Flask structured logging tests

Run with python -m flask run

Setting the logging framework

Use the LOG_FRAMEWORK env-var to on of

• json_logging
• loguru
• structlog

Debug mode

Enabling debug mode will turn off structured logging to improve developer experience. It will thus log normally and you won't have to parse JSON while developing.

Evaluation

Time of writing 2023-09-22.

	json_logging	loguru	structlog
uses stdlib logging	yes	no	no
last commit	2022-11-5	2023-09-11	2023-09-19
maintained	???	yes	yes
Structured logging	yes	yes	yes
OOB Correlation-ID logging	yes	no	no
Correlation-ID generation	yes	no	no
Logger naming	yes	yes	maybe?
optional structured logging	yes	yes	yes

Optional requirements

	json_logging	loguru	structlog
log colors	no	yes	yes

json_logging

Is the easiest to use with Flask. The only downside is that it's not clear whether it's still maintained.

Example log output

{
  "written_at": "2023-09-22T08:13:52.968Z",
  "written_ts": 1695370432968949000,
  "type": "request",
  "correlation_id": "f7b3e876-591f-11ee-832c-98597af1829f",
  "remote_user": "-",
  "request": "/",
  "referer": "-",
  "x_forwarded_for": "-",
  "protocol": "HTTP/1.1",
  "method": "GET",
  "remote_ip": "127.0.0.1",
  "request_size_b": -1,
  "remote_host": "127.0.0.1",
  "remote_port": 39438,
  "request_received_at": "2023-09-22T08:13:52.968Z",
  "response_time_ms": 0,
  "response_status": 200,
  "response_size_b": 12,
  "response_content_type": "text/html; charset=utf-8",
  "response_sent_at": "2023-09-22T08:13:52.968Z"
}

loguru

A good logger that works out of the box on new projects without any setup. Its selling point is that you use a singler logger everywhere. Import it and call log.info or whatever and you're done.

However, it is its own logging framework and doesn't use the logging from stdlib. That means custom code has to be written to handle libraries that do log using the stdlib.

Log level configuration involves adding a logger with a filter at the desired level. One then has to remember to remove other loggers, otherwise there are multiple loggers. It isn't as simple as stdlib logger.setLevel() and you're done.

Example log output

{
  "text": "2023-09-22 10:12:58.082 | INFO     | endpoints.hello:say_hello:8 - about to say hello\n",
  "record": {
    "elapsed": {
      "repr": "0:00:12.520662",
      "seconds": 12.520662
    },
    "exception": null,
    "extra": {},
    "file": {
      "name": "hello.py",
      "path": "/home/michael/projects/other/flaskProject/endpoints/hello.py"
    },
    "function": "say_hello",
    "level": {
      "icon": "ℹ️",
      "name": "INFO",
      "no": 20
    },
    "line": 8,
    "message": "about to say hello",
    "module": "hello",
    "name": "endpoints.hello",
    "process": {
      "id": 234206,
      "name": "MainProcess"
    },
    "thread": {
      "id": 139874679666368,
      "name": "Thread-1 (process_request_thread)"
    },
    "time": {
      "repr": "2023-09-22 10:12:58.082316+02:00",
      "timestamp": 1695370378.082316
    }
  }
}

structlog

Another self-rolled logging framework, which comes with the same caveats as loguru. It doesn't use the "one to rule them all" approach though.

It takes more fiddling with the formatter for stdlib than with loguru and I haven't figured it out yet. The logs are minimal.

Example log output