what's your go-to strategy for distributed tracing instrumentation?

hey folks, ive been wrestling with how to best instrument our microservices for distributed tracing. we've tried a few different libraries, but getting consistent context propagation and meaningful span data across services feels like a constant battle. are you using auto-instrumentation, manual instrumentation, or a hybrid approach? what libraries or frameworks have you found most reliable, and what are some common pitfalls to avoid when setting this up at scale?