Why Isn’t There a Python Auto-Instrumentor for OpenTelemetry That Covers Every Function?

If you want to analyze time spent in different code parts, using a profiler might be a better bet. Profiler libraries can help you monitor performance without the clutter that comes from excessive instrumentation. Just keep in mind that running these profilers often requires isolated instances due to their performance overhead.

Creating a span for every function might not be the best approach because it could lead to an overwhelming amount of trace data. The existing instrumentors like FastAPI focus on the key parts of the application that matter most, helping you to get meaningful insights. Instead of automatic spans for every function, try instrumenting crucial parts of your app, like file reads, conversions, and writes. This way, you get valuable data without drowning in noise.

There are actually tools that can achieve similar outcomes, but they often come with a cost. For example, Pydantic Logfire offers a good solution with free access to the first 10 million spans, which is pretty generous! If you need real-time insights without excessive noise, it’s worth looking into.

OpenTelemetry is still developing its standards and APIs, focusing on specific frameworks. They suggest using specialized instrumentation libraries for your API framework. While the idea of a generic auto-instrumentor for all functions sounds promising, it would create significant noise and may not provide usable metrics, which is why the community hasn't adopted it yet.

You might want to reconsider the blanket approach. Instrumentation is typically set up for specific functions where you anticipate needing data. For instance, you could just use a profiler for deep insights and keep it focused rather than trying to instrument every single function.