kubernetes-sigs/metrics-server
Metrics Server
Metrics Server is a lightweight, single-purpose daemon that collects CPU and memory usage data from every node and pod in a Kubernetes cluster and exposes those metrics through a standard Kubernetes API endpoint. It registers as an aggregated extension API server behind the Kubernetes apiserver, making resource utilization data available to the Horizontal Pod Autoscaler and Vertical Pod Autoscaler for automatic replica count and resource request adjustments.
The project distinguishes itself by operating as a focused, in-cluster resource metrics collector that polls kubelet summary endpoints across all nodes, with resource consumption scaling linearly to support clusters of up to 5,000 nodes. It secures all API-to-node traffic through HTTPS, client certificate validation, and tunnel-based proxy connections, while delegating authorization decisions to the Kubernetes apiserver via access reviews. Multiple replicas can run with leader election to maintain availability during failures.
Beyond its core metrics collection and autoscaling pipeline, Metrics Server exposes node and pod resource usage through the standard metrics API for command-line querying with tools like top, and supports extension API registration for custom metric endpoints. The project provides a complete autoscaling data source that feeds container-level resource usage to both horizontal and vertical autoscalers, with configurable scaling behaviors and stabilization windows.
Features
- Background Resource Metric Collectors - A lightweight daemon that gathers per-node and per-pod resource utilization data and serves it via a standard API endpoint.
- Kubelet Metric Pollers - Polls kubelet summary endpoints on each node to gather CPU and memory usage for pods and nodes.
- Kubernetes Horizontal Pod Autoscalers - Provides the resource metrics pipeline that feeds CPU and memory usage to the Horizontal Pod Autoscaler.
- Resource Metric Feeds - Supplies per-pod resource metrics to the horizontal autoscaler so it can scale replicas based on CPU or memory load.
- Vertical Autoscaler Data Feeds - Provides container-level resource usage data to the vertical autoscaler for automatic CPU and memory request recommendations.
- Vertical Autoscaler Data Sources - Supplies container-level resource usage data to the Vertical Pod Autoscaler for automatic CPU and memory request recommendations.
- API Service Registrations - Maps URL paths to external services so the Kubernetes apiserver proxies matching requests to them.
- Pod Autoscaling - Feeds container resource metrics to horizontal and vertical autoscalers for automatic replica and resource adjustments.
- Resource-Metric-Based Auto Scaling - Adjusts the number of running pods based on average CPU or memory usage relative to configured resource requests.
- Autoscaler Data Feeds - Delivers resource usage data to autoscalers so they can adjust replica counts and resource requests based on current demand.
- Client Certificate Authentication - Validates incoming proxy requests using signed client certificates to ensure they originate from the Kubernetes apiserver.
- Kubelet Certificate Validations - Validates kubelet HTTPS certificates against a root CA to secure API-to-node connections.
- Kubernetes Node-to-API Server Communications - Secures all node and pod traffic to the API server with HTTPS, client certificates, and service account tokens.
- Custom Metric API Servers - Registers as an aggregated API server to proxy custom metric requests through the Kubernetes apiserver.
- Kubernetes Node Metrics - Collects CPU and memory usage from every node and pod and exposes them through a standard Kubernetes API.
- Kubelet API Scraping - Collects CPU and memory usage data from every node by polling kubelet summary endpoints.
- Custom Metrics API Exposures - Exposes resource metrics through a read-only API following the Kubernetes custom metrics API conventions.
- Resource Metrics - Exposes CPU and memory usage for every pod in the cluster through a standard API endpoint.
- Extension API Server Registrations - Registers as an aggregated extension API server behind the Kubernetes apiserver for custom metric endpoints.
- Scaling Metric Aggregation - Evaluates several metrics simultaneously and applies the largest recommended replica count to the target workload.
- Container-Specific Metric Scalers - Adjusts replica count based on resource usage of a specific container within each pod, ignoring sidecar containers.
- Scaling Behavior Configurations - Controls the rate of replica changes during scale-up and scale-down, and sets stabilization windows to prevent unnecessary fluctuations.
- CLI Metric Queries - Exposes node and pod resource usage via a standard API queryable with a top command.
- Sequential Node Pollers - Scans all nodes sequentially or in batches, keeping resource consumption proportional to cluster size.
- Control Plane Tunneling - Secures API-to-node traffic with HTTPS, client certificates, and tunneled proxy connections.
- High Availability Deployments - Deploys multiple replicas with leader election to keep autoscaling pipelines available during failures.
- Leader Election - Runs multiple replicas with leader election so only one instance collects metrics while others stand by.
- Large-Scale Cluster Monitoring - Collects resource metrics across clusters with up to 5,000 nodes while maintaining linear resource consumption.
- Secure Tunnels - Routes API-to-node traffic through persistent tunnels to secure communication over untrusted networks.
- Kubernetes Access Review Delegations - Delegates authorization decisions to the API server using access reviews to control access to node endpoints.
- Proxy Request Certificate Authentications - Validates proxied requests originate from trusted sources using signed client certificates.
- Kubernetes Resource Path Authorizations - Maps HTTP request paths to resource attributes so access to node endpoints can be individually permitted or denied.
- Custom Metric Autoscalers - Adjusts replica count using any custom metric exposed through the metrics API, not limited to built-in resources.
- Kubernetes Access Review Delegations - Delegates authorization of proxied requests to the API server by submitting access reviews.
- Monitoring and Visualization - Scalable source of container metrics for autoscaling pipelines.
