Surveillance avec Prometheus et Grafana

When running applications in Kubernetes, it's crucial to know how they're performing. Are your pods healthy? Is the cluster overloaded? Monitoring gives you these answers.

In this lesson, we'll explore two powerful tools: Prometheus for collecting metrics and Grafana for visualizing them.

Prometheus is an open-source monitoring system that collects metrics from configured targets at given intervals, evaluates rule expressions, displays the results, and can trigger alerts.

• It uses a pull model: Prometheus scrapes (pulls) metrics from HTTP endpoints.
• It stores data as time-series data, meaning values are recorded over time.

Prometheus gathers metrics from various sources:

• Exporters: Specialized tools that expose existing metrics in a Prometheus-readable format (e.g., node_exporter for host metrics, kube-state-metrics for Kubernetes object metrics).
• Instrumented applications: Applications can directly expose a /metrics endpoint.

Prometheus then scrapes these endpoints.

Deploying Prometheus in Kubernetes often involves using Helm charts or the Prometheus Operator. These help automate the setup and configuration.

Key for Kubernetes monitoring is service discovery. Prometheus can automatically find and scrape pods or services by using Kubernetes API integration, often configured via ServiceMonitor or PodMonitor custom resources.

While Prometheus collects and stores metrics, Grafana is the visualization layer. It allows you to:

• Create rich, interactive dashboards.
• Query multiple data sources (like Prometheus).
• Set up alerts based on visualized data.

Grafana makes complex data easy to understand at a glance.

To visualize Prometheus data in Grafana, you first need to add Prometheus as a data source.

Inside Grafana, you'll specify the URL of your Prometheus server. Once connected, you can use Grafana's interface to write queries using PromQL (Prometheus Query Language) to retrieve specific metrics.

Building a dashboard in Grafana is straightforward:

1. Create a new dashboard.
2. Add a new panel.
3. Select Prometheus as the data source.
4. Write a PromQL query (e.g., sum(rate(node_cpu_seconds_total{mode="idle"}[5m])) by (instance) to see idle CPU usage).
5. Choose a visualization type (graph, gauge, table, etc.).

You can combine many panels to create a comprehensive view.

What should you monitor in Kubernetes?

• Node Health: CPU, memory, disk usage, network I/O.
• Pod Status: Restarts, CPU/memory usage, network traffic.
• API Server: Request latency, error rates.
• Controller/Scheduler: Queue lengths, operation durations.

These metrics provide insights into your cluster's overall health and performance.

Dashboards aren't just pretty pictures; they're powerful diagnostic tools. Learn to:

• Spot trends: Is resource usage consistently increasing?
• Identify anomalies: Sudden spikes or drops often indicate issues.
• Correlate events: Did a deployment cause a performance degradation?

Effective monitoring helps you proactively address problems before they impact users.

You've learned that Prometheus is excellent for collecting and storing time-series metrics, while Grafana provides powerful, customizable dashboards for visualizing this data.

Together, they form a robust monitoring solution for Kubernetes, helping you understand your cluster's health and application performance. Next, explore advanced PromQL, setting up alerts, and integrating with other tools!