fix(logs)!: move JSON parsing after user-defined processors

[swiatekm]

It's easier for users to write processors if they can assume the log body is a string, as opposed to having to work with arbitrary structured data.

Checklist

• Changelog updated or skip changelog label added
• Documentation updated
• Template tests added for new features
• Integration tests added or modified for major features

[sumo-drosiek]

It's easier for users to write processors if they can assume the log body is a string

Hmm, it is problematic if they would like to operate on the log structure 🤔

[swiatekm]

It's easier for users to write processors if they can assume the log body is a string

Hmm, it is problematic if they would like to operate on the log structure 🤔

They can explicitly parse it from json themselves in that case. And going the other way is apparently impossible in OTTL, or at least I haven't found a way. All the body manipulation functions assume the body is a string and do nothing otherwise. If you want to do things like masking PII, working with the serialized JSON is much easier.

[andrzej-stencel]

Isn't this a breaking change?

[swiatekm]

Isn't this a breaking change?

It's a fix, as the current behaviour is not intended, and the new behaviour will be better for the majority of users. However, some users may depend on the current behaviour. I think the change is beneficial enough in the aggregate that it should go into v3. I've added an extended note to the changelog and marked the change as breaking to make things clearer.