Course contentsShow
← System Design
Lesson 1241 of 1,91931. Observability: TracingFree lesson

Jaeger Architecture and Components

Learn Jaeger's agent-collector-query architecture and how it ingests, stores, and visualizes trace data at scale.

Jaeger Architecture and Components

What you'll learn: How Jaeger's distributed architecture ingests, stores, and visualizes trace data at massive scale.

The Three-Tier Architecture

Jaeger is built around three core components that work together like a conveyor belt in a factory:

Jaeger Agent acts as a local daemon on each host (typically a sidecar). It listens for spans sent over UDP from your instrumented applications, batches them, and forwards them upstream. Think of it as a local post office collecting mail before sending it to a regional hub.

Jaeger Collector receives traces from multiple agents, validates them, runs processing pipelines (enrichment, indexing), and writes them to persistent storage. This is your regional distribution center—handling high throughput from many sources and organizing data for long-term storage.

Jaeger Query Service provides the API and UI for retrieving and visualizing traces. It reads from the storage backend and lets you search by trace ID, service name, operation, or tags. This is your search engine and display layer.

Storage Backends

Jaeger doesn't dictate storage—it supports Cassandra, Elasticsearch, Kafka (as a buffer), and in-memory stores for testing. The collector writes spans to your chosen backend, and the query service reads from it.

Why This Architecture Scales

By separating concerns—local collection (agent), aggregation and persistence (collector), and querying (query service)—Jaeger handles millions of spans per second. Agents prevent application blocking, collectors scale horizontally, and query services can be replicated independently.

The agent-collector pattern also keeps your application's tracing overhead minimal: fire-and-forget UDP spans locally, with all heavy lifting happening out-of-band.

Key Takeaway: Jaeger's agent-collector-query architecture separates trace ingestion, storage, and retrieval into scalable, independently deployable components that handle high-volume distributed tracing without blocking application code.