I'm seeing “WARNING: ~async_file_dispatcher_base() detects stuck async_io_op in total of X ops” during process close. What does this mean?

I'm seeing “WARNING: `~async_file_dispatcher_base()` detects stuck `async_io_op` in total of X ops” during process close. What does this mean?

This means that you scheduled ops with a dispatcher which did not complete in a timely fashion before you tried to destroy that dispatcher. This generally indicates a bug in your code, which can include:

An op is still running in some thread pool and you didn't use when_all_p() or get() to wait for it to complete before trying to destroy the dispatcher.
An op's precondition never completed and therefore the op was never started.

Tracking down the cause of the latter in particular is very similar to tracking down race conditions i.e. it's hard, and we as the authors of Boost.AFIO know just how hard! Try recompiling AFIO with the macro BOOST_AFIO_OP_STACKBACKTRACEDEPTH set to a reasonable depth like 8 will have AFIO take a stack backtrace for every op allocated which it then will print during the stuck ops warnings. This can be helpful to identify which ops exactly are stuck, and then you can figure out which preconditions of theirs are the cause.

Note that BOOST_AFIO_OP_STACKBACKTRACEDEPTH has two implementations on POSIX, one based on glibc backtrace() and the other based on libunwind. The latter is more portable, but requires explicit linking with libunwind, so we have defaulted to the former.