Fillmore
The Python sentry-sdk has a before_send hook that lets you scrub Sentry events before they're sent. Fillmore makes it easier to set up a before_send scrubber and test it.
Code: | https://github.com/willkg/fillmore |
---|---|
Issues: | https://github.com/willkg/fillmore/issues |
License: | MPL v2 |
Documentation: | https://fillmore.readthedocs.io/ |
Goals
Goals of Fillmore:
- make it easier to configure Sentry event scrubbing in a way that you can reason about
- make it easier to test your scrubbing code so you know it's working over time
- scrub in a resilient manner and default to emitting some signal when it kicks up errors so you know when your error handling code is kicking up errors
From that, Fillmore has the following features:
- lets you specify keys to scrub in a Sentry event
- resilient to errors--if it fails, it will emit a signal that you can see and alert on
- links to relevant Sentry documentation, projects, and other things
- testing infrastructure to use in your integration tests
Install
Run:
$ pip install fillmore
Quickstart
Example:
# Create a scrubber from fillmore.scrubber import Scrubber, ScrubRule, build_scrub_query_string scrubber = Scrubber( scrub_rules=[ ScrubRule( key_path="request.headers", keys=["Auth-Token", "Cookie"], scrub_function="scrub", ), ScrubRule( key_path="request", keys=["query_string"], scrub_function=build_scrub_query_string(params=["code", "state"]), ), ScrubRule( key_path="exception.values.[].stacktrace.frames.[].vars", keys=["username", "password"], scrub_function="scrub", ), ] ) # Set up Sentry with the scrubber sentry_sdk.init( sentry_dsn="somedsn", ..., before_send=scrubber, )
Now you've got a scrubber. However, how do you know it's scrubbing the right stuff? How will you know if something changes and it's no longer scrubbing the right stuff?
Fillmore comes with a test harness you can use. For example, say you have a app called "myapp":
# myapp/libsentry.py import sentry_sdk # Create a scrubber from fillmore.scrubber import Scrubber, ScrubRule, build_scrub_query_string scrubber = Scrubber( scrub_rules=[ ScrubRule( key_path="request.headers", keys=["Auth-Token", "Cookie"], scrub_function="scrub", ), ScrubRule( key_path="request", keys=["query_string"], scrub_function=build_scrub_query_string(params=["code", "state"]), ), ScrubRule( key_path="exception.values.[].stacktrace.frames.[].vars", keys=["username", "password"], scrub_function="scrub", ), ] )
Then you can test it like this:
# myapp/tests/test_libsentry.py from myapp.libsentry import scrubber from fillmore.test import SentryTestHelper def test_scrubber(): helper = SentryTestHelper() with helper.session_context() as helper_with_context: helper_with_context.init(scrubber=scrubber) try: username = "foo" raise Exception("intentional") except Exception as exc: sentry_sdk.capture_exception(exc) (event,) = helper_with_context.events error = event["exception"]["values"][0] assert error["type"] == "Exception" assert error["value"] == "intentional" assert error["stacktrace"]["frames"][0]["vars"]["username"] == "[Scrubbed]"
This kicks up an exception in this context which sentry captures. If you need to test scrubbing for other contexts, you'll need to set that up differently. See Fillmore documentation for details and recipes.
Why this? Why not other libraries?
Other libraries:
Have an awkward API that is hard to reason about.
I'm not scrubbing Sentry events for fun. I need to be able to write scrubbing configuration that is exceptionally clear about what it is and isn't doing.
Don't covers large portions of the Sentry event structure.
I need scrubbers that cover the entire event structure as well as some of the curious cases like the fact that cookie information shows up twice and can be encoded as a string.
Aren't resilient.
The scrubber is running in the context of Sentry reporting an error. If it also errors out, then you can end up in situations where you never see errors and have no signal that something is horribly wrong. We need scrubbing code to be extremely resilient and default to emitting a signal that it's broken.
Don't include testing infrastructure.
I'm not scrubbing Sentry events for fun. I need to know that the scrubbing code is working correctly and that it continues to work as we upgrade Python, sentry_sdk, and other things.
Having testing infrastructure for making this easier is really important.