Kubeadapt

Frequently Asked Questions

General Questions

What is Kubeadapt?

Kubeadapt is a Kubernetes cost monitoring and optimization platform that helps you:

  • Understand where your Kubernetes money goes
  • Identify opportunities to reduce costs (20-40% average savings)
  • Prevent unexpected cost increases before they hit production

For a more comprehensive overview of Kubeadapt's platform and features, please refer to the Introduction section.

Is Kubeadapt read-only?

Yes, Kubeadapt is read-only.

What Kubeadapt does:

  • Reads cluster metrics
  • Calculates costs
  • Shows recommendations
  • Detects anomalies
  • Sends alerts
  • Automatically opens pull requests (and auto-merges if you enable it), integrating recommendations into your GitOps workflow. This way, updates are picked up by tools like ArgoCD or FluxCD during their regular repository scans. Everything is auditable through Git.

What Kubeadapt does NOT do:

  • Modify your workloads
  • Delete resources
  • Change configurations
  • Auto-scale anything

You stay in control. Kubeadapt shows you what to optimize, you decide when and how to implement recommendations.

What Kubernetes distributions are supported?

Fully supported:

  • AWS EKS
  • Google GKE
  • Azure AKS
  • Self-hosted (kubeadm, kops, Rancher, etc.)
  • OpenShift

Requirements:

  • Kubernetes 1.24 or higher
  • Outbound HTTPS access (agent → Kubeadapt Cloud)

Data & Security

What data does Kubeadapt collect?

Collected from your cluster:

Kubeadapt collects ~122-172+ metrics from your Kubernetes cluster via Prometheus scraping:

Prometheus Internal Metrics (~50-100+ metrics):

Monitoring & Observability:

  • All Prometheus internal metrics (prometheus_* wildcard pattern)
  • Scrape Status (up)
  • Current Time (time)

These include TSDB health, scrape statistics, rule evaluation metrics, target discovery, and HTTP request metrics.

Kubernetes State Metrics (39 metrics):

Pod & Container Metadata:

  • Pod Info (namespace, node, host IP)
  • Pod Status Phase (Running, Pending, Failed, etc.)
  • Pod Owner References (Deployment, StatefulSet, DaemonSet)
  • Container Info (image, image ID)
  • Container Status (running, waiting, terminated)
  • Container Restart Counts
  • Container Waiting/Termination Reasons

Resource Allocation:

  • Container CPU Requests
  • Container Memory Requests
  • Container CPU Limits
  • Container Memory Limits
  • Node Allocatable Resources (CPU, memory, pods)
  • Node Capacity (CPU cores, memory bytes)

Workload Status:

  • Deployment Creation Time
  • Deployment Replicas (desired, ready)
  • StatefulSet Replicas (desired, ready, generation)
  • DaemonSet Status (ready, desired, current)
  • Job Status (succeeded, failed, active, completions)
  • CronJob Suspend Status
  • HorizontalPodAutoscaler Info

Storage:

  • PersistentVolume Info (storage class)
  • PersistentVolume Capacity
  • PersistentVolumeClaim Info
  • PersistentVolumeClaim Requests
  • Pod Volume Bindings

Cluster Metadata:

  • Namespace Labels
  • Node Info (internal IP, OS, architecture)
  • Node Labels (instance type, zone, region)
  • ReplicaSet Owner References

Container Resource Usage (13 metrics):

CPU & Memory:

  • Container CPU Usage (seconds total)
  • Container Memory Working Set
  • Machine CPU Cores

Network:

  • Container Network Receive Bytes
  • Container Network Transmit Bytes

Filesystem I/O:

  • Container Filesystem Usage
  • Container Filesystem Limits
  • Container Filesystem Read Operations
  • Container Filesystem Write Operations
  • Container Filesystem Read Bytes
  • Container Filesystem Write Bytes
  • Container Filesystem I/O Time
  • Container Filesystem Current I/O Operations

Cost Metrics (10 metrics):

Pricing:

  • Node CPU Hourly Cost
  • Node RAM Hourly Cost
  • Node GPU Hourly Cost
  • PersistentVolume Hourly Cost
  • Cluster Management Cost

Cost Context:

  • Cluster Info (version, provider)
  • Node Spot Instance Detection
  • Node Labels (for cost allocation)
  • PersistentVolumeClaim Info
  • PersistentVolumeClaim Storage Requests

Node-Level Metrics (8 metrics):

CPU:

  • Node CPU Usage (seconds per mode)

Memory:

  • Node Total Memory
  • Node Available Memory

Filesystem:

  • Node Filesystem Total Size
  • Node Filesystem Available Space
  • Node Filesystem Free Space

Disk I/O:

  • Node Disk I/O Time
  • Node Disk Current I/O Operations

GPU Metrics (3 metrics - Optional):

When gpu-operator is enabled (requires NVIDIA GPU Operator):

GPU Utilization & Performance:

  • GPU Compute Utilization (DCGM_FI_DEV_GPU_UTIL)
  • GPU Memory Copy Utilization (DCGM_FI_DEV_MEM_COPY_UTIL)
  • GPU Power Usage (DCGM_FI_DEV_POWER_USAGE)

These metrics are scraped from NVIDIA DCGM Exporter for GPU cost optimization.

Network Cost Metrics (2 metrics - Optional):

When networkCost feature is enabled via eBPF agent:

Pod-to-Pod Network Traffic:

  • Connection Traffic Bytes Total (per IP pair, protocol)
  • Connection Traffic Packets Total (per IP pair, protocol)

These metrics track egress network traffic between pods for accurate network cost attribution.

NOT collected:

  • Application logs
  • Application code
  • Secrets or ConfigMaps
  • Environment variables
  • Application data
  • User traffic or requests
  • Container exec commands
  • Image contents
  • Pod logs or events

Does Kubeadapt automatically optimize my cluster?

No. Kubeadapt is read-only and provides recommendations only. Though, it can inject into your SDLC through a GitOps Friendly set of features.

Why read-only?

  • You maintain full control
  • No risk of automated changes breaking things
  • You decide what/when/how to optimize

Implementation: Follow Right-sizing Guide to implement recommendations safely.

Can Kubeadapt prevent cost overruns?

Yes, with Cost Prevention features:

1. Cost Firewall

  • Budget thresholds (80%, 90%, 100%)
  • Spike detection (+20% rapid increase)
  • Anomaly detection (unusual spending patterns)
  • Delivered to Slack, email, PagerDuty

2. Cost Gates (GitOps integration)

  • Analyze Kubernetes YAML in pull requests
  • Estimate cost impact before merge
  • AutoPR/Automerge capabilities for implemented deployments
  • Block PRs exceeding thresholds
  • Comment with cost analysis

3. Cost Alerts (monitoring only)

  • Monitors new deployments
  • Alerts if deployment exceeds cost thresholds
  • Shows what would have been blocked (read-only)

All prevention features provide visibility and alerting, not enforcement.

Can Kubeadapt track costs by team or project?

Yes, we have a cost query engine that provides our users with a very flexible way to query costs across different types of entities such as containers, pods, deployments, labels, and more.

Does Kubeadapt support multi-cluster environments?

Yes, multi-cluster is fully supported. You can install our agents on your Kubernetes clusters and monitor all of them from a single dashboard.

Can I get cost alerts in Slack?

Yes, Slack integration is fully supported.

Does Kubeadapt have an API?

Currently, Kubeadapt does not offer a public API. However, we are actively considering API access based on user demand and future requirements. If you have a specific use case or need API access, please let us know so we can prioritize accordingly.

Documentation: Settings → API Tokens → API Docs

What if I have multiple clusters?

Yes, multi-cluster is a core feature of Kubeadapt.

Can I try before purchasing?

Yes, we have a free plan.

Start: https://app.kubeadapt.io/signup

Still Have Questions?

Search docs: https://kubeadapt.io/docs/v1 Email support: authors@kubeadapt.io

Back to Docs
Edit this page