OpenTelemetry Demo 3.0 Breaks Dashboards on Purpose and Adds an Agentic AI Stack

The new release renames every custom attribute, replaces the old compose file and adds a fully instrumented LangGraph agent, MCP server and chatbot, plus continuous profiling and OpAMP.
OpenTelemetry Demo 3.0 Breaks Dashboards on Purpose and Adds an Agentic AI Stack
OpenTelemetry Demo 3.0 Breaks Dashboards on Purpose and Adds an Agentic AI Stack

The OpenTelemetry project has shipped version 3.0 of its official demo application, and it knowingly broke things to get there. The announcement post, titled “We broke the OTel demo,” warns users that “if you had dashboards with custom metrics, everything is broken now.”

“This was no accident,” wrote Juliano Costa of Datadog, who authored the post. He described the release as clearing out accumulated pieces that “were getting in the way of where we wanted to go.”

The two breaking changes

Every custom telemetry attribute moved from the app.* namespace to demo.*, so app.product.id is now demo.product.id. The demo started using app.* in 2022, before the name was reserved; in 2023 the semantic conventions claimed it for client-side applications. Martin Thwaites brought OpenTelemetry Weaver into the project to help with the rename, and Florian Bourgey, who joined through a Bloomberg mentorship program, drove the change across the codebase.

The single docker-compose.yaml is also gone. It has been replaced by a compose.yaml split across several files, so users can run anything from the full stack to leaner setups. Costa said the demo is the most forked project in the OpenTelemetry organization, with 6,859 forks, and the new layout lets those forks layer their changes on top instead of carrying full copies of the config. The project asked all 45 vendors listed in its README to update their setup instructions and said it will drop unchecked entries 60 days after the release.

What’s new

The headline addition is a three-part AI stack: a LangGraph ReAct agent that handles multi-turn requests and picks tools, a Model Context Protocol server that exposes demo features as tools, and a chatbot UI. Every tool call, LLM interaction and reasoning step is traced from the chatbot through the agent and MCP server to the other microservices, and a Collector processor converts Traceloop/OpenLLMetry telemetry into the official GenAI semantic conventions.

Version 3.0 also adds continuous profiling with the firepit backend, so traces, metrics, logs and profiles share one pipeline. An OpAMP server shows the Collector reporting its version and active configuration over WebSocket, viewable at localhost:8080/opamp, as groundwork for remote configuration later.

Other changes mirror messier production setups. The Ad service now exposes a Prometheus endpoint scraped by the Collector, and the Product Catalog service uses SQLCommenter to embed trace context in PostgreSQL queries. The Locust load generator was replaced with k6 and a custom xk6-otel extension, cutting the memory limit from 1,500 MiB to 512 MiB.

A new telemetry sanity-testing framework checks traces in Jaeger, metrics in Prometheus and logs in OpenSearch. Costa credited it with giving maintainers the confidence to merge a backlog of dependency updates that Piotr Kiełkowicz started working through, which took the demo’s OpenSSF Scorecard from 62 flagged vulnerabilities to single digits and resolved more than 300 CVEs.

Subscribe for Updates

ObservabilityTrends Newsletter

News and updates for oberservability-driven developers and professionals.

By signing up for our newsletter you agree to receive content related to ientry.com / webpronews.com and our affiliate partners. For additional information refer to our terms of service.

Notice an error?

Help us improve our content by reporting any issues you find.

Get the WebProNews newsletter delivered to your inbox

Get the free daily newsletter read by decision makers

Subscribe
Advertise with Us

Ready to get started?

Get our media kit

Advertise with Us