distributed tracing: how granular is granular enough

hey folks, ive been thinkin alot bout distributed tracing lately, esp in the context of microservices. weve got some basic tracing setup, but im wonderin how much detail is *too* much detail. what are ur teams doin? are u tracn every single rpc, or are u more selective? what factors influence ur decision on what to instrument and what to leave as is. lookin fwd to hearin ur experiences