fix(core): Use maxValueLength in extra error data integration

[AbhiPrasad]

fixes #12168

We should truncate extra error data keys using maxValueLength to make sure event payload size stays reasonably small.

[AbhiPrasad]

Actually now that I'm looking at this again, const value = error[key] means that value could be anything. What is the cleanest way here to make sure we truncate all the string? Should I adjust the normalize function?

[mydea]

I think we can just do something like this:

Suggested change

	extraErrorInfo[key] = truncate(isError(value) ? value.toString() : value, maxValueLength);
	extraErrorInfo[key] = isError(value) || typeof value === 'string' ? truncate(`${value}`, maxValueLength) : value;

?

[AbhiPrasad]

The issue here is that it'll only truncate string values at the top-most level. String values that are nested within an object will not get truncated.

[mydea]

Hmm I see. Yeah then if we want to fix this I guess we have to update normalize to also handle max value length!

[AbhiPrasad]

Given the normalize update will mean refactoring how base client and event processing works, I'm electing for the quick fix and just using your change on the extraErrorInfo[key] setter. This also unblocks the user.

I will create a GH issue and write down what we need to do to make a more robust fix.

[WesCossick]

Any update on this? We're starting to retroactively discover more and more lost errors that our engineering team needs to be aware of due to this bug.

[AbhiPrasad]

Hi @WesCossick sorry for the delay - there were some other items that took my attention away from this recently. I will try to get to this tomorrow and release by next week.

The source of the problems are the extraErrorDataIntegration integration. For now I recommend removing this integration to make sure you get no errors dropped.

As another workaround, you can make sure that the fields set by extraErrorDataIntegration on the sentry event are trimmed accordingly via the beforeSend option in Sentry.init.

[lforst]

Am I dumb or could we use our normalizeToSize() function to truncate the extra data to a sane length?