CloudWatchLogFormatter mutates logged dicts

[medley56]

Summary

When using the CloudWatchLogFormatter with its add_log_record_attrs functionality, it mutates dictionary objects that are passed to logging calls.

Reproducible Example

import logging
import watchtower

logger = logging.getLogger()  # root logger
logger.setLevel(logging.DEBUG)
added_attrs = ('lineno', 'funcName', 'levelname', 'created', 'module')
formatter = watchtower.CloudWatchLogFormatter(add_log_record_attrs=added_attrs)

stream = logging.StreamHandler()
stream.setFormatter(formatter)
logger.addHandler(stream)

logged_value = {"key": "value"}
print(logged_value)  # unchanged
logger.info(logged_value)
print(logged_value)  # mutated

Suggested Mitigation

It would be a performance hit but we could deepcopy the dict before adding all the attributes. This is what I'm doing on my custom implementation of a JsonFormatter.