Only thread blocked in io_service.run() is not processing event loop.

[Tanner Sansbury <twsansbury@…>]

There is a concurrency issue in task_io_service (Boost 1.48+), that can leave a thread blocked in io_service.run(), waiting for its wakeup_event to be signaled. If thread A invokes io_service::poll(), runs the reactor while thread B invokes io_service.run(), becoming an idle thread (op_queue_ is empty), then if no work is ready to run, thread A io_service::poll() returns 0 without waking up the idle thread B. This results in thread B remaining blocked on io_service::run(), but never processing the event loop.

A side-by-side visualization illustrating the problem:

          poll thread                  |          main thread
---------------------------------------+---------------------------------------
  lock()                               | 
  do_poll_one()                        |                          
  |-- pop task_operation_ from         |
  |   queue_op_                        |
  |-- unlock()                         | lock()
  |-- create task_cleanup              | do_run_one()
  |-- service reactor (non-block)      | `-- queue_op_ is empty
  |-- ~task_cleanup()                  |     |-- set thread as idle
  |   |-- lock()                       |     `-- unlock()
  |   `-- queue_op_.push(              |
  |       task_operation_)             |
  `-- task_operation_ is               | 
      queue_op_.front()                |
      `-- return 0                     | // still waiting on wakeup_event
  unlock()                             |

Attached files are:

• Patch file that modifies task_io_service.ipp to sleep immediately after do_poll_one runs the reactor. This is only used to increase the changes of reproducing the concurrency problem.
• Test application:
  • An asynchronous work loop via a timer that will print "." every 3 seconds.
  • Spawn off a single thread that will poll the io_service.
  • Delay to allow the new thread time to poll io_service, and have main call io_service::run()}} while the poll thread sleeps in {{{task_io_service::do_poll_one().
• Patch with suggested fix that invokes wake_one_thread_and_unlock() before do_poll_one returning 0 if no work is to be done after running the reactor.

[Tanner Sansbury <twsansbury@…>]

Patch task_io_service to increase changes of reproducing concurrency problem.

[Tanner Sansbury <twsansbury@…>]

Test application used with test.patch task_io_service.

[Tanner Sansbury <twsansbury@…>]

Proposed fix.

[chris_kohlhoff]

Fixed on trunk in [84348]. Merged to release in [84388]