Smart Alerting: Catch Cost Issues Before They Hit the Bill

Alerts that find cost problems, not just report them

Cluster spend drifts quietly. A bad rollout doubles CPU requests, a forgotten job runs all weekend, a new workload lands at $900/month — and you find out when the invoice arrives.

Smart Alerting watches your spend continuously and tells you the moment something moves. Set it up once with three building blocks — channels, policies, rules — then the alerts come to you.

The building blocks

Rules — the detectors

A five-step wizard: name it, pick a scope, choose alert types, attach a notification policy, review. One rule per scope, multiple alert types per rule. Scopes go from broad to narrow:

Organization > Cluster > Namespace > Team > Department

Team and Department scopes ride on your cost attribution, so you can alert "Team Payments crossed 90% of budget" without touching a single YAML label. Rules can be muted for 4h, 24h, 7d, or indefinitely when you already know about something.

Channels — where alerts land

Send a test delivery to any channel before you rely on it, and watch channel health at a glance: Healthy, Degraded, or Failing.

Notification Policies — routing and cadence

Group rules into reusable policies to control how and when alerts go out:

• Delivery mode: Immediate, Daily digest, or In-app only
• Daily cap: overflow rolls into a single digest instead of flooding the channel
• Quiet hours: restrict delivery to a time window; alerts queue until it opens

Edit the policy once and every rule attached to it follows.

What makes it smart

Self-learning cost-spike detection. No thresholds to guess. Spike detection adapts to each scope's own spending baseline and fires when spend jumps clear of the expected band. New scopes need about 7 days of cost history before detection activates.

AI incident analysis. Every incident has an "Explain this incident" button that produces a plain-language what-happened, why-it-matters, and next-steps — with the causal chain and cost drivers laid out. See Kubeadapt AI for how that works, and why none of your utilization metrics ever leave your workspace.

Noise control. Optimization suggestions notify you without creating incidents. Daily caps and quiet hours keep channels calm. New-workload alerts auto-exclude operator- and CRD-owned resources so you're not paged for the control plane.

Alert types

Incident-creating:

• Cost spike — spend vs. the adaptive baseline
• Budget threshold — monthly, quarterly, or fiscal-year budgets, with early warnings at 50 / 80 / 90 / 100%; loaded cost (CPU + RAM + GPU + overhead) or workload-only
• New expensive workload — newly deployed high-cost workloads, with a configurable minimum ($750/month by default)

Notify-only optimization alerts — fan out into workload rightsizing, Spot-eligible workloads, and orphaned resources (idle PVs, load balancers, disks, static IPs, snapshots). These land in your channel but never open an incident.

Coming soon

Unused resources (idle-workload digest) ships in v1.1, and Network Costs optimization alerts (cross-AZ / inter-cluster spend) are on the way. Everything else above is live today.

<!-- callout:note -->

Incidents: the full trail

Every alert that fires becomes an incident you can track. Filter by status, time range, or type. Each incident shows the delta $/day, how long it's been firing, and the top contributing workload. Work them with Acknowledge, Snooze, and Resolve — states move through Pending, Firing, Acknowledged, Snoozed, and Resolved.

A firing incident reads like this:

Cost spike on Team Payments — checkout-worker CPU requests went 80 → 220 cores, +$620/day vs. baseline.

Getting started

Go to Alerts. Connect a channel, create a notification policy, then add your first rule — the built-in guide walks you through all three in about two minutes. For details, see Smart Alerting.