Observability
OpenSearch provides observability capabilities for monitoring applications, infrastructure, and AI agents. You can ingest logs, traces, and metrics, explore and visualize them in OpenSearch Dashboards, and set up alerts that notify you of issues.
Getting started
To get started, choose the type of data that you want to explore. The log and trace analytics tutorials ingest sample data into OpenSearch and explore it in OpenSearch Dashboards.
Get started with log analytics
Send web logs to OpenSearch using Fluent Bit and Data Prepper, and then query and visualize them in OpenSearch Dashboards.
Configuring observability
To configure observability for your applications, follow these steps. If your data is already in OpenSearch, start at step 2.
Send logs and traces to OpenSearch or connect a Prometheus data source for metrics.
Use the classic apps in the Observability menu or the apps in an Observability workspace and find the app for each task.
To automatically detect anomalies, create monitors that notify you when your data meets a condition.
Monitoring applications
OpenSearch provides two ways to monitor applications: application performance monitoring (APM) for microservices and web applications, and agent monitoring for generative AI applications.
| APM | Agent monitoring | |
|---|---|---|
| Purpose | Monitor microservices and web applications | Monitor AI agents and large language models (LLMs) |
| Metrics | Rate, error, and duration (RED) metrics | Token usage, model calls, agent steps |
| Visualization | Service maps, latency charts, error tracking | Execution graphs (DAGs), trace trees, timelines |
| Conventions | OpenTelemetry standard conventions | OpenTelemetry generative AI semantic conventions |
| Best for | APIs, microservices, web services | Chatbots, AI agents, LLM applications |
APM
APM monitors distributed applications using service topology, RED metrics, and performance tracking. APM requires an Observability workspace, an OpenTelemetry Collector, Data Prepper, and Prometheus.
APM also includes service-level objectives (SLOs) that let you define availability and latency targets for your services and track their error budgets.
Agent monitoring
Agent monitoring traces LLM calls, tool invocations, and token usage in generative AI applications. Agent monitoring requires an Observability workspace, an OpenTelemetry Collector, Data Prepper, and an application instrumented using the OpenTelemetry generative AI semantic conventions.
Alerting and detection
OpenSearch provides the following tools for detecting issues and sending notifications. They work with any index in OpenSearch, including observability data.
Alerting
Create monitors that query your data on a schedule, define triggers for alert conditions, and perform actions when alerts fire.
Anomaly detection
Detect anomalies in your time-series data using machine learning with the Random Cut Forest (RCF) algorithm.
OpenSearch Observability Stack
The OpenSearch Observability Stack provides a complete, preconfigured observability platform that you can run locally using Docker Compose. The Observability Stack includes:
- All APM and agent trace capabilities.
- A GenAI SDK for Python or TypeScript instrumentation.
- An Agent Health tool for local debugging and evaluation.
- A Docker Compose setup with example applications.
- A preconfigured OpenTelemetry Collector, OpenSearch Data Prepper, and Prometheus.
Reference
- Observability security – Control which users can view and modify observability objects, such as notebooks and visualizations.
- Simple Schema for Observability – The index naming conventions and field mappings that some classic observability apps require.