class Logger[source]

An object to dispatch logging messages to configured handlers.

The Logger is the core objet of loguru, every logging configuration and usage pass through a call to one of its methods. There is only one logger, so there is no need to retrieve one before usage.

Handlers to which send log messages are added using the start() method. Note that you can use the Logger right after import as it comes pre-configured. Messages can be logged with different severity levels and using braces attributes like the str.format() method do.

Once a message is logged, a “record” is associated with it. This record is a dict wich contains several information about the logging context: time, function, file, line, thread, level… It also contains the __name__ of the module, this is why you don’t need named loggers.

You should not instantiate a Logger by yourself, use from loguru import logger instead.

start(sink, *, level='DEBUG', format='<green>{time:YYYY-MM-DD HH:mm:ss.SSS}</green> | <level>{level: <8}</level> | <cyan>{name}</cyan>:<cyan>{function}</cyan>:<cyan>{line}</cyan> - <level>{message}</level>', filter=None, colorize=None, serialize=False, backtrace=True, enqueue=False, catch=True, **kwargs)[source]

Start sending log messages to a sink adequately configured.

  • sink (file-like object, str, pathlib.Path, function, logging.Handler or class) – An object in charge of receiving formatted logging messages and propagating them to an appropriate endpoint.
  • level (int or str, optional) – The minimum severity level from which logged messages should be send to the sink.
  • format (str or function, optional) – The template used to format logged messages before being sent to the sink.
  • filter (function or str, optional) – A directive used to optionally filter out logged messages before they are send to the sink.
  • colorize (bool, optional) – Whether or not the color markups contained in the formatted message should be converted to ansi codes for terminal coloration, ore stripped otherwise. If None, the choice is automatically made based on the sink being a tty or not.
  • serialize (bool, optional) – Whether or not the logged message and its records should be first converted to a JSON string before being sent to the sink.
  • backtrace (bool, optional) – Whether or not the formatted exception should use stack trace to display local variables values. This probably should be set to False in production to avoid leaking sensitive data.
  • enqueue (bool, optional) – Whether or not the messages to be logged should first pass through a multiprocess-safe queue before reaching the sink. This is useful while logging to a file through multiple processes.
  • catch (bool, optional) – Whether or not errors occuring while sink handles logs messages should be caught or not. If True, an exception message is displayed on sys.stderr but the exception is not propagated to the caller, preventing sink from stopping working.
  • **kwargs – Additional parameters that will be passed to the sink while creating it or while logging messages (the exact behavior depends on the sink type).

If and only if the sink is a file, the following parameters apply:

  • rotation (str, int, datetime.time, datetime.timedelta or function, optional) – A condition indicating whenever the current logged file should be closed and a new one started.
  • retention (str, int, datetime.timedelta or function, optional) – A directive filtering old files that should be removed during rotation or end of program.
  • compression (str or function, optional) – A compression or archive format to which log files should be converted at closure.
  • delay (bool, optional) – Whether or not the file should be created as soon as the sink is configured, or delayed until first logged message. It defaults to False.
  • mode (str, optional) – The openning mode as for built-in open() function. It defaults to "a" (open the file in appending mode).
  • buffering (int, optional) – The buffering policy as for built-in open() function. It defaults to 1 (line buffered file).
  • encoding (str, optional) – The file encoding as for built-in open() function. If None, it defaults to locale.getpreferredencoding().
  • **kwargs – Others parameters are passed to the built-in open() function.

int – An identifier associated with the starteds sink and which should be used to stop() it.


Extended summary follows.

The sink parameter

The sink handles incomming log messages and proceed to their writing somewhere and somehow. A sink can take many forms:

  • A file-like object like sys.stderr or open("somefile.log", "w"). Anything with a .write() method is considered as a file-like object. If it has a .flush() method, it will be automatically called after each logged message. If it has a .stop() method, it will be automatically called at sink termination.
  • A file path as str or pathlib.Path. It can be parametrized with some additional parameters, see bellow.
  • A simple function like lambda msg: print(msg). This allows for logging procedure entirely defined by user preferences and needs.
  • A built-in logging.Handler like logging.StreamHandler. In such a case, the Loguru records are automatically converted to the structure expected by the logging module.
  • A class object that will be used to instantiate the sink using **kwargs attributes passed. Hence the class should instantiate objects which are therefore valid sinks.

The logged message

The logged message passed to all started sinks is nothing more than a string of the formatted log, to which a special attribute is associated: the .record which is a dict containing all contextual information possibly needed (see bellow).

Logged messages are formatted according to the format of the started sink. This format is usually a string containing braces fields to display attributes from the record dict.

If fine-grained control is needed, the format can also be a function which takes the record as parameter and return the format template string. However, note that in such a case, you should take care of appending the line ending and exception field to the returned format, while "\n{exception}" is automatically appended for convenience if format is a string.

The filter attribute can be used to control which messages are effectively passed to the sink and which one are ignored. A function can be used, accepting the record as an argument, and returning True if the message should be logged, False otherwise. If a string is used, only the records with the same name and its children will be allowed.

The record dict

The record is just a Python dict, accessible from sinks by message.record, and usable for formatting as "{key}". Some record’s values are objects with two or more attibutes, those can be formatted with "{key.attr}" ("{key}" would display one by default). Formatting directives like "{key: >3}" also works and is specially useful for time (see bellow).

Key Description Attributes
elapsed The time elapsed since the start of the program See datetime.timedelta
exception The formatted exception if any, None otherwise type, value, traceback
extra The dict of attributes bound by the user None
file The file where the logging call was made name (default), path
function The function from which the logging call was made None
level The severity used to log the the message name (default), no, icon
line The line number in the source code None
message The logged message (not yet formatted) None
module The module where the logging call was made None
name The __name__ where the logging call was made None
process The process in which the logging call was made name, id (default)
thread The thread in which the logging call was made name, id (default)
time The local time when the logging call was made See datetime.datetime

The time formatting

The time field can be formatted using more human-friendly tokens. Those constitute a subset of the one used by the Pendulum library by @sdispater. To escape a token, just add square brackets around it.

  Token Output
Year YYYY 2000, 2001, 2002 … 2012, 2013
YY 00, 01, 02 … 12, 13
Quarter Q 1 2 3 4
Month MMMM January, February, March …
MMM Jan, Feb, Mar …
MM 01, 02, 03 … 11, 12
M 1, 2, 3 … 11, 12
Day of Year DDDD 001, 002, 003 … 364, 365
DDD 1, 2, 3 … 364, 365
Day of Month DD 01, 02, 03 … 30, 31
D 1, 2, 3 … 30, 31
Day of Week dddd Monday, Tuesday, Wednesday …
ddd Mon, Tue, Wed …
d 0, 1, 2 … 6
Days of ISO Week E 1, 2, 3 … 7
Hour HH 00, 01, 02 … 23, 24
H 0, 1, 2 … 23, 24
hh 01, 02, 03 … 11, 12
h 1, 2, 3 … 11, 12
Minute mm 00, 01, 02 … 58, 59
m 0, 1, 2 … 58, 59
Second ss 00, 01, 02 … 58, 59
s 0, 1, 2 … 58, 59
Fractional Second S 0 1 … 8 9
SS 00, 01, 02 … 98, 99
SSS 000 001 … 998 999
SSSS… 000[0..] 001[0..] … 998[0..] 999[0..]
SSSSSS 000000 000001 … 999998 999999
Timezone Z -07:00, -06:00 … +06:00, +07:00
ZZ -0700, -0600 … +0600, +0700
Seconds timestamp X 1381685817, 1234567890.123
Microseconds timestamp x 1234567890123

The file sinks

If the sink is a str or a pathlib.Path, the corresponding file will be opened for writing logs. The path can also contains a special "{time}" field that will be formatted with the current date at file creation.

The rotation check is made before logging each messages. If there is already an existing file with the same name that the file to be created, then the existing file is renamed by appending the date to its basename to prevent file overwritting. This parameter accepts:

  • an int which corresponds to the maximum file size in bytes before that the current logged file is closed and a new one started over.
  • a datetime.timedelta which indicates the frequency of each new rotation.
  • a datetime.time which specifies the hour when the daily rotation should occur.
  • a str for human-friendly parametrization of one of the previously enumerated types. Examples: "100 MB", "0.5 GB", "1 month 2 weeks", "4 days", "10h", "monthly", "18:00", "sunday", "w0", "monday at 12:00", …
  • a function which will be called before logging. It should accept two arguments: the logged message and the file object, and it should return True if the rotation should happen now, False otherwise.

The retention occurs at rotation or at sink stop if rotation is None. Files are selected according to their basename, if it is the same that the sink file, with possible time field being replaced with .*. This parameter accepts:

  • an int which indicates the number of log files to keep, while older files are removed.
  • a datetime.timedelta which specifies the maximum age of files to keep.
  • a str for human-friendly parametrization of the maximum age of files to keep. Examples: "1 week, 3 days", "2 months", …
  • a function which will be called before the retention process. It should accept the list of log files as argument and process to whatever it wants (moving files, removing them, etc.).

The compression happens at rotation or at sink stop if rotation is None. This parameter acccepts:

  • a str which corresponds to the compressed or archived file extension. This can be one of: "gz", "bz2", "xz", "lzma", "tar", "tar.gz", "tar.bz2", "tar.xz", "zip".
  • a function which will be called before file termination. It should accept the path of the log file as argument and process to whatever it wants (custom compression, network sending, removing it, etc.).

The color markups

To add colors to your logs, you just have to enclose your format string with the appropriate tags. This is based on the great ansimarkup library from @gvalkov. Those tags are removed if the sink don’t support ansi codes.

The special tag <level> (abbreviated with <lvl>) is transformed according to the configured color of the logged message level.

Here are the available tags (note that compatibility may vary depending on terminal):

Color (abbr) Styles (abbr)
Black (k) Bold (b)
Blue (e) Dim (d)
Cyan (c) Normal (n)
Green (g) Italic (i)
Magenta (m) Underline (u)
Red (r) Strike (s)
White (w) Reverse (r)
Yellow (y) Blink (l)
  Hide (h)


Description Examples
Foreground Background
Basic colors <red>, <r> <GREEN>, <G>
Light colors <light-blue>, <le> <LIGHT-CYAN>, <LC>
Xterm colors <fg 86>, <fg 255> <bg 42>, <bg 9>
Hex colors <fg #00005f>, <fg #EE1> <bg #AF5FD7>, <bg #fff>
RGB colors <fg 0,95,0> <bg 72,119,65>
Stylizing <bold>, <b> , <underline>, <u>
Shorthand (FG, BG) <red, yellow>, <r, y>
Shorthand (Style, FG, BG) <bold, cyan, white>, <b,,w>, <b,c,>

The environment variables

The default values of sink parameters can be entirely customized. This is particularly useful if you don’t like the log format of the pre-configured sink.

Each of the start() default parameter can be modified by setting the LOGURU_[PARAM] environment variable. For example on Linux: export LOGURU_FORMAT="{time} - {message}" or export LOGURU_ENHANCE=NO.

The default levels attributes can also be modified by setting the LOGURU_[LEVEL]_[ATTR] environment variable. For example, on Windows: setx LOGURU_DEBUG_COLOR="<blue>" or setx LOGURU_TRACE_ICON="🚀".

If you want to disable the pre-configured sink, you can set the LOGURU_AUTOINIT variable to False.


>>> logger.start(sys.stdout, format="{time} - {level} - {message}", filter="sub.module")
>>> logger.start("file_{time}.log", level="TRACE", rotation="100 MB")
>>> def my_sink(message):
...     record = message.record
...     update_db(message, time=record.time, level=record.level)
>>> logger.start(my_sink)
>>> from logging import StreamHandler
>>> logger.start(StreamHandler(sys.stderr), format="{message}")
>>> class RandomStream:
...     def __init__(self, seed, threshold):
...         self.threshold = threshold
...         random.seed(seed)
...     def write(self, message):
...         if random.random() > self.threshold:
...             print(message)
>>> stream_object = RandomStream(seed=12345, threhold=0.25)
>>> logger.start(stream_object, level="INFO")
>>> logger.start(RandomStream, level="DEBUG", seed=34567, threshold=0.5)

Stop logging to a previously started sink.

Parameters:handler_id (int or None) – The id of the sink to stop, as it was returned by the start() method. If None, all sinks are stopped. The pre-configured sink is guaranteed to have the index 0.


>>> i = logger.start(sys.stderr, format="{message}")
>>> logger.info("Logging")
>>> logger.stop(i)
>>> logger.info("No longer logging")
catch(exception=<class 'Exception'>, *, level='ERROR', reraise=False, message="An error has been caught in function '{record[function]}', process '{record[process].name}' ({record[process].id}), thread '{record[thread].name}' ({record[thread].id}):")[source]

Return a decorator to automatically log possibly caught error in wrapped function.

This is useful to ensure unexpected exceptions are logged, the entire program can be wrapped by this method. This is also very useful to decorate threading.Thread.run() methods while using threads to propagate errors to the main logger thread.

Note that the visibility of variables values (which uses the cool better_exceptions library from @Qix-) depends on the backtrace option of each configured sinks.

The returned object can also be used as a context manager.

  • exception (Exception, optional) – The type of exception to intercept. If several types should be caught, a tuple of exceptions can be used too.
  • level (str or int, optional) – The level name or severity with which the message should be logged.
  • reraise (bool, optional) – Whether or not the exception should be raised again and hence propagated to the caller.
  • message (str, optional) – The message that will be automatically logged if an exception occurs. Note that it will be formatted with the record attribute.

decorator / context manager – An object that can be used to decorate a function or as a context manager to log exceptions possibly caught.


>>> @logger.catch
... def f(x):
...     100 / x
>>> def g():
...     f(10)
...     f(0)
>>> g()
ERROR - An error has been caught in function 'g', process 'Main' (367), thread 'ch1' (1398):
Traceback (most recent call last, catch point marked):
  File "program.py", line 12, in <module>
    └ <function g at 0x7f225fe2bc80>
> File "program.py", line 10, in g
    └ <function f at 0x7f225fe2b9d8>
  File "program.py", line 6, in f
    100 / x
          └ 0
ZeroDivisionError: division by zero
>>> with logger.catch(message="Because we never know..."):
...    main()  # No exception, no logs
opt(*, exception=None, record=False, lazy=False, ansi=False, raw=False, depth=0)[source]

Parametrize a logging call to slightly change generated log message.

  • exception (bool, tuple or Exception, optional) – It if does not evaluate as False, the passed exception is formatted and added to the log message. It could be an Exception object or a (type, value, traceback) tuple, otherwise the exception information is retrieved from sys.exc_info().
  • record (bool, optional) – If True, the record dict contextualizing the logging call can be used to format the message by using {record[key]} in the log message.
  • lazy (bool, optional) – If True, the logging call attribute to format the message should be functions which will be called only if the level is high enough. This can be used to avoid expensive functions if not necessary.
  • ansi (bool, optional) – If True, logged message will be colorized according to the markups it possibly contains.
  • raw (bool, optional) – If True, the formatting of each sink will be bypassed and the message will be send as is.
  • depth (int, optional) – Specify which stacktrace should be used to contextualize the logged message. This is useful while using the logger from inside a wrapped function to retrieve worthwhile information.

Logger – A logger wrapping the core logger, but transforming logged message adequately before sending.


>>> try:
...     1 / 0
... except ZeroDivisionError:
...    logger.opt(exception=True).debug("Exception logged with debug level:")
[18:10:02] DEBUG in '<module>' - Exception logged with debug level:
Traceback (most recent call last, catch point marked):
> File "<stdin>", line 2, in <module>
ZeroDivisionError: division by zero
>>> logger.opt(record=True).info("Current line is: {record[line]}")
[18:10:33] INFO in '<module>' - Current line is: 1
>>> logger.opt(lazy=True).debug("If sink <= DEBUG: {x}", x=lambda: math.factorial(2**5))
[18:11:19] DEBUG in '<module>' - If sink <= DEBUG: 263130836933693530167218012160000000
>>> logger.opt(ansi=True).warning("We got a <red>BIG</red> problem")
[18:11:30] WARNING in '<module>' - We got a BIG problem
>>> logger.opt(raw=True).debug("No formatting\n")
No formatting
>>> def wrapped():
...     logger.opt(depth=1).info("Get parent context")
>>> def func():
...     wrapped()
>>> func()
[18:11:54] DEBUG in 'func' - Get parent context

Bind attributes to the extra dict of each logged message record.

This is used to add custom context to each logging call.

Parameters:**kwargs – Mapping between keys and values that will be added to the extra dict.
Returns:Logger – A logger wrapping the core logger, but which sends record with the customized extra dict.


>>> logger.start(sys.stderr, format="{extra[ip]} - {message}")
>>> class Server:
...     def __init__(self, ip):
...         self.ip = ip
...         self.logger = logger.bind(ip=ip)
...     def call(self, message):
...         self.logger.info(message)
>>> instance_1 = Server("")
>>> instance_2 = Server("")
>>> instance_1.call("First instance") - First instance
>>> instance_2.call("Second instance") - Second instance
level(name, no=None, color=None, icon=None)[source]

Add, update or retrieve a logging level.

Logging levels are defined by their name to which a severity no, an ansi color and an icon are associated and possibly modified at run-time. To log() to a custom level, you should necessarily use its name, the severity number is not linked back to levels name (this implies that several levels can share the same severity).

To add a new level, all parameters should be passed so it can be properly configured.

To update an existing level, pass its name with the parameters to be changed.

To retrieve level information, the name solely suffices.

  • name (str) – The name of the logging level.
  • no (int) – The severity of the level to be added or updated.
  • color (str) – The color markup of the level to be added or updated.
  • icon (str) – The icon of the level to be added or updated.

Level – A namedtuple containing information about the level.


>>> level = logger.level("ERROR")
Level(no=40, color='<red><bold>', icon='❌')
>>> logger.start(sys.stderr, format="{level.no} {icon} {message}")
>>> logger.level("CUSTOM", no=15, color="<blue>", icon="@")
>>> logger.log("CUSTOM", "Logging...")
15 @ Logging...
>>> logger.level("WARNING", icon=r"/!\")
>>> logger.warning("Updated!")
30 /!\ Updated!

Disable logging of messages comming from name module and its children.

Developers of library using Loguru should absolutely disable it to avoid disrupting users with unrelated logs messages.

Parameters:name (str) – The name of the parent module to disable.


>>> logger.info("Allowed message by default")
[22:21:55] Allowed message by default
>>> logger.disable("my_library")
>>> logger.info("While publishing a library, don't forget to disable logging")

Enable logging of messages comming from name module and its children.

Logging is generally disabled by imported library using Loguru, hence this function allows users to receive these messages anyway.

Parameters:name (str) – The name of the parent module to re-allow.


>>> logger.disable("__main__")
>>> logger.info("Disabled, so nothing is logged.")
>>> logger.enable("__main__")
>>> logger.info("Re-enabled, messages are logged.")
[22:46:12] Re-enabled, messages are logged.
configure(*, handlers=None, levels=None, extra=None, activation=None)[source]

Configure the core logger.

  • handlers (list of dict, optional) – A list of each handler to be started. The list should contains dicts of params passed to the start() function as keyword arguments. If not None, all previously started handlers are first stopped.
  • levels (list of dict, optional) – A list of each level to be added or updated. The list should contains dicts of params passed to the level() function as keyword arguments. This will never remove previously created levels.
  • extra (dict, optional) – A dict containing additional parameters bound to the core logger, useful to share common properties if you call bind() in several of your files modules. If not None, this will remove previously configured extra dict.
  • activation (list of tuple, optional) – A list of (name, state) tuples which denotes which loggers should be enabled (if state is True) or disabled (if state is False). The calls to enable() and disable() are made accordingly to the list order. This will not modify previously activated loggers, so if you need a fresh start preprend your list with ("", False) or ("", True).

list of int – A list containing the identifiers of possibly started sinks.


>>> logger.configure(
...     handlers=[dict(sink=sys.stderr, format="[{time}] {message}"),
...            dict(sink="file.log", enqueue=True, serialize=True)],
...     levels=[dict(name="NEW", no=13, icon="¤", color="")],
...     extra={"common_to_all": "default"},
...     activation=[("my_module.secret": False, "another_library.module": True)]
... )
[1, 2]
static parse(file, pattern, *, cast={}, chunk=65536)[source]

Parse raw logs and extract each entry as a dict.

The logging format has to be specified as the regex pattern, it will then be used to parse the file and retrieve each entries based on the named groups present in the regex.

  • file (str, pathlib.Path or file-like object) – The path of the log file to be parsed, or alternatively an already opened file object.
  • pattern (str or re.Pattern) – The regex to use for logs parsing, it should contain named groups which will be included in the returned dict.
  • cast (function or dict, optional) – A function that should convert in-place the regex groups parsed (a dict of string values) to more appropiate types. If a dict is passed, its should be a mapping between keys of parsed log dict and the function that should be used to convert the associated value.
  • chunk (int, optional) – The number of bytes read while iterating through the logs, this avoid having to load the whole file in memory.

dict – The dict mapping regex named groups to matched values, as returned by re.Match.groupdict() and optionally converted according to cast argument.


>>> reg = r"(?P<lvl>[0-9]+): (?P<msg>.*)"    # If log format is "{level.no} - {message}"
>>> for e in logger.parse("file.log", reg):  # A file line could be "10 - A debug message"
...     print(e)                             # => {'lvl': '10', 'msg': 'A debug message'}
>>> caster = dict(lvl=int)                   # Parse 'lvl' key as an integer
>>> for e in logger.parse("file.log", reg, cast=caster):
...     print(e)                             # => {'lvl': 10, 'msg': 'A debug message'}
>>> def cast(groups):
...     if "date" in groups:
...         groups["date"] = datetime.strptime(groups["date"], "%Y-%m-%d %H:%M:%S")
>>> with open("file.log") as file:
...     for log in logger.parse(file, reg, cast=cast):
...         print(log["date"], log["something_else"])
trace(_message, *args, **kwargs)

Log _message.format(*args, **kwargs) with severity 'TRACE'.

debug(_message, *args, **kwargs)

Log _message.format(*args, **kwargs) with severity 'DEBUG'.

info(_message, *args, **kwargs)

Log _message.format(*args, **kwargs) with severity 'INFO'.

success(_message, *args, **kwargs)

Log _message.format(*args, **kwargs) with severity 'SUCCESS'.

warning(_message, *args, **kwargs)

Log _message.format(*args, **kwargs) with severity 'WARNING'.

error(_message, *args, **kwargs)

Log _message.format(*args, **kwargs) with severity 'ERROR'.

critical(_message, *args, **kwargs)

Log _message.format(*args, **kwargs) with severity 'CRITICAL'.

log(_level, _message, *args, **kwargs)[source]

Log _message.format(*args, **kwargs) with severity _level.

exception(_message, *args, **kwargs)[source]

Convenience method for logging an 'ERROR' with exception information.