เครื่องมือสังเกตระบบ Kubernetes

Kubernetes environments are dynamic and complex. Pods come and go, services scale, and nodes can fail. Traditional monitoring struggles with this constant change.

• Observability in K8s means understanding the health and performance of your entire cluster, from nodes to individual application containers.
• We need specialized tools to collect logs, metrics, and traces from these ever-changing components effectively.

To keep your K8s cluster healthy and your applications running smoothly, you need to monitor several key areas:

• Resource Utilization: CPU, memory, disk, and network usage across nodes and pods.
• Application Health: Readiness and liveness probes, error rates, latency of your deployed apps.
• Cluster Components: Health of the control plane (API server, scheduler, etcd) and worker nodes.
• Network Traffic: Ingress/egress, DNS resolution, and service-to-service communication.

Prometheus is the leading open-source monitoring system for Kubernetes. It excels at collecting time-series metrics from configured targets at regular intervals.

• It uses a pull model: Prometheus actively scrapes metrics from endpoints exposed by your applications and Kubernetes components.
• It integrates with K8s service discovery to automatically find new targets (pods, services) to scrape.
• Grafana is often paired with Prometheus to create powerful, customizable dashboards for visualizing these metrics.

Prometheus uses Kubernetes' native service discovery to automatically find metric endpoints. For example, it can discover kube-state-metrics, which exposes metrics about the state of K8s objects (pods, deployments, etc.).

You can check where Prometheus components might be running in your cluster (assuming a common installation namespace):

For centralized logging in Kubernetes, Fluentd and its lightweight cousin, Fluent Bit, are popular choices. They ensure logs from ephemeral containers aren't lost.

• They run as DaemonSets on each node, collecting logs from all containers on that node.
• They can parse logs, add valuable K8s metadata (like pod name, namespace), and forward them to a centralized logging backend (e.g., Elasticsearch, Loki).
• Fluent Bit is often preferred for its smaller footprint and lower resource consumption in cloud-native environments.

Fluent Bit is typically deployed as a DaemonSet, ensuring a log collector runs on every node and captures all container logs. This ensures comprehensive log coverage.

Here's how you might check the status of a Fluent Bit DaemonSet:

Distributed tracing helps visualize requests flowing through multiple microservices in Kubernetes. Jaeger and Zipkin are common open-source tracing systems.

• Applications are instrumented (often using OpenTelemetry SDKs) to send trace data to a collector.
• Collectors (e.g., OpenTelemetry Collector) can run as DaemonSets or Deployments within your K8s cluster to receive and process trace data.
• These tools are crucial for identifying latency bottlenecks and errors across service boundaries in complex K8s deployments.

Before diving into full-fledged observability platforms, Kubernetes offers powerful built-in commands for quick insights and initial debugging:

• kubectl top node: Shows CPU and memory usage for nodes.
• kubectl top pod: Shows CPU and memory usage for pods.
• kubectl describe pod <pod-name>: Provides detailed information about a specific pod, including events, status, and resource requests/limits.

These are invaluable for immediate troubleshooting and resource assessment.

We've explored essential tools and strategies for observing Kubernetes clusters effectively:

• Prometheus & Grafana are the go-to for collecting and visualizing cluster and application metrics.
• Fluentd/Fluent Bit provide robust, centralized log collection from containers and nodes.
• Jaeger/Zipkin are critical for distributed tracing, helping understand complex microservice interactions.
• Native kubectl commands offer quick, on-the-spot insights into your cluster's state.

Combining these tools provides a powerful, unified view into your cloud-native applications and infrastructure.