Skip to content

Alerts: watch it in plain English

Describe what you want watched in a plain sentence and Rigel turns it into a live background alert that pings your phone.

Type what you want to watch as a plain sentence, like “text me if the postgres database goes down”, and Rigel turns it into a live watch that runs in the background. When the rule trips you get a notification on your phone. The in-cluster agent does the watching itself, so there’s no Prometheus or Alertmanager to stand up for the event-based rules.

Where: Alerts. Write a rule in plain English, or pick a condition, and it starts watching.

Writing a rule

Describe the thing you care about in one sentence and Rigel builds the watch for you: “page me when a pod crash-loops in the payments namespace”, “let me know if a node goes NotReady”, “alert me if node memory goes over 90%”. You don’t wire up queries or thresholds by hand, you say what “bad” looks like and Rigel keeps an eye on it.

Conditions you can watch

  • Pod restarts: a container restarting.
  • Crash loops: a pod stuck in CrashLoopBackOff.
  • OOM kills: a container killed for running out of memory.
  • Pods pending too long: workloads that can’t get scheduled.
  • Node NotReady: a node dropping out of the cluster.
  • Deployment degraded: a workload running below its desired replicas.
  • Metric thresholds: node CPU % and memory % crossing a limit you set.

Where alerts go

When a rule trips you get a notification through whichever channel you’ve set up:

  • Signal
  • Matrix
  • Discord
  • Slack
  • A generic webhook for anything else.

Cutting the noise

  • Quiet hours: set a window so you don’t get paged overnight.
  • Per-rule cooldowns: after a rule fires it holds off before firing again, so a flapping pod doesn’t bury you in messages.

Notes & limitations

  • Alerts are evaluated by the in-cluster agent, so this needs the agent installed. If the agent isn’t running, nothing is being watched.
  • The event-based rules (restarts, crash loops, OOM kills, pending pods, NotReady nodes, degraded deployments) work on their own, no metrics backend required.
  • The metric-threshold rules (node CPU %, node memory %) need a metrics source in the cluster, either metrics-server or a Prometheus-compatible backend. Without one, those thresholds have nothing to read.
  • Notifications land on your phone through a chat channel or webhook. There’s no SMS or phone-call paging.