If all worker threads are busy and waiting for an event, we must ensure that
a job delivering that event gets executed. This new method has this property
for CRITICAL jobs, using a worker if we have one, but executing the job directly
if not.
During daemon shutdown, users might call remove() after processor.set_threads(0)
has been called. This gets problematic, as a watch event might be unable
to signal completion when no threads are available anymore. Work around this
issue by cancelling waiters once processor.cancel() has been called.
If a lock is held when queue_job() is called and the same lock is
required during the destruction of a job, holding the internal lock
in the processor while calling destroy() could result in a deadlock.
This avoids race conditions between calls to cancel() and jobs that like
to be rescheduled. If jobs were able to reschedule themselves it would
theoretically be possible that two worker threads have the same job
assigned (the one currently executing the job and the one executing the
same but rescheduled job if it already is time to execute it), this means
that cancel() could be called twice for that job.
Creating a new job based on the current one and reschedule that is also
OK, but rescheduling itself is more efficient for jobs that need to be
executed often.
This ensures that no threads are active when plugins and the rest of the
daemon are unloaded.
callback_job_t was simplified a lot in the process as its main
functionality is now contained in processor_t. The parent-child
relationships were abandoned as these were only needed to simplify job
cancellation.
Jobs are now destroyed by the processor, but they are allowed to
reschedule themselves. That is, parts of the reschedule functionality
already provided by callback_job_t is moved to the processor. Not yet
fully supported is JOB_REQUEUE_DIRECT and canceling jobs.
Note: job_t.destroy() is now called not only for queued jobs but also
after execution or cancellation of jobs. job_t.status can be used to
decide what to do in said method.
Warnings like
comparison of unsigned expression < 0 is always false
are reported with -Wextra when enum types that are compiled to an
unsigned type (which is up to the compiler) are checked for negativity.
During destruction the main thread locks the mutex in processor_t and
waits on a condvar for threads to have terminated. Because the mutex
has also to be locked to decrement the thread count the condvar cannot
be signaled before doing that as otherwise the main thread might already
be waiting to join the threads while locking the mutex and thus causing
a deadlock.
Martin Willi