Sentry processors to scrub sensitive data

[weisk]

as stated in the sentry docs, http://sentry.readthedocs.org/en/latest/developer/client/index.html#scrubbing-data

It would be really helpful to be able to add such processors to the raven-js client. For example ignoring password fields by default.

[dcramer]

@weisk we dont capture any user data in raven.js afaik?

[weisk]

Well, we just log all failed ajax requests, sending the whole request and response to Sentry. Yes we can work around it by manually rejecting fields in the request body, but it would be much more convenient to have a feature as the sentry 'processors'.

The python client, for example, has the feature:
http://raven.readthedocs.org/en/latest/config/index.html?#raven.processors.SanitizePasswordsProcessor

[mattrobenolt]

Yeah, I plan on just documenting a way to do this for those that need it.

I'd rather not include it in core since most people don't need it, unlike how we collect data on the server side.

You can hook in a processor through the dataCallback config parameter. It's a chance to mutate the final object before it's sent off.

[weisk]

(^ sorry, clicked by accident)

Ok thanks @mattrobenolt ! I really missed that dataCallback option, it will do the trick!

[mattrobenolt]

Ah, cool. I should definitely port the one over from python and document it for inclusion on your own, or distribute as a plugin or something.

[mk0x9]

@mattrobenolt what's the status of the issue?

[benvinegar]

@mk0x9 – note that since this ticket was created in 2014, the Sentry server itself does automatic scrubbing of personally-identifiable information (PII).

Alternatively, you can do it manually by using dataCallback to inspect (and remove) potentially sensitive outbound data.

[dmose]

One of the reasons client-side scrubbing is compelling is because there have been instances in the past of organizations (e.g. governments) trying to force software vendors to log data that they have on the server-side.

One problem with using dataCallback is that it's fragile in the face of evolving data formats. Which is to say that if new fields are added containing messages, or things like stack argument collection in JS gets better, then it's unclear how someone maintaining a scrubber would know to update their scrubber.

[dmose]

Presumably, the python code that wants to be ported to fix this is https://github.com/getsentry/raven-python/blob/c69abaa1506a426802e979c7f8c6a4ab0ef9c4d7/raven/processors.py