CODEWITHSUNDEEP
clinuxmultithreadingpthreadsposix
shodanex
shodanex

Reputation: 15446

master / worker threads and signal handling

I am writing a program, with a master thread and some worker threads, and I would like to get the signal handling right. My problem is the following :

Master thread starts and do all allocation

Master thread sets a SIGINT signal handler

Master threads start worker threads. worker threads don't need special cleanup, however they can be sleeping on system call or semaphore.

When SIGINT is received, my understanding is that only one thread receives it. So if thread are sleeping on system call or semaphore, they wont be waked up, and I won't be able to pthread_join my worker threads and do all the necessary cleanup in my master thread.

Can the following signal handler resolve my problem ?

void term(int sig)
{
    g_do_cleanup = 1;
    pthread_kill(worker_1_id, some_other_signal);
    ...
    pthread_kill(worker_2_id, some_other_signal);
}

What I expect is that upon receiving SIGINT, all thread will be signaled with another signal, get out of their blocking call, see the g_do_cleanup flag and gracefully exit.

Any comment or link on how to do this properly is welcome.

Edit : I am NOT looking for a way to wakeup multiple thread waiting on a particular condition, so I don't think the pthread_cond_signal approach is what I am looking for. What I want is :

Upvotes: 6

Views: 7540

Answers (3)

Nikolai Fetissov
Nikolai Fetissov

Reputation: 84199

This is what I usually do in the main app thread:


/* before starting other threads */
sigset_t sigs;
sigemptyset( &sigs );
sigaddset( &sigs, SIGTERM );
sigaddset( &sigs, SIGINT );
/* add other signals to handle */
if ( pthread_sigmask( SIG_BLOCK, &sigs, 0 )) { /* handle error */ }
/* can start threads now */
...
/* in the main event loop */
siginfo_t info;
if ( sigwaitinfo( &sigs, &info ) == -1 ) { /* handle error */ }
switch( info.si_signo )
{
  case SIGTERM:
  case SIGINT: /* raise shutdown event */ break;
  default: /* other actions - log rotation, etc. */
}
...

This way signals become regular application events - no special signal context restrictions, etc. Waiting could also be done with a timeout via sigtimedwait, though BSD derivatives don't support it.

Upvotes: 4

Duck
Duck

Reputation: 27562

Your plan seems fine. You are forcing one thread to handle system signals. You'll need to mask the signals except some-other-signal in your worker threads with pthread-sigmask().

I think I would go with a separate thread to handle the process signal rather than master. Have it wait forever on a semaphore or sigwait() or whatever while waiting to run the signal handler. Move the pthread-kill code out of the signal handler. Just set the switch and have the signal-waiting thread send the other_signal to the worker threads and exit itself.

Upvotes: 1

T.E.D.
T.E.D.

Reputation: 44814

It sounds to me like what you want is a pthread condition variable, so that you can wake any number of multiple threads by singalling one "event" from your master thread.

See the man pages for PTHREAD_COND_DESTROY and PTHREAD_COND_BROADCAST for more information.

Upvotes: 2

Related Questions