Esempi in C++ (Cpp) per jobacct_gather_g_suspend_poll

Esempio n. 1

File: req.c Progetto: VURM/slurm

static int
_handle_suspend(int fd, slurmd_job_t *job, uid_t uid)
{
	int rc = SLURM_SUCCESS;
	int errnum = 0;

	debug("_handle_suspend for job %u.%u",
	      job->jobid, job->stepid);

	debug3("  uid = %d", uid);
	if (!_slurm_authorized_user(uid)) {
		debug("job step suspend request from uid %ld for job %u.%u ",
		      (long)uid, job->jobid, job->stepid);
		rc = -1;
		errnum = EPERM;
		goto done;
	}

	if (job->cont_id == 0) {
		debug ("step %u.%u invalid container [cont_id:%"PRIu64"]",
			job->jobid, job->stepid, job->cont_id);
		rc = -1;
		errnum = ESLURMD_JOB_NOTRUNNING;
		goto done;
	}

	jobacct_gather_g_suspend_poll();

	/*
	 * Signal the container
	 */
	pthread_mutex_lock(&suspend_mutex);
	if (suspended) {
		rc = -1;
		errnum = ESLURMD_STEP_SUSPENDED;
		pthread_mutex_unlock(&suspend_mutex);
		goto done;
	} else {
		/* SIGTSTP is sent first to let MPI daemons stop their
		 * tasks, then we send SIGSTOP to stop everything else */
		if (slurm_container_signal(job->cont_id, SIGTSTP) < 0) {
			verbose("Error suspending %u.%u (SIGTSTP): %m",
				job->jobid, job->stepid);
		} else
			sleep(1);

		if (slurm_container_signal(job->cont_id, SIGSTOP) < 0) {
			verbose("Error suspending %u.%u (SIGSTOP): %m",
				job->jobid, job->stepid);
		} else {
			verbose("Suspended %u.%u", job->jobid, job->stepid);
		}
		suspended = true;
	}
	pthread_mutex_unlock(&suspend_mutex);

done:
	/* Send the return code and errno */
	safe_write(fd, &rc, sizeof(int));
	safe_write(fd, &errnum, sizeof(int));
	return SLURM_SUCCESS;
rwfail:
	return SLURM_FAILURE;
}

Esempio n. 2

File: req.c Progetto: lipari/slurm

static int
_handle_suspend(int fd, slurmd_job_t *job, uid_t uid)
{
	int rc = SLURM_SUCCESS;
	int errnum = 0;

	debug("_handle_suspend for job %u.%u",
	      job->jobid, job->stepid);

	debug3("  uid = %d", uid);
	if (!_slurm_authorized_user(uid)) {
		debug("job step suspend request from uid %ld for job %u.%u ",
		      (long)uid, job->jobid, job->stepid);
		rc = -1;
		errnum = EPERM;
		goto done;
	}

	if (job->cont_id == 0) {
		debug ("step %u.%u invalid container [cont_id:%"PRIu64"]",
			job->jobid, job->stepid, job->cont_id);
		rc = -1;
		errnum = ESLURMD_JOB_NOTRUNNING;
		goto done;
	}

	jobacct_gather_g_suspend_poll();

	/*
	 * Signal the container
	 */
	pthread_mutex_lock(&suspend_mutex);
	if (suspended) {
		rc = -1;
		errnum = ESLURMD_STEP_SUSPENDED;
		pthread_mutex_unlock(&suspend_mutex);
		goto done;
	} else {
		/* SIGTSTP is sent first to let MPI daemons stop their tasks,
		 * then wait 2 seconds, then send SIGSTOP to the spawned
		 * process's container to stop everything else.
		 *
		 * In some cases, 1 second has proven insufficient. Longer
		 * delays may help insure that all MPI tasks have been stopped
		 * (that depends upon the MPI implementaiton used), but will
		 * also permit longer time periods when more than one job can
		 * be running on each resource (not good). */
		if (slurm_container_signal(job->cont_id, SIGTSTP) < 0) {
			verbose("Error suspending %u.%u (SIGTSTP): %m",
				job->jobid, job->stepid);
		} else
			sleep(2);

		if (slurm_container_signal(job->cont_id, SIGSTOP) < 0) {
			verbose("Error suspending %u.%u (SIGSTOP): %m",
				job->jobid, job->stepid);
		} else {
			verbose("Suspended %u.%u", job->jobid, job->stepid);
		}
		suspended = true;
	}
	pthread_mutex_unlock(&suspend_mutex);

done:
	/* Send the return code and errno */
	safe_write(fd, &rc, sizeof(int));
	safe_write(fd, &errnum, sizeof(int));
	return SLURM_SUCCESS;
rwfail:
	return SLURM_FAILURE;
}