[pkg/ottl] Validate that all parts of a Key are used by a Context

[TylerHelmuth]

Component(s)

pkg/ottl

Is your feature request related to a problem? Please describe.

OTTL currently validates that a context "consumes" all segments in a path, but does not validate that all a path's keys are used. This means users can do something like name["foo"] in the span context and get back the name of the span, completely ignoring the foo key. In this scenario the context is allowing parsing to succeed when it shouldnt. To simplify this process for all the different contexts, we can build something into OTTL, similar to Path validation, but for Keys.

Things to consider:

• Should we validate just the first key? - [pkg/ottl] Don't check that all keys are used #30047 (comment)
• How will key validation interact with dynamic indexing? If we don't know the value of the key until hot path execution, how should we validate that all keys are being considered?
• Should we have a way for contexts to declare that they've reached a point where the map is arbitrarily deep? So a context exposing a path of underlying type map[string]string would not declare this since stringMap["key"]["nestedKey"] would be invalid. But a context would declare this for an attributes path since the type is map[string]any and attributes["key1"]["key2"][...] is valid. - Originally from [pkg/ottl] Don't check that all keys are used #30047 (comment)

[TylerHelmuth]

@evan-bradley I though about this some more and I think there is a clean solution if we go back to keys being represented by a slice instead of a linked list.

Unlike the path segments, the keys are used during hot-path execution which means the next concept doesn't fit them as well. If a context needs keys, it needs all of them so it can feed them into the Map/Slice functions (or something else).

I think we can solve this problem by having OTTL record whether the context asked for all the keys or not. If a context has a requirement to only allow a specific number of keys, such as ottlspan's trace_state, then the context can be responsible for counting the number of keys. If a context doesn't request the keys for a given path, we'll be able to fail. This means situations like replace_match(body["metadata"]["uid"], "*", "12345") will be allowed but name["invalid"] will fail.

[TylerHelmuth]

Which is also pretty annoying because ottl.Key interface ends up really similar to the previous struct from the grammar and now I have to change a bunch of tests again lol