Kost

Kubernetes clusters waste **30–50% of compute** from over-provisioning. Kost sits in your cluster, polls the Metrics API every 15 minutes, and tells you exactly what's wasting money — with the `kubectl` command to fix it.

---

Overview

Kost is a single-pod Go agent built for teams that want to know where their Kubernetes spend is being wasted without installing a dashboard-first platform or handing usage data to a third party.

Core design rules

• Push over pull — alerts come to you via Slack, you don't go check a dashboard.
• Zero config to start — every field has a default, the agent reports out of the box.
• Every alert is actionable — a finding without a fix command isn't useful.
• Minimal footprint — single pod, ~25MB image, no database.
• Honest scope — pricing is approximate and documented as such; it's a signal, not a source of truth.

---

Why Kost?

Feature	Kost	Typical Platforms
Zero config	✅	⚠️
Single pod	✅	❌
No database	✅	❌
Push-first alerts	✅	⚠️
Built-in dashboard	✅	✅
Ready-to-run fixes	✅	⚠️

---

Features

Detection & Analysis

• Polls the Kubernetes Metrics API and pod specs on a configurable interval (default 15m).
• Compares actual CPU/memory usage against each container's resources.requests.
• Flags workloads where the request exceeds actual usage by 1.5× and estimated waste clears $5/month.
• Resolves the owner chain (Pod → ReplicaSet → Deployment) so fix commands target the correct resource.
• Generates right-sizing suggestions at actual usage × 1.3, with a 0.1 core / 0.1 GB floor.

Reporting & Observability

• JSON reports to stdout on every cycle.
• Prometheus metrics exposed on /metrics.
• Dark-themed HTML dashboard on /dashboard.
• In-memory rolling history of the last 200 reports via /api/reports.
• Optional Slack alerts via webhook.

Deployment & Operations

• Single replica Deployment.
• ~25MB image.
• Runs as non-root (UID 65534).
• Liveness and readiness probes.
• Graceful shutdown.
• Backup script for manifests, RBAC and metrics snapshot.

Security & RBAC

• Read-only ClusterRole.
• No write access.
• No delete permissions.
• No pod exec.
• No secret reads.
• Read-only ConfigMap mount.

---

Tech Stack

Layer	Technologies
Language	Go 1.22+
K8s Client	client-go, apimachinery, Metrics API (dynamic client)
Metrics	Hand-written Prometheus text format
Dashboard	Embedded HTML/CSS/vanilla JS (go:embed)
Alerts	Slack Incoming Webhooks
Deployment	Kubernetes, Docker (multi-stage)
Code Quality	SonarCloud, go vet

---

Architecture Diagram

flowchart TD

A[Kubernetes API]
B[Metrics API]

A --> C[Kost Agent]
B --> C

C --> D[Analyzer]

D --> E[JSON]
D --> F[Prometheus]
D --> G[Dashboard]
D --> H[Slack]

Loading

---

Quick Start

Clone the Repository

git clone https://github.com/nirjxr26/Kost.git
cd Kost

Deploy to Kubernetes

# Create namespace
kubectl create ns kost

# Deploy resources
kubectl apply -f deploy/rbac.yaml \
  -f deploy/configmap.yaml \
  -f deploy/deployment.yaml \
  -f deploy/service.yaml

Every flagged workload includes the exact kubectl set resources command—copy, paste, and apply. Zero configuration is required.

Verify Deployment

kubectl logs -n kost deployment/kost --tail=20

Example output:

{
  "cluster": "prod",
  "waste_monthly": 2847,
  "healthy": false,
  "findings": [
    {
      "workload": "auth-service",
      "namespace": "prod",
      "waste_monthly": 421,
      "fix": "kubectl set resources deployment/auth-service -n prod --requests=cpu=0.3,memory=1434Mi"
    }
  ]
}

Access the UI

Port Forward

kubectl port-forward -n kost deployment/kost 8080:8080

Endpoint	URL
Dashboard	http://localhost:8080/dashboard
Metrics	http://localhost:8080/metrics
Reports API	http://localhost:8080/api/reports
Health	http://localhost:8080/health

Local Development

go build -o kost ./cmd/kost/
go vet ./...
go test ./...

make image
make push
make deploy

Optional: Slack Alerts

kubectl create secret generic kost-slack -n kost \
  --from-literal=SLACK_WEBHOOK_URL="https://hooks.slack.com/services/T00/B00/your-webhook"

---

Project Structure

├── cmd/
│   └── kost/
├── internal/
│   ├── config/
│   ├── k8s/
│   ├── analyze/
│   └── report/
├── deploy/
├── scripts/
│   └── backup.sh

License

MIT