trace high-radius tag usage

[WeihanLi]

We should avoid too many radius for metrics, likes we should not include customerId in the metrics, should we also avoid it in trace tags also?
Should we exclude url.path tag when the url path may contain customerId?

[martincostello]

For metrics, high-cardinality data (like an arbitrary ID) is not recommended to be collected as attributes on a metric count towards its identity. For example foo.bar with user=1 is considered distinct from foo.bar with user=2, so if user is an arbitrary 32-bit integer it can cause a cardinality explosion and could lead to a large number of active series in a metrics back-end. This in turn could have a significant impact on your telemetry costs. A finite set (e.g. HTTP methods, GET/POST etc.) wouldn't typically be of concern.

The same consideration with respect to identity doesn't apply to traces as the identity for a span is its unique ID, so the number of attributes that a trace has doesn't affect its identity. Traces are often billed based on raw data storage costs, so what is logged isn't usually a factor - just how much.

Separate to costs, whether any particular data is safe to be stored in your telemetry back-end is up to your individual requirements and policies with consideration for things such as PII/PCI/HPAA etc.