Should span Status description be of low cardinality?

[jcarres-mdsol]

Currently the spec about span status mentions that the Description SHOULD be documented and predictable. and Description that provides a descriptive message of the Status.

We are using a library which adds to the description something like "failed to connect to queue 'uuid'" where 'uuid' is the whole uuid. This creates many unique descriptions which would prevent us from using description as a metric label, something that is on the other hand desirable in http spans.

Should the spec add a note description SHOULD be of low cardinality or description cardinality should follow the guidance for span names ?

[Oberon00]

You mention the span name, but you mean the description inside the span status, right?

[jcarres-mdsol]

Yes sorry, I've fixed the first link, it is all about the description of the span status

[trask]

there's been a bit of prior discussion on this, see #3496

I have the feeling that we may need to relax the restriction on span status description based on feedback from @Oberon00 and @jsuereth

prevent us from using description as a metric label

check out the brand new error.type attribute which is approximately the metric equivalent of span status description

[jmacd]

@trask I agree we should relax the restriction on span status -- and that some other well-specified attribute value should be used when metrics are requested.

[svrnm]

is this still being discussed?

[trask]

I think a PR would be useful to clarify that span status description does not need to be low cardinality, since this has been unclear previously.