Add attribute noting details of scrubbed values

[alexmojaki]

For #248

The idea is to use this to generate code for the user to whitelist values from scrubbing. Here's some example code which I imagine being ported to the frontend:

attributes = {
    'logfire.scrubbed': [
        {'path': ['attributes', 'usage'], 'matched_substring': 'token'},
        {'path': ['attributes', 'prompt'], 'matched_substring': 'secret'},
    ],
}

print(f"""
import logfire

def scrubbing_callback(match: logfire.ScrubMatch):
{"\n".join(f"""\
    if match.path == {tuple(match['path'])} and match.pattern_match.group(0) == {repr(match['matched_substring'])}:
        return match.value
"""
           for match in attributes['logfire.scrubbed'])}

logfire.configure(scrubbing=logfire.ScrubbingOptions(callback=scrubbing_callback))
""")

This generates:

import logfire

def scrubbing_callback(match: logfire.ScrubMatch):
    if match.path == ('attributes', 'usage') and match.pattern_match.group(0) == 'token':
        return match.value

    if match.path == ('attributes', 'prompt') and match.pattern_match.group(0) == 'secret':
        return match.value


logfire.configure(scrubbing=logfire.ScrubbingOptions(callback=scrubbing_callback))

[cloudflare-workers-and-pages]

Deploying logfire-docs with    Cloudflare Pages

Latest commit:	e545df4
Status:	✅  Deploy successful!
Preview URL:	https://372a62bb.logfire-docs.pages.dev
Branch Preview URL:	https://alex-scrubbed-note.logfire-docs.pages.dev

View logs

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

📢 Thoughts on this report? Let us know!

[samuelcolvin]

The code here looks good.

What do we need to change in the backend?

The PR body seems kind of confusing, and/or related to something slightly different. Am I being dumb or should or be updated.

[alexmojaki]

The PR body seems kind of confusing, and/or related to something slightly different. Am I being dumb or should or be updated.

It's related. The first bit of code is a sample of attributes that this PR adds to a span:

attributes = {
    'logfire.scrubbed': [
        {'path': ['attributes', 'usage'], 'matched_substring': 'token'},
        {'path': ['attributes', 'prompt'], 'matched_substring': 'secret'},
    ],
}

The rest of that code block is an example of how the attributes can be used to generate sample Python code which is displayed to the user for them to copy and paste to whitelist values from being scrubbed. That's the ultimate goal here.

That code generation logic could also be baked into the SDK, but I think it makes more sense to do that part in the frontend.

The last code block is the output of the first, i.e. the sample code which would be generated and shown to the user.

The backend doesn't really need anything. But since the new attribute is stored as a string and it's not mentioned in the JSON schema, I think the backend should explicitly check for this attribute and parse the JSON just in case someone wants to query it or something.