Logging in Python: A Comparison of the Top 6 Libraries

While Python offers a robust and feature-rich logging solution within its standard library, the third-party logging ecosystem presents a compelling array of alternatives. Depending on your requirements, these external libraries might be more suitable for your logging needs.

Therefore, this article will consider Python's top six logging solutions for tracking application and library behaviour. We will begin with a discussion of the standard logging module, then examine five other logging frameworks created by the Python community.

Let's get started!

1. The standard logging module

Python distinguishes itself from most programming languages by including a fully-featured logging framework in its standard library. This logging solution effectively caters to the needs of both library and application developers and it incorporates the following severity levels: DEBUG, INFO, WARNING, ERROR, and CRITICAL. Thanks to the default logger, you can immediately begin logging without any preliminary setup:

import logging

logging.debug("A debug message")
logging.info("An info message")
logging.warning("A warning message")
logging.error("An error message")
logging.critical("A critical message")

This default (or root) logger operates at the WARNING level, meaning that only logging calls whose severity equals or exceeds WARNING will produce an output:

WARNING:root:A warning message
ERROR:root:An error message
CRITICAL:root:A critical message

This configuration ensures that only potentially important messages are shown, reducing the noise in the log output. However, you can customize the log level and fine-tune the logging behavior as needed. The recommended way to use the logging module involves creating a custom logger through the getLogger() method:

import logging

logger = logging.getLogger(__name__)

. . .

Once you have a custom logger, you can customize its output through the Handler, Formatter, and Filter classes provided by the logging module.

• Handlers decide the output destination and can be customized based on the log level. Multiple handlers can also be added to a logger to simultaneously send log messages to different destinations.

• Formatters determine the format of the records produced by a logger. However, there are no predefined formats like JSON, Logfmt, etc. You have to combine the available log record attributes to build your own formats. The default format for the root logger is %(levelname)s:%(name)s:%(message)s. However, custom loggers default to just %(message)s.

• Filters are used by handler and logger objects to filter log records. They provide greater control than log levels over which log records should be processed and ignored. They also allow you to enhance or modify the records somehow before the logs are sent to their final destination. For example, you can create a custom filter that redacts sensitive data in your logs.

Here's an example that logs to the console and a file using a custom logger:

import sys
import logging

logger = logging.getLogger("example")
logger.setLevel(logging.DEBUG)

# Create handlers for logging to the standard output and a file
stdoutHandler = logging.StreamHandler(stream=sys.stdout)
errHandler = logging.FileHandler("error.log")

# Set the log levels on the handlers
stdoutHandler.setLevel(logging.DEBUG)
errHandler.setLevel(logging.ERROR)

# Create a log format using Log Record attributes
fmt = logging.Formatter(
    "%(name)s: %(asctime)s | %(levelname)s | %(filename)s:%(lineno)s | %(process)d >>> %(message)s"
)

# Set the log format on each handler
stdoutHandler.setFormatter(fmt)
errHandler.setFormatter(fmt)

# Add each handler to the Logger object
logger.addHandler(stdoutHandler)
logger.addHandler(errHandler)

logger.info("Server started listening on port 8080")
logger.warning(
    "Disk space on drive '/var/log' is running low. Consider freeing up space"
)

try:
    raise Exception("Failed to connect to database: 'my_db'")
except Exception as e:
    # exc_info=True ensures that a Traceback is included
    logger.error(e, exc_info=True)

When you execute the above program, the following log messages are printed to the console as expected:

example: 2023-07-23 14:42:18,599 | INFO | main.py:30 | 187901 >>> Server started listening on port 8080
example: 2023-07-23 14:14:47,578 | WARNING | main.py:28 | 143936 >>> Disk space on drive '/var/log' is running low. Consider freeing up space
example: 2023-07-23 14:14:47,578 | ERROR | main.py:34 | 143936 >>> Failed to connect to database: 'my_db'
Traceback (most recent call last):
  File "/home/ayo/dev/betterstack/demo/python-logging/main.py", line 32, in <module>
    raise Exception("Failed to connect to database: 'my_db'")
Exception: Failed to connect to database: 'my_db'

The error.log file is also created, and it should contain the ERROR log alone since the minimum level on the errHandler was set to ERROR:

example: 2023-07-23 14:14:47,578 | ERROR | main.py:34 | 143936 >>> Failed to connect to database: 'my_db'
Traceback (most recent call last):
  File "/home/ayo/dev/betterstack/demo/python-logging/main.py", line 32, in <module>
    raise Exception("Failed to connect to database: 'my_db'")
Exception: Failed to connect to database: 'my_db'

At the time of writing, the logging module cannot produce structured logs unless you implement some additional code. Thankfully, there is an easier and better way to get structured output: the python-json-logger library.

pip install python-json-logger

Once installed, you may utilize it as follows:

import sys
import logging
from pythonjsonlogger import jsonlogger

. . .

# The desired Log Record attributes must be included here, and they can be
# renamed if necessary
fmt = jsonlogger.JsonFormatter(
    "%(name)s %(asctime)s %(levelname)s %(filename)s %(lineno)s %(process)d %(message)s",
    rename_fields={"levelname": "severity", "asctime": "timestamp"},
)

# Set the log format on each handler
stdoutHandler.setFormatter(fmt)
errHandler.setFormatter(fmt)

. . .

If you modify the previous example with the highlighted lines above, you will observe the following output upon execution:

{"name": "example", "filename": "main.py", "lineno": 31, "process": 179775, "message": "Server started listening on port 8080", "severity": "INFO", "timestamp": "2023-07-23 14:39:03,265"}
{"name": "example", "filename": "main.py", "lineno": 32, "process": 179775, "message": "Disk space on drive '/var/log' is running low. Consider freeing up space", "severity": "WARNING", "timestamp": "2023-07-23 14:39:03,265"}
{"name": "example", "filename": "main.py", "lineno": 38, "process": 179775, "message": "Failed to connect to database: 'my_db'", "exc_info": "Traceback (most recent call last):\n  File \"/home/ayo/dev/betterstack/demo/python-logging/main.py\", line 36, in <module>\n    raise Exception(\"Failed to connect to database: 'my_db'\")\nException: Failed to connect to database: 'my_db'", "severity": "ERROR", "timestamp": "2023-07-23 14:39:03,265"}

Contextual data can also be added at log point through the extra property on a level method like this:

logger.info(
    "Server started listening on port 8080",
    extra={"python_version": 3.10, "os": "linux", "host": "fedora 38"},
)

These extra fields are automatically inserted in the log record if they do not clash with any of the default attribute names. Otherwise, you'll get a KeyError exception.

{"name": "example", "filename": "main.py", "lineno": 31, "process": 195301, "message": "Server started listening on port 8080", "python_version": 3.1, "os": "linux", "host": "fedora 38", "severity": "INFO", "timestamp": "2023-07-23 14:45:42,472"}

As you can see, the built-in logging module is capable and extensible for various logging needs. However, its initial configuration and customization can be cumbersome since you have to create and configure loggers, handlers, and formatters before you can start logging effectively.

Please see our comprehensive Python logging guide and the official documentation to learn more about the logging module's features and best practices.

2. Loguru

Loguru is the most popular third-party logging framework for Python with over 15k GitHub stars at the time of writing. It aims to simplify the logging process by pre-configuring the logger and making it really easy to customize via its add() method. Initiating logging with Loguru is a breeze; just install the package and import it, then call one of its level methods as follows:

pip install loguru

from loguru import logger

logger.trace("Executing program")
logger.debug("Processing data...")
logger.info("Server started successfully.")
logger.success("Data processing completed successfully.")
logger.warning("Invalid configuration detected.")
logger.error("Failed to connect to the database.")
logger.critical("Unexpected system error occurred. Shutting down.")

The default configuration logs a semi-structured and colorized output to the standard error. It also defaults to DEBUG as its minimum level, which explains why the TRACE output isn't recorded.

Loguru's inner workings are easily customized through the add() function which handles everything from formatting the logs to setting their destination. For example, you can log to the standard output, change the default level to INFO, and format your logs as JSON using the configuration below:

from loguru import logger
import sys

logger.remove(0) # remove the default handler configuration
logger.add(sys.stdout, level="INFO", serialize=True)

. . .

{"text": "2023-07-17 15:26:21.597 | INFO     | __main__:<module>:9 - Server started successfully.\n", "record": {"elapsed": {"repr": "0:00:00.006401", "seconds": 0.006401}, "exception": null, "extra": {}, "file": {"name": "main.py", "path": "/home/ayo/dev/betterstack/demo/python-logging/main.py"}, "function": "<module>", "level": {"icon": "ℹ️", "name": "INFO", "no": 20}, "line": 9, "message": "Server started successfully.", "module": "main", "name": "__main__", "process": {"id": 3852028, "name": "MainProcess"}, "thread": {"id": 140653618894656, "name": "MainThread"}, "time": {"repr": "2023-07-17 15:26:21.597156+02:00", "timestamp": 1689600381.597156}}}

The default JSON output produced by Loguru can be quite verbose, but it's easy to serialize the log messages using a custom function like this:

from loguru import logger
import sys
import json

def serialize(record):
    subset = {
        "timestamp": record["time"].timestamp(),
        "message": record["message"],
        "level": record["level"].name,
        "file": record["file"].name,
        "context": record["extra"],
    }
    return json.dumps(subset)

def patching(record):
    record["extra"]["serialized"] = serialize(record)

logger.remove(0)

logger = logger.patch(patching)
logger.add(sys.stderr, format="{extra[serialized]}")

logger.bind(user_id="USR-1243", doc_id="DOC-2348").debug("Processing document")

{"timestamp": 1689601339.628792, "message": "Processing document", "level": "DEBUG", "file": "main.py", "context": {"user_id": "USR-1243", "doc_id": "DOC-2348"}}

Contextual logging is also fully supported in Loguru. You've already seen the bind() method above which allows the addition of contextual data at log point. You can also use it to create child loggers for logging records that share the same context:

child = logger.bind(user_id="USR-1243", doc_id="DOC-2348")
child.debug("Processing document")
child.warning("Invalid configuration detected. Falling back to defaults")
child.success("Document processed successfully")

Notice how the user_id and doc_id fields are present in all three records:

{"timestamp": 1689601518.884659, "message": "Processing document", "level": "DEBUG", "file": "main.py", "context": {"user_id": "USR-1243", "doc_id": "DOC-2348"}}
{"timestamp": 1689601518.884706, "message": "Invalid configuration detected. Falling back to defaults", "level": "WARNING", "file": "main.py", "context": {"user_id": "USR-1243", "doc_id": "DOC-2348"}}
{"timestamp": 1689601518.884729, "message": "Document processed successfully", "level": "SUCCESS", "file": "main.py", "context": {"user_id": "USR-1243", "doc_id": "DOC-2348"}}

On the other hand, its contextualize() method eases the addition of contextual fields to all log records within a specific scope or context. For example, the snippet below demonstrates adding a unique request ID attribute to all logs created as a result of that request:

from loguru import logger
import uuid

def logging_middleware(get_response):
    def middleware(request):
        request_id = str(uuid.uuid4())

        with logger.contextualize(request_id=request_id):
            response = get_response(request)
            response["X-Request-ID"] = request_id
            return response

    return middleware

Loguru also supports all the features you'd expect from a good logging framework such as logging to files with automatic rotation and compression, custom log levels, exception handling, logging to multiple destinations at once, and much more. It also provides a migration guide for users coming from the standard logging module.

Please see the official documentation and our dedicated Loguru guide and to learn more about using Loguru to create a production-ready logging setup for Python applications.

3. Structlog

Structlog is a logging library dedicated to producing structured output in JSON or Logfmt. It supports a colorized and aesthetically enhanced console output for development environments, but also allows for complete customization of the log format to meet diverse needs. You may install the Structlog package using the command below:

pip install structlog

The simplest possible usage of Structlog involves calling the get_logger() method and using any of the level methods on the resulting logger:

import structlog

logger = structlog.get_logger()

logger.debug("Database query executed in 0.025 seconds")
logger.info(
    "Processing file 'data.csv' completed. 1000 records were imported",
    file="data.csv",
    elapsed_ms=300,
    num_records=1000,
)
logger.warning(
    "Unable to load configuration file 'config.ini'. Using default settings instead",
    file="config.ini",
)

try:
    1 / 0
except ZeroDivisionError as e:
    logger.exception(
        "Division by zero error occurred during calculation. Check the input values",
        exc_info=e,
    )
logger.critical("Application crashed due to an unhandled exception")

The default configuration of a Structlog logger is quite friendly for development environments. The output is colorized, and any included contextual data is placed in key=value pairs. Additionally, tracebacks are neatly formatted and organized so that it's much easier to spot the cause of the issue.

A unique behavior of Structlog is that it doesn't filter records by their levels. This is why all the levels above were written to the console. However, it's easy enough to configure a default level through the configure() method like this:

import structlog
import logging

structlog.configure(wrapper_class=structlog.make_filtering_bound_logger(logging.INFO))

Structlog is compatible with the log levels in the standard logging module hence the usage of the logging.INFO constant above. You can also use the number associated with the level directly:

structlog.configure(wrapper_class=structlog.make_filtering_bound_logger(20))

The logger returned by the get_logger() function is called a Bound Logger because you can bind contextual values to it. Once the key/value pairs are bound, they will be included in each subsequent log entry produced by the logger.

import structlog
import platform

logger = structlog.get_logger()

logger = logger.bind(python_version=platform.python_version(), os="linux")

. . .

2023-07-23 17:20:10 [debug    ] Database query executed in 0.025 seconds os=linux python_version=3.11.4
2023-07-23 17:20:10 [info     ] Processing file 'data.csv' completed. 1000 records were imported elapsed_ms=300 file=data.csv num_records=1000 os=linux python_version=3.11.4
. . .

Bound loggers also include a chain of processor functions that transform and enrich log records as they pass through the logging pipeline. For example, you can log in JSON using the following configuration:

import structlog
import platform

structlog.configure(
    processors=[
        structlog.processors.TimeStamper(fmt="iso"),
        structlog.processors.add_log_level,
        structlog.processors.JSONRenderer(),
    ]
)

. . .

Each processor is executed in the declaration order, so TimeStamper() is called first to add an ISO-8601 formatted timestamp to each entry, then the severity level is added through add_log_level, and finally the entire record is serialized as JSON by calling JSONRenderer(). You will observe the following output after making the highlighted modifications to the program:

{"python_version": "3.11.4", "os": "linux", "event": "Database query executed in 0.025 seconds", "timestamp": "2023-07-23T15:32:21.590688Z", "level": "debug"}
{"python_version": "3.11.4", "os": "linux", "file": "data.csv", "elapsed_ms": 300, "num_records": 1000, "event": "Processing file 'data.csv' completed. 1000 records were imported", "timestamp": "2023-07-23T15:32:21.590720Z", "level": "info"}
. . .

Another cool thing Structlog can do is automatically format tracebacks so that they are also serialized in JSON format. You only need to use the dict_tracebacks processor like this:

structlog.configure(
    processors=[
        structlog.processors.TimeStamper(fmt="iso"),
        structlog.processors.add_log_level,
        structlog.processors.dict_tracebacks,
        structlog.processors.JSONRenderer(),
    ]
)

Whenever exceptions are logged, you will observe that the records are enriched with well-formatted information about the exception making it easy to analyze in a log management service.

{"python_version": "3.11.4", "os": "linux", "event": "Division by zero error occurred during calculation. Check the input values", "timestamp": "2023-07-23T16:07:50.127241Z", "level": "error", "exception": [{"exc_type": "ZeroDivisionError", "exc_value": "division by zero", "syntax_error": null, "is_cause": false, "frames": [{"filename": "/home/ayo/dev/betterstack/demo/python-logging/main.py", "lineno": 32, "name": "<module>", "line": "", "locals": {"__name__": "__main__", "__doc__": "None", "__package__": "None", "__loader__": "<_frozen_importlib_external.SourceFileLoader object at 0x7fdb22df2ed0>", "__spec__": "None", "__annotations__": "{}", "__builtins__": "<module 'builtins' (built-in)>", "__file__": "/home/ayo/dev/betterstack/demo/python-logging/main.py", "__cached__": "None", "structlog": "\"<module 'structlog' from '/home/ayo/.local/lib/python3.11/site-packages/structlo\"+15", "platform": "<module 'platform' from '/usr/lib64/python3.11/platform.py'>", "logging": "<module 'logging' from '/usr/lib64/python3.11/logging/__init__.py'>", "logger": "\"<BoundLoggerFilteringAtDebug(context={'python_version': '3.11.4', 'os': 'linux'}\"+249", "e": "ZeroDivisionError('division by zero')"}}]}]}

I've only scratched the surface of what Structlog has to offer, so ensure to check out its documentation to learn more.

4. Eliot

Eliot is a unique Python logging solution that aims not only to provide a record of an event that occurred in the program, but also outputs a causal chain of actions leading to the event. You can install Eliot with pip as follows:

pip install eliot

One of Eliot's key concepts is an action which represents any task that can start and finish successfully or fail with an exception. When you start an action, two log records are produced: one to indicate the start of the action, and the other to indicate its success or failure. The best way to demonstrate this model is through an example:

import sys
from eliot import start_action, to_file

to_file(sys.stdout)


def calculate(x, y):
    with start_action(action_type="multiply"):
        return x * y


calculate(10, 5)

The start_action function is used here to indicate the start of a new action. Once the calculate() function is executed, two logs are sent to the destination configured by to_file():

{"action_status": "started", "timestamp": 1690213156.7701144, "task_uuid": "a9a47808-15a9-439b-8335-b88d50013f75", "action_type": "multiply", "task_level": [1]}
{"action_status": "succeeded", "timestamp": 1690213156.7701554, "task_uuid": "a9a47808-15a9-439b-8335-b88d50013f75", "action_type": "multiply", "task_level": [2]}

Eliot produces structured JSON output by default, and the following records are included:

• task_uuid: The unique task identifier that produced the message.
• action_status: Indicates the status of the action.
• timestamp: The UNIX timestamp of the message.
• task_level: The location of the message within the task's tree of actions.
• action_type: The provided action_type argument.

You can add additional fields to both the start message and the success message of an action like this:

def calculate(x, y):
    # additional fields here are added to the start message of the action alone
    with start_action(action_type="multiply", x=x, y=y) as action:
        result = x * y
        # fields added here show up only in the success message of the action
        action.add_success_fields(result=result)
        return result

{"x": 10, "y": 5, "action_status": "started", "timestamp": 1690213820.4083755, "task_uuid": "09df3632-96d2-4dd8-b782-1926cd87ccc9", "action_type": "multiply", "task_level": [1]}
{"result": 50, "action_status": "succeeded", "timestamp": 1690213820.4084144, "task_uuid": "09df3632-96d2-4dd8-b782-1926cd87ccc9", "action_type": "multiply", "task_level": [2]}

Another way to log the inputs and results of a function is through the log_call decorator:

from eliot import log_call, to_file

to_file(sys.stdout)


@log_call
def calculate(x, y):
    return x * y


calculate(10, 5)

In this case, the action_type will be a concatenation of the module and function name, but the remaining fields will be the same as before:

{"x": 10, "y": 5, "action_status": "started", "timestamp": 1690214038.799868, "task_uuid": "2c78b304-12a1-474a-8b95-e80deadb8dde", "action_type": "__main__.calculate", "task_level": [1]}
{"result": 50, "action_status": "succeeded", "timestamp": 1690214038.7999015, "task_uuid": "2c78b304-12a1-474a-8b95-e80deadb8dde", "action_type": "__main__.calculate", "task_level": [2]}

You can customize the behavior of the log_call decorator by changing the action_type field, and excluding certain arguments or the result:

@log_call(action_type="CALC", include_args=["x"], include_result=False)

If an uncaught exception is detected within the context of an action, the action will be marked as failed, and an exception message will be logged instead of a success message:

import sys
from eliot import log_call, to_file

to_file(sys.stdout)


@log_call
def calculate(x, y):
    return x / y


try:
    calculate(1, 0)
except ZeroDivisionError as e:
    print("division by zero detected")

Instead of a success message, you'll now observe an exception message accompanied with a reason:

{"x": 1, "y": 0, "action_status": "started", "timestamp": 1690215830.1103916, "task_uuid": "f267b0f5-8c07-4828-a973-0a8a273f272d", "action_type": "__main__.calculate", "task_level": [1]}
{"exception": "builtins.ZeroDivisionError", "reason": "division by zero", "action_status": "failed", "timestamp": 1690215830.1104264, "task_uuid": "f267b0f5-8c07-4828-a973-0a8a273f272d", "action_type": "__main__.calculate", "task_level": [2]}

When you need to log isolated messages within the context of an action, you can use the log method as follows:

def calculate(x, y):
    with start_action(action_type="multiply") as ctx:
        ctx.log(message_type="mymsg", msg="a standalone message")
        return x * y

{"msg": "a standalone message", "timestamp": 1690217318.2063951, "task_uuid": "500b06e6-c0ba-42b4-9d6c-466ea3f1634d", "task_level": [2], "message_type": "mymsg"}

Eliot does not have the concept of log levels, so you can only add the level field manually if needed:

def calculate(x, y):
    with start_action(action_type="multiply", level="INFO") as ctx:
        ctx.log(message_type="mymsg", msg="a standalone message", level="INFO")
        return x * y

Another neat feature of Eliot is its ability to visualize its logs through the eliot-tree command-line tool.

pip install eliot-tree

Once you've installed eliot-tree, you can pipe the JSON logs produced by Eliot to the command as follows:

python main.py | eliot-tree

If you're logging to a file, you can pass the file as an argument to the tool:

eliot-tree <file>

There is so much more to Eliot than can be covered here so ensure to check out its documentation to learn more.

5. Logbook

Logbook describes itself as a cool replacement for Python's standard library logging module, whose aim is to make logging fun. You can install it in your project using the following command:

pip install logbook

Getting started with Logbook is also really straightforward:

import sys
import logbook

logger = logbook.Logger(__name__)

handler = logbook.StreamHandler(sys.stdout, level="INFO")
handler.push_application()

logger.info("Successfully connected to the database 'my_db' on host 'ubuntu'")

logger.warning("Detected suspicious activity from IP address: 111.222.333.444")

[2023-07-24 21:41:50.932575] INFO: __main__: Successfully connected to the database 'my_db' on host 'ubuntu'
[2023-07-24 21:41:50.932623] WARNING: __main__: Detected suspicious activity from IP address: 111.222.333.444

As shown above, the logbook.Logger method is used to create a new logger channel.. This logger provides access to level methods like info() and warning() for writing log messages. All the log levels in the logging module are supported, with the addition of NOTICE level, which sits between INFO and WARNING.

Logbook also uses the Handler concept to determine the destination and formatting of the logs. The StreamHandler class sends logs to any output stream (the standard output in this case), and other handlers are available for logging to files, Syslog, Redis, Slack etc.

However, unlike the standard logging module, you are discouraged from registering handlers on the logger directly. Instead, you're supposed to bind the handler to the process, thread, or greenlet stack through the push_application(), push_thread(), and push_greenlet() methods respectively. The corresponding pop_application(), pop_thread(), and pop_greenlet() methods also exist for unregistering handlers:

handler = MyHandler()
handler.push_application()
# everything logged here here goes to that handler
handler.pop_application()

You can also bind a handler for the duration of a with-block. This ensures that logs created within the block are sent only to the specified handler:

with handler.applicationbound():
    logger.info(...)

with handler.threadbound():
    logger.info(...)

with handler.greenletbound():
    logger.info(...)

Log formatting is also done through handlers. A format_string property exists on each handler for this purpose, and it accepts properties on the LogRecord class:

import sys
import logbook

logger = logbook.Logger(__name__)

handler = logbook.StreamHandler(sys.stdout, level="INFO")
handler.format_string = "{record.channel} | {record.level_name} | {record.message}"
handler.push_application()

logger.info("Successfully connected to the database 'my_db' on host 'ubuntu'")

logger.warning("Detected suspicious activity from IP address: 111.222.333.444")

__main__ | INFO | Successfully connected to the database 'my_db' on host 'ubuntu'
__main__ | WARNING | Detected suspicious activity from IP address: 111.222.333.444

Unfortunately, structured logging isn't supported in any of Logbook's built-in handlers. You'd have to implement it yourself via a custom handler. For more details, see the Logbook documentation.

6. Picologging

Mirosoft's Picologging library is a relatively new addition to Python's logging ecosystem. Positioned as a high-performance drop-in replacement for the standard logging module, it boasts a remarkable 4-10 times speed improvement, as stated in its GitHub Readme. To integrate it into your project, you can install it with the following command:

pip install picologging

Picologging shares the same familiar API as the logging module in Python and it uses the same log record attributes for formatting:

import sys
import picologging as logging

logger = logging.getLogger(__name__)
logger.setLevel(logging.INFO)

stdout_handler = logging.StreamHandler(sys.stdout)
fmt = logging.Formatter(
    "%(name)s: %(asctime)s | %(levelname)s | %(process)d >>> %(message)s"
)

stdout_handler.setFormatter(fmt)
logger.addHandler(stdout_handler)

logger.info(
    "Successfully connected to the database '%s' on host '%s'", "my_db", "ubuntu20.04"
)

logger.warning("Detected suspicious activity from IP address: %s", "111.222.333.444")

__main__: 2023-07-24 05:46:38,-2046715687 | INFO | 795975 >>> Successfully connected to the database 'my_db' on host 'ubuntu20.04'
__main__: 2023-07-24 05:46:38,-2046715687 | WARNING | 795975 >>> Detected suspicious activity from IP address: 111.222.333.444

Picologging's documentation emphasizes that it is currently in an early-alpha state, so you should hold off on using it in production. Nevertheless, it is already showing some promise when it comes to performance improvements to the standard logging module according to these benchmarks. Please see the documentation for more information on its features and limitations.

Final thoughts

Our primary recommendation for logging in Python is to use Loguru due to its impressive features and user-friendly API. However, it's crucial to familiarize yourself with the built-in logging module, as it remains a powerful and widely used solution.

Structlog is another robust option that merits consideration, and Eliot can also be a good choice, provided its lack of log levels isn't a significant concern for your use case. On the other hand, Picologging is currently in its early development stages, and Logbook lacks native support for structured logging, making them less advisable for logging in production.

Thanks for reading, and happy logging!

Article by

Ayooluwa Isaiah

Ayo is the Head of Content at Better Stack. His passion is simplifying and communicating complex technical ideas effectively. His work was featured on several esteemed publications including LWN.net, Digital Ocean, and CSS-Tricks. When he’s not writing or coding, he loves to travel, bike, and play tennis.

Got an article suggestion? Let us know

Next article

Logging in Go with Slog: The Ultimate Guide

Learn everything about the new Slog package which brings high-performance structured, and leveled logging to the Go standard library

→

This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.