Curious, do people actually use high cardinality data to make decisions or it is more of the case to store all labels if needed in future
There's always going to be a large amount of data, especially observability data, that's never used to make significant decisions. However we want high cardinality data both to increase the chance that the information we're storing will be useful, and to make compression and other data management easier. For example if we're storing fully paramaterized URLs as metrics names we'll never be able to consolidate or average that data.
A brief guide with some examples since this seems to come up every time we discuss DIY metrics tools, structured logs, and parsing.