Span attributes cardinality

[mladedav]

I know metrics and logs suffer when cardinality is high. But since it may still contain useful information, logs can have only some predefined tags used for indexing and the rest is put inside a structured log line.

Is the same issue considered for traces? I've seen some concern for cardinality of span name (e.g. semantic conventions for HTTP mention this explicitly), but not for other fields. For example there are predefined attributes for client IP, host identifier, route and and so on, which together will often uniquely identify a span.

Is there also an expectation that the backend would be configured so that searching would be optimized only for a subset of attributes?

Or is the idea of searching by anything but trace ID already wrong and the expected use is to look up traces only after I've obtained trace ID either from a log line or an exemplar?