Simple solution that works
The logging functions will convert the '%s' to string representation (and if the object happens to be a container, then repr() will be used for the contained objects)
import logging
logging.basicConfig(level=logging.DEBUG, filename="sample.log")
logging.debug('Sample dict log: %s', {'name' : "John", 'age': 10})
How it shows in the log file:
DEBUG:root:Sample dict log: {'age': 10, 'name': 'John'}
If you have custom objects (e.g. instances of your own classes), you should implement a sensible __str__ and/or __repr__ for it, and the above will work without any hacks.
More on this here What is the difference between __str__ and __repr__?
A Note on performance
Notice that
logging.debug('%s', some_dict) is not same as
logging.debug('{0}'.format(some_dict))
In the first one, the function is passed 2 arguments '%s' and the original some_dict.
In the second one, the function is passed one argument which is the already-converted some_dict to a string.
Because a function needs the arguments evaluated, the second example will always do the conversion, even if logging configs have turned off the logging of debug messages.
That’s an unnecessary performance penalty.
In the the first example, the logging.debug function can decide to do the conversion, or not.
JSON is not very good for this
For objects that aren’t JSON-like (object instances, datetime, etc), json.dumps() would require a custom serializer e.g. with datetime objects you get:
TypeError: datetime.datetime(...) is not JSON serializable
This goes for any type that isn’t supported by json.dumps()
Whereas the logging module would figure out a good representation for the object