This is a Python implementation of the "Circuit Breaker" Pattern (https://martinfowler.com/bliki/CircuitBreaker.html). Inspired by Michael T. Nygard's highly recommendable book Release It! (https://pragprog.com/titles/mnee2/release-it-second-edition/).
The project is available on PyPI. Simply run:
$ pip install circuitbreaker
This is the simplest example. Just decorate a function with the @circuit decorator:
from circuitbreaker import circuit
@circuit
def external_call():
...
Async functions are also supported:
@circuit
async def external_call():
...
This decorator sets up a circuit breaker with the default settings. The circuit breaker:
- monitors the function execution and counts failures
- resets the failure count after every successful execution (while it is closed)
- opens and prevents further executions after 5 subsequent failures
- switches to half-open and allows one test-execution after 30 seconds recovery timeout
- closes if the test-execution succeeded
- considers all raised exceptions (based on class
Exception) as an expected failure - is named "external_call" - the name of the function it decorates
A failure is a raised exception, which was not caught during the function call.
By default, the circuit breaker listens for all exceptions based on the class Exception.
That means, that all exceptions raised during the function call are considered as an
"expected failure" and will increase the failure count.
It is important, to be as specific as possible, when defining the expected exception.
The main purpose of a circuit breaker is to protect your distributed system from a cascading failure.
That means, you probably want to open the circuit breaker only, if the integration point on the other
end is unavailable. So e.g. if there is an ConnectionError or a request Timeout.
If you are e.g. using the requests library (https://docs.python-requests.org/) for making HTTP calls,
its RequestException class would be a great choice for the expected_exception parameter.
The logic for treating thrown exceptions as failures can also be customized by passing a callable. The callable will be passed the exception type and value, and should return True if the exception should be treated as a failure.
All recognized exceptions will be re-raised anyway, but the goal is, to let the circuit breaker only recognize those exceptions which are related to the communication to your integration point.
When it comes to monitoring (see Monitoring), it may lead to falsy conclusions, if a
circuit breaker opened, due to a local OSError or KeyError, etc.
The following configuration options can be adjusted via decorator parameters. For example:
from circuitbreaker import circuit
@circuit(failure_threshold=10, expected_exception=ConnectionError)
def external_call():
...
By default, the circuit breaker opens after 5 subsequent failures. You can adjust this value with the failure_threshold parameter.
By default, the circuit breaker stays open for 30 seconds to allow the integration point to recover.
You can adjust this value with the recovery_timeout parameter.
By default, the circuit breaker listens for all exceptions which are based on the Exception class.
You can adjust this with the expected_exception parameter. It can be either an exception class, an iterable of an exception classes,
or a callable.
Use a callable if the logic to flag exceptions as failures is more complex than a type check. For example:
# Assume we are using the requests library
def is_not_http_error(thrown_type, thrown_value):
return issubclass(thrown_type, RequestException) and not issubclass(thrown_type, HTTPError)
def is_rate_limited(thrown_type, thrown_value):
return issubclass(thrown_type, HTTPError) and thrown_value.status_code == 429
@circuit(expected_exception=is_not_http_error)
def call_flaky_api(...):
rsp = requests.get(...)
rsp.raise_for_status()
return rsp
@circuit(expected_exception=is_rate_limited)
def call_slow_server(...):
rsp = requests.get(...)
rsp.raise_for_status()
return rsp
```
By default, the circuit breaker name is the name of the function it decorates. You can adjust the name with parameter name.
By default, the circuit breaker will raise a CircuitBreaker exception when the circuit is opened.
You can instead specify a function to be called when the circuit is opened. This function can be specified with the
fallback_function parameter and will be called with the same parameters as the decorated function would be.
The fallback type of call must also match the decorated function. For instance, if the decorated function is an
async generator, the fallback_function must be an async generator as well.
If you apply circuit breakers to a couple of functions and you always set specific options other than the default values,
you can extend the CircuitBreaker class and create your own circuit breaker subclass instead:
from circuitbreaker import CircuitBreaker
class MyCircuitBreaker(CircuitBreaker):
FAILURE_THRESHOLD = 10
RECOVERY_TIMEOUT = 60
EXPECTED_EXCEPTION = RequestException
Now you have two options to apply your circuit breaker to a function. As an Object directly:
@MyCircuitBreaker()
def external_call():
...
Please note, that the circuit breaker class has to be initialized, you have to use a class instance as decorator (@MyCircuitBreaker()), not the class itself (@MyCircuitBreaker).
Or via the decorator proxy:
@circuit(cls=MyCircuitBreaker)
def external_call():
...
To keep track of the health of your application and the state of your circuit breakers, every circuit breaker registers itself at the CircuitBreakerMonitor. You can receive all registered circuit breakers via CircuitBreakerMonitor.get_circuits().
To get an aggregated health status, you can ask the Monitor via CircuitBreakerMonitor.all_closed(). Or you can retrieve the currently open circuits via CircuitBreakerMonitor.get_open() and the closed circuits via CircuitBreakerMonitor.get_closed().
