SyncHistogram::refresh() freezes

hdrhistogram = { version = "7.5.2", default-features = false, features = ["sync"] }

I'm getting recorders within a tokio::spawn closure:

time_per_one_recorder
    .get_or(|| {
        let time_per_one = time_per_one.lock().expect("time_per_one lock on updater");
        std::cell::RefCell::from(time_per_one.recorder())
    })
    .borrow_mut()
    .record(duration_ms)
    .expect("record to histogram");

When all tasks (1 million of them, i.e. that many record() calls) were processed and there's nothing else to record, I'm doing a refresh:

let mut time_per_one = time_per_one.lock().unwrap();
println!("=> Merging histograms...");
time_per_one.refresh();
println!("=> Merging done");

My program hangs forever on the refresh(), because the merging message is the last thing I'm seeing on stdout. According to top, my process doesn't do anything. I never tried debugging (as in gdb) for Rust programs so can't tell more at this point, will try it later.

By the way, everything used to be working fine whenever I called refresh() roughly once a second during the task processing. The freeze started happening when I decided one big merge when everything's done is better because it stops blocking recorders amid task processing.

Is this a crate bug, or maybe I'm using the crate wrong?

Thanks.

The report centers on SyncHistogram::refresh() and recorder() calls inside a tokio::spawn closure; start by reproducing the one-million-record workload and tracing the refresh path. Determine whether the freeze comes from crate synchronization or the calling pattern, then confirm a fix or documented usage correction with a test or reproducible result.