This is in part related to some of the leaks:
https://svn.boost.org/trac/boost/ticket/12220
We are still compiling with BOOST_THREAD_VERSION=4, which might prevent
the boost fix from being effective. (Need to check both xenial and
zesty.)
** Bug watch added: Boost Trac #12220
Before classifying this as a definite leak, it'd be good to work out
some way to return control to the event loop for some period after this
clean-up.
Cancelling a job is potentially an asynchronous operation, so the future
returned by cancel() needs to be kept alive until the future is ready.
We
