Search⌘ K
AI Features
Home
Courses
System Design Patterns

Server-Side Monitoring

Explore server-side monitoring to understand how metrics, structured logs, and distributed traces maintain backend health in scalable systems. Learn to use RED and USE frameworks for metrics, choose monitoring tools, instrument services, and execute effective incident response to improve reliability and observability across distributed backend architectures.

We'll cover the following...

In the previous lesson, we explored how client-side monitoring captures what users actually experience through real user monitoring, synthetic checks, and error tracking. Those signals reveal symptoms. As illustrated below, a single user request passing through edge infrastructure can trigger an HTTP 500 Internal Server Error, a clear signal that the backend has failed, even if the user's browser is functioning perfectly. Diagnosing root causes, however, requires visibility into the backend: infrastructure health, application logic, and inter-service communication.

Server-side errors
Server-side errors

Server-side monitoring is the practice of collecting, correlating, and acting on metrics, logs, and traces emitted by backend services to maintain data consistency and system reliability across distributed architectures. This lesson covers four areas: the key metrics that expose infrastructure and application health, the tools used to collect and visualize telemetry, how to set up structured logs and distributed traces, and how to respond to server-side incidents effectively.

Key metrics for server-side monitoring

Two complementary frameworks govern what to measure on the server side.

  • RED method: It targets request-driven services such as APIs and microservices. Each RED metric maps to a specific dimension of service behavior.

    • Request rate measures throughput as the number of requests processed per second, revealing demand patterns and capacity needs.

    • Error rate captures the percentage of responses returning 4xx or 5xx status codes, directly reflecting service correctness.

    • Request duration tracks latency at percentile boundaries (p50, p95, and p99) rather than averages, because a p99 spike in one microservice cascades into tail latency across the entire call chain.

  • USE method: It targets infrastructure resources like CPU, memory, disk, and network interfaces. Applying both ensures no category of signal is missed. Infrastructure-level USE metrics catch problems ...