The RateLimitingFilter is a filter for the Python logging system that allows you to restrict the rate at which messages can pass through your logging handlers.
The filter can be useful if you're using a handler such as Python's logging.handlers.SMTPHandler to send error notification emails. Error notification emails provide a useful means of keeping an eye on the health of a running system, but these emails have the potential to overload a mailbox if they start arriving in quick succession due to some kind of critical failure.
The RateLimitingFilter can help prevent mailbox overload by throttling messages based on a configurable rate, whilst allowing for initial bursts of messages which can be a useful indicator that something somewhere has broken.
$ pip install ratelimitingfilter
or
$ git clone https://github.com/wkeeling/ratelimitingfilter.git
$ cd ratelimitingfilter
$ python setup.py install
A typical use case may be to throttle error notification emails sent by the logging.handlers.SMTPHandler. Here's an example of how you might set it up:
import logging.handlers
import time
from ratelimitingfilter import RateLimitingFilter
logger = logging.getLogger('throttled_smtp_example')
# Create an SMTPHandler
smtp = logging.handlers.SMTPHandler(mailhost='smtp.example.com',
fromaddr='from@example.com',
toaddrs='to@example.com',
subject='An error has occurred')
smtp.setLevel(logging.ERROR)
# Create a formatter and set it on the handler
formatter = logging.Formatter('%(asctime)s - %(name)s - %(levelname)s - %(message)s')
smtp.setFormatter(formatter)
# Create an instance of the RateLimitingFilter, and add it to the handler
throttle = RateLimitingFilter()
smtp.addFilter(throttle)
# Add the handler to the logger
logger.addHandler(smtp)
# Logged errors will now be restricted to 1 every 30 seconds
while True:
logger.error('An error message')
time.sleep(2)Creating an instance of the RateLimitingFilter without any arguments like in the example above will restrict the flow of messages to 1 every 30 seconds.
You can customize the flow rate by supplying your own values for the rate, per and burst attributes. For example, to allow a rate of 1 message every 2 minutes with a periodic burst of up to 5 messages:
throttle = RateLimitingFilter(rate=1, per=120, burst=5)
smtp.addFilter(throttle)It is possible to pass some additional configuration options to the RateLimitingFilter initializer for further control over message throttling.
Perhaps you want to selectively throttle particular error messages whilst allowing other messages to pass through freely. This might be the case if there is part of the application which you know can generate large volumes of errors, whilst the rest of the application is unlikely to.
One way to achieve this might be to use separate loggers, one configured with rate limiting, one without, for the different parts of the application. Alternatively, you can use a single logger and configure the RateLimitingFilter to match only those messages that you want to throttle.
Applying selective rate limiting allows for constant visbility of lower volume errors whilst keeping the higher volume errors in check.
The RateLimitingFilter supports two ways to selectively throttle messages:
You can pass a list of substrings to the RateLimitingFilter which it will use to match messages to apply to.
config = {'match': ['some error', 'a different error']}
throttle = RateLimitingFilter(rate=1, per=60, burst=1, **config)
smtp.addFilter(throttle)
# Can be rate limited
logger.error('some error occurred')
# Can be rate limited
logger.error('a different error occurred')
# Will not be rate limited
logger.error('something completely different happened')This is an experimental feature.
You can let the RateLimitingFilter automatically throttle messages by setting the match option to auto.
config = {'match': 'auto'}
throttle = RateLimitingFilter(rate=1, per=60, burst=1, **config)The filter will then attempt to identify messages based on their content in order to figure out whether to throttle them or not. It will tolerate slight differences in content when identifying messages. So for example, if error messages are being rapidly logged that are the same apart from a timestamp, or perhaps an incrementing id, then these messages will be treated as the same as far as rate limiting is concerned.
MIT
Feedback and improvements are more than welcome. Please submit a pull request!