Skip to content
Observability Stack

Platform Overview

The OpenSearch Observability Stack is built on open standards with OpenTelemetry at its core. Every component is open-source and runs in Docker containers on your machine.

Architecture

Section titled “Architecture”

Architecture diagram showing microservices and infrastructure sending OTLP to the OTel Collector, which exports to Data Prepper. Data Prepper writes to OpenSearch and Prometheus, both queried by OpenSearch Dashboards.

Data flow

Section titled “Data flow”
  1. Instrumentation: Your applications send traces, logs, and metrics via OTLP to the OTel Collector.
  2. Collection: The OTel Collector batches, processes, and routes data to downstream systems.
  3. Processing: Data Prepper ingests trace data, builds service maps, and computes RED metrics (request rate, error rate, duration).
  4. Storage: OpenSearch indexes traces and logs. Prometheus stores time-series metrics.
  5. Visualization: OpenSearch Dashboards provides trace exploration, agent trace views, PromQL-based metric charts, and service maps.

Key design decisions

Section titled “Key design decisions”
  • OpenTelemetry-native: All data ingestion uses OTel protocols and semantic conventions. No proprietary agents.
  • GenAI semantic conventions: AI agent traces use the standard gen_ai.* OTel attributes, enabling interoperability with any OTel-compatible tool.
  • PPL and PromQL: Query traces and logs with PPL (Piped Processing Language) and metrics with PromQL.
  • Local-first: The entire stack runs on your machine via Docker Compose. No cloud account or external dependencies required.