Monitors

A monitor watches one query over a sliding window and alerts when a condition holds.

• Query: a metric (name, optional attrs, and avg, sum, min, max or count), or logs (optional service, minSev, contains, attrs), whose value is the number of matching records.
• Window: 1 to 60 minutes.
• Condition: > or < a threshold.

Pulse evaluates every monitor once a minute. When the condition holds over the window, the monitor goes from ok to alerting and Pulse sends a trigger; when it stops holding, it goes back to ok and Pulse sends a resolve. A metric window with no data at all changes nothing.

{ "name": "Queue backing up",
  "spec": { "query": { "signal": "metrics", "name": "queue.depth", "agg": "avg" }, "windowMin": 5, "op": ">", "threshold": 1000 } }

{ "name": "Checkout errors",
  "spec": { "query": { "signal": "logs", "service": "checkout", "minSev": 17 }, "windowMin": 10, "op": ">", "threshold": 20 } }

A new monitor judges data that arrives after it was created, so its first full window ends windowMin minutes later.

Manage them in Monitors in the app, with GET/POST /api/v1/monitors and PATCH/DELETE /api/v1/monitors/:id, or with the MCP tools list_monitors, create_monitor, update_monitor and delete_monitor. Alerts go where Alerts and Shift says.

Next: Alerts and Shift · This page as Markdown