#tracing #pyo3 #open-telemetry #python-module #python-bindings #python

pyo3-tracing-subscriber

A Python module for configuring and initializing tracing subscribers from Python

8 releases

0.2.0-rc.0 Nov 14, 2023
0.1.2-rc.0 Mar 2, 2024
0.1.1 Dec 18, 2023
0.1.0 Dec 16, 2023
0.1.0-rc.1 Nov 30, 2023

#264 in Debugging

Apache-2.0

110KB
1.5K SLoC

PyO3 Tracing Subscriber

Background

What this is

pyo3_tracing_subscriber provides a PyModule that can be added to upstream pyo3 extension modules in order to support the configuration and initialization of Rust tracing subscribers from Python.

What this is not

  • Any initialized tracing subscriber imported from your upstream package will not collect traces from any other pyo3 extension module. In other words, any pyo3 extension module will need to separately export tracing configuration and context managers, which in turn must be separately initialized in order to capture Rust traces from respective pyo3 extension modules.
  • Currently, only three tracing subcriber layers are supported:
    • tracing_subscriber::fmt which writes traces to file (or stdout) in a human readable format.
    • opentelemetry-stdout which writes traces to file (or stdout) in OTLP format. Available only with the layer-otel-otlp-file feature.
    • opentelemetry-otlp which sends traces to an OpenTelemetry OTLP endpoint. Available only with the layer-otel-otlp feature.
  • This does not propagate OpenTelemetry contexts from Python into Rust (or vice versa). Use the pyo3-opentelemetry crate for that feature.

Usage

For a complete functioning example, see the examples/pyo3-opentelemetry-lib/src/lib.rs example within this crate's repository.

Given a pyo3 extension module named "my_module" that would like to expose the tracing subscriber configuration and context manager classes from "my_module._tracing_subscriber", from Rust:

use pyo3::prelude::*;

#[pymodule]
fn my_module(py: Python, m: &PyModule) -> PyResult<()> {
    // Add your own Python classes, functions and modules.

    let tracing_subscriber = PyModule::new(py, "_tracing_subscriber")?;
    pyo3_tracing_subscriber::add_submodule(
        "my_module._tracing_subscriber",
        py,
        tracing_subscriber,
    )?;
    m.add_submodule(tracing_subscriber)?;
    Ok(())
}

Then a user could initialize a tracing subscriber that logged to stdout from Python:

import my_module
from my_module._tracing_subscriber import (
    GlobalTracingConfig,
    SimpleConfig,
    Tracing,
    subscriber,
)
from pyo3_opentelemetry_lib._tracing_subscriber.layers import file


def main():
    tracing_configuration = GlobalTracingConfig(
        export_process=SimpleConfig(
            subscriber=subscriber.Config(
                layer=file.Config()
            )
        )
    )
    with Tracing(config=config):
        result = my_module.example_function()
        my_module.other_example_function(result)

if __name__ == '__main__':
    main()

Building Python Stub Files

This crate provides a convenient method for adding stub files to your Python source code with the stubs feature.

Given a pyo3 extension module named "my_module" that uses the pyo3-tracing-subscriber crate to expose tracing subscriber configuration and context manager classes from "my_module._tracing_subscriber", in the upstream build.rs file:

use pyo3_tracing_subscriber_stubs::write_stub_files;

fn main() {
    let target_dir = std::path::Path::new("./my_module/_tracing_subscriber");
    std::fs::remove_dir_all(target_dir).unwrap();
    write_stub_files(
        "my_module",
        "_tracing_subscriber",
        target_dir,
        true, // layer_otel_otlp_file feature enabled
        true, // layer_otel_otlp feature enabled
    )
    .unwrap();
}

Dependencies

~16–25MB
~351K SLoC