I'm using libuv and when calling uv_queue_work(..., work_cb, after_work_cb), for some reason, libuv takes ~50ms to call after_work_cb(...) after the end of work_cb(..), causing a significant latency overhead (testing on linux).
Is there a known workaround to substantially reduce this delay? Otherwise, I'll have to refactor the code to be synchronous, which would harm scalability, since work_cb(...) blocks for a couple of milliseconds, which it spends mostly on waiting for file system operations.