Prometheus Architecture and Data Model
What You'll Learn
You'll learn how Prometheus structures its time-series database, uses labels to organize metrics, and efficiently stores and queries monitoring data.
Core Architecture
Prometheus is a pull-based monitoring system with a built-in time-series database (TSDB). At its core, it periodically scrapes HTTP endpoints (called targets) to collect metrics, stores them locally, and makes them queryable via PromQL.
The key components:
- Scraper: Pulls metrics from configured endpoints at fixed intervals (e.g., every 15 seconds)
- TSDB: Stores time-series data efficiently on disk in compressed chunks
- Query Engine: Executes PromQL queries against stored data
- Alertmanager (separate component): Handles alert routing and notifications
The Label-Based Data Model
Prometheus doesn't use hierarchical namespaces like traditional databases. Instead, every time-series is uniquely identified by:
- A metric name (e.g.,
http_requests_total) - A set of key-value labels (e.g.,
{service="api", method="GET", status="200"})
Think of labels as dimensions on a spreadsheet—they let you slice and dice the same metric across different attributes.
http_requests_total{service="api", method="GET", status="200"} 1234
http_requests_total{service="api", method="POST", status="201"} 567
Each unique combination of metric name and labels creates a separate time-series. Prometheus indexes these efficiently, allowing fast lookups even with millions of series.
Storage Efficiency
Prometheus uses delta-of-delta encoding and compression to minimize disk usage. Data is written in two-hour blocks, then compacted into larger blocks over time. This structure optimizes both write performance and query speed for recent data—exactly what monitoring needs.
Key Takeaway: Prometheus's pull-based architecture, label-driven data model, and purpose-built TSDB make it exceptionally efficient for storing and querying high-cardinality metrics with minimal operational overhead.