Mutex with tracing enabled can hold parent spans open?

[xuorig]

I've been investigating an issue with some traces not getting closed in another repository and tracked it down to a tokio Mutex (with tracing feature enabled).

If a Mutex is created within the context of a particular, "request" trace, and then stored in memory to be reused by other requests or traces, it seems like that first request's span will never be closed, since the Mutex and its inner resource_span holds a "reference" (incremented the ref_count of the trace) to the parent span. (When using a tracing-subscriber)

Wondering if this is a known gotcha, and whether we should be careful to hold a tokio Mutex across requests? If not, any other ideas besides using const_new or switching to a std::sync::Mutex?

[xuorig]

Closing this for now, since this seems like a more general problem than I thought apollographql/router#4074 (comment).