Auto-instrumentation: No-code approach to instrument Python

In the previous post we saw instrumenting a python application with OTel emitting metrics and traces along the way. While this is good, it can still be intimidating to an instrumentation-newbie.
To get over this barrier and to allow a quick-start, some languages have auto-instrumentation support. One only needs to install some libraries to get going.

Our sample web-app

import datetime
import flask

######################
## initialization
######################
app = flask.Flask(__name__)
start = datetime.datetime.now()

######################
## routes
######################
@app.route('/', methods=['GET'])
def root():
  return flask.jsonify({'message': 'flask app root/'})

@app.route('/healthz', methods=['GET'])
def healthz():
  now = datetime.datetime.now()
  return flask.jsonify({'message': f'up and running since {(now - start)}'})

######################
if __name__ == '__main__':
######################
  app.run(debug=True, host='0.0.0.0', port=5000)

Install necessary libraries

$ pip install flask
$ pip install opentelemetry-distro opentelemetry-instrumentation-flask

Note: some corner case made my install opentelemetry-api as well, but is not needed as per official documentation.

Initialize (to install extra libraries as needed)
$ opentelemetry-bootstrap -a install

Run your code

$ opentelemetry-instrument --traces_exporter console --metrics_exporter console flask run
 * Serving Flask app 'app'
 * Debug mode: on
WARNING: This is a development server. Do not use it in a production deployment. Use a production WSGI server instead.
 * Running on all addresses (0.0.0.0)
 * Running on http://127.0.0.1:5000
 * Running on http://192.168.1.7:5000
Press CTRL+C to quit
 * Restarting with stat
 * Debugger is active!
 * Debugger PIN: 121-673-590
127.0.0.1 - - [17/Oct/2022 07:21:05] "GET /healthz HTTP/1.1" 200 -
{
    "name": "/healthz",
    "context": {
        "trace_id": "0xd0850752865577d2d8cd11aaef169574",
        "span_id": "0x29c8ad5fd974de41",
        "trace_state": "[]"
    },
    "kind": "SpanKind.SERVER",
    "parent_id": null,
    "start_time": "2022-10-17T01:52:45.522806Z",
    "end_time": "2022-10-17T01:52:45.523615Z",
    "status": {
        "status_code": "UNSET"
    },
    "attributes": {
        "http.method": "GET",
        "http.server_name": "127.0.0.1",
        "http.scheme": "http",
        "net.host.port": 5000,
        "http.host": "localhost:5000",
        "http.target": "/healthz",
        "net.peer.ip": "127.0.0.1",
        "http.user_agent": "curl/7.79.1",
        "net.peer.port": 55838,
        "http.flavor": "1.1",
        "http.route": "/healthz",
        "http.status_code": 200
    },
    "events": [],
    "links": [],
    "resource": {
        "attributes": {
            "telemetry.sdk.language": "python",
            "telemetry.sdk.name": "opentelemetry",
            "telemetry.sdk.version": "1.13.0",
            "telemetry.auto.version": "0.34b0",
            "service.name": "unknown_service"
        },
        "schema_url": ""
    }
}
{"resource_metrics": [{"resource": {"attributes": {"telemetry.sdk.language": "python", "telemetry.sdk.name": "opentelemetry", "telemetry.sdk.version": "1.13.0", "telemetry.auto.version": "0.34b0", "service.name": "unknown_service"}, "schema_url": ""}, "scope_metrics": [], "schema_url": ""}]}

As of this post, most popular frameworks like Django, FastAPI, Flask have instrumentation libraries for HTTP context propagation

Code size implications

Auto-instrumentation does add some extra libraries. This was the result in my case

$ du -sh manual/venv/ auto/venv/
 29M    manual/venv/
 30M    auto/venv/

---

Always refer to the official documentation which is up-to-date
Official documentation