alert. See Per-row alerts.

Guard — a condition evaluated before the measurements; if it fails, the measurements are not reported. See Conditions.

Incident — a human response to one or more alerts. See Incidents.

Lease — the hold a worker takes on a queued retry, so a crashed attempt is picked up once rather than spinning.

Notify mode — per_subject or summary. Whether a grouped rule sends one message per subject or one for the run.

Organisation — the tenant. Every row belongs to exactly one.

Pool — a Cloudflare origin pool. A group of servers a load balancer can steer traffic to.

Probe — a check the agent runs from a server, rather than from the control plane. See Probes.

Profile — the agent's machine-side capability cap: observe, operate or full. Set on the machine; cannot be widened remotely.

Recovery — how long a condition must be false before the alert closes.

Rollup — one alert standing for many breaching subjects.

Run window — the days and hours a rule may run at all. Outside it, the query is not run. See Schedules.

Service — a thing that serves requests and can be up or down.

Severity — info, low, warning, high, critical.

Silence — suppressing notifications for one alert or fingerprint.

SLO — a target for a service level, with an error budget.

Steering — how a Cloudflare load balancer decides which pool gets a request.

Warn threshold — a milder level on the same condition, so one rule fires at warning first and escalates itself.