int
main (int argc, char *argv[])
{
int error_code = 0, i;
if (argc < 2) {
printf ("Usage: %s job_id\n", argv[0]);
exit (1);
}
for (i=1; i<argc; i++) {
error_code = slurm_kill_job ((uint32_t) atoi(argv[i]),
SIGKILL, 0);
if (error_code) {
char msg[64];
sprintf(msg, "slurm_kill_job(%.12s)",argv[i]);
slurm_perror (msg);
}
}
return (error_code);
}
int main (int argc, char *argv[])
{
int i, min_nodes = 1, max_nodes = 1, nodes, tasks = 0, rc = 0;
job_desc_msg_t job_req;
resource_allocation_response_msg_t *job_resp;
slurm_step_ctx_params_t step_params[1];
slurm_step_ctx_t *ctx = NULL;
slurm_step_launch_params_t launch[1];
char *task_argv[3];
int *fd_array = NULL;
int num_fd;
if (argc > 1) {
i = atoi(argv[1]);
if (i > 0)
min_nodes = i;
}
if (argc > 2) {
i = atoi(argv[2]);
if (i > 0)
max_nodes = i;
}
if (max_nodes < min_nodes)
max_nodes = min_nodes;
/* Create a job allocation */
slurm_init_job_desc_msg( &job_req );
job_req.min_nodes = min_nodes;
job_req.max_nodes = max_nodes;
job_req.user_id = getuid();
job_req.group_id = getgid();
job_req.time_limit = 1;
if (slurm_allocate_resources(&job_req, &job_resp)) {
slurm_perror ("slurm_allocate_resources");
printf("INFO: min_nodes=%u max_nodes=%u user_id=%u group_id=%u",
job_req.min_nodes, job_req.max_nodes,
job_req.user_id, job_req.group_id);
exit(0);
}
printf("job_id %u\n", job_resp->job_id);
fflush(stdout);
/* Wait for allocation request to be satisfied */
if ((job_resp->node_list == NULL) ||
(strlen(job_resp->node_list) == 0)) {
printf("Waiting for resource allocation\n");
fflush(stdout);
while ((job_resp->node_list == NULL) ||
(strlen(job_resp->node_list) == 0)) {
sleep(5);
if (slurm_allocation_lookup_lite(job_resp->job_id,
&job_resp) &&
(slurm_get_errno() != ESLURM_JOB_PENDING)) {
slurm_perror("slurm_confirm_allocation");
exit(0);
}
}
}
nodes = job_resp->node_cnt;
if (argc > 3)
tasks = atoi(argv[3]);
if (tasks < 1)
tasks = nodes * TASKS_PER_NODE;
if (tasks < nodes) {
fprintf(stderr, "Invalid task count argument\n");
exit(1);
}
printf("Starting %d tasks on %d nodes\n", tasks, nodes);
fflush(stdout);
/*
* Create a job step context.
*/
slurm_step_ctx_params_t_init(step_params);
step_params->job_id = job_resp->job_id;
step_params->min_nodes = nodes;
step_params->task_count = tasks;
ctx = slurm_step_ctx_create(step_params);
if ((ctx == NULL) &&
(slurm_get_errno() == ESLURM_PROLOG_RUNNING)) {
printf("SlurmctldProlog is still running, "
"sleep and try again\n");
sleep(10);
ctx = slurm_step_ctx_create(step_params);
}
if (ctx == NULL) {
slurm_perror("slurm_step_ctx_create");
rc = 1;
goto done;
}
/*
* Hack to run one task per node, regardless of what we set up
* when we created the job step context.
*/
if (slurm_step_ctx_daemon_per_node_hack(ctx) != SLURM_SUCCESS) {
slurm_perror("slurm_step_ctx_daemon_per_node_hack");
rc = 1;
goto done;
}
/*
* Launch the tasks using "user managed" IO.
* "user managed" IO means a TCP stream for each task, directly
* connected to the stdin, stdout, and stderr the task.
*/
slurm_step_launch_params_t_init(launch);
task_argv[0] = "./test7.3.io";
launch->argv = task_argv;
launch->argc = 1;
launch->user_managed_io = true; /* This is the key to using
"user managed" IO */
if (slurm_step_launch(ctx, launch, NULL) != SLURM_SUCCESS) {
slurm_perror("slurm_step_launch");
rc = 1;
goto done;
}
if (slurm_step_launch_wait_start(ctx) != SLURM_SUCCESS) {
slurm_perror("slurm_step_launch_wait_start");
rc =1;
goto done;
}
slurm_step_ctx_get(ctx, SLURM_STEP_CTX_USER_MANAGED_SOCKETS,
&num_fd, &fd_array);
/* Interact with launched tasks as desired */
_do_task_work(fd_array, tasks);
for (i = 0; i < tasks; i++) {
close(fd_array[i]);
}
slurm_step_launch_wait_finish(ctx);
/* Terminate the job killing all tasks */
done: slurm_kill_job(job_resp->job_id, SIGKILL, 0);
/* clean up storage */
slurm_free_resource_allocation_response_msg(job_resp);
if (ctx)
slurm_step_ctx_destroy(ctx);
exit(0);
}
static void
slurmdrmaa_job_control( fsd_job_t *self, int action )
{
slurmdrmaa_job_t *slurm_self = (slurmdrmaa_job_t*)self;
job_desc_msg_t job_desc;
fsd_log_enter(( "({job_id=%s}, action=%d)", self->job_id, action ));
fsd_mutex_lock( &self->session->drm_connection_mutex );
TRY
{
switch( action )
{
case DRMAA_CONTROL_SUSPEND:
if(slurm_suspend(fsd_atoi(self->job_id)) == -1) {
fsd_exc_raise_fmt( FSD_ERRNO_INTERNAL_ERROR,"slurm_suspend error: %s,job_id: %s",slurm_strerror(slurm_get_errno()),self->job_id);
}
slurm_self->user_suspended = true;
break;
case DRMAA_CONTROL_HOLD:
/* change priority to 0*/
slurm_init_job_desc_msg(&job_desc);
slurm_self->old_priority = job_desc.priority;
job_desc.job_id = atoi(self->job_id);
job_desc.priority = 0;
job_desc.alloc_sid = 0;
if(slurm_update_job(&job_desc) == -1) {
fsd_exc_raise_fmt( FSD_ERRNO_INTERNAL_ERROR,"slurm_update_job error: %s,job_id: %s",slurm_strerror(slurm_get_errno()),self->job_id);
}
break;
case DRMAA_CONTROL_RESUME:
if(slurm_resume(fsd_atoi(self->job_id)) == -1) {
fsd_exc_raise_fmt( FSD_ERRNO_INTERNAL_ERROR,"slurm_resume error: %s,job_id: %s",slurm_strerror(slurm_get_errno()),self->job_id);
}
slurm_self->user_suspended = false;
break;
case DRMAA_CONTROL_RELEASE:
/* change priority back*/
slurm_init_job_desc_msg(&job_desc);
job_desc.priority = INFINITE;
job_desc.job_id = atoi(self->job_id);
if(slurm_update_job(&job_desc) == -1) {
fsd_exc_raise_fmt( FSD_ERRNO_INTERNAL_ERROR,"slurm_update_job error: %s,job_id: %s",slurm_strerror(slurm_get_errno()),self->job_id);
}
break;
case DRMAA_CONTROL_TERMINATE:
if(slurm_kill_job(fsd_atoi(self->job_id),SIGKILL,0) == -1) {
fsd_exc_raise_fmt( FSD_ERRNO_INTERNAL_ERROR,"slurm_terminate_job error: %s,job_id: %s",slurm_strerror(slurm_get_errno()),self->job_id);
}
break;
default:
fsd_exc_raise_fmt(
FSD_ERRNO_INVALID_ARGUMENT,
"job::control: unknown action %d", action );
}
fsd_log_debug(("job::control: successful"));
}
FINALLY
{
fsd_mutex_unlock( &self->session->drm_connection_mutex );
}
END_TRY
fsd_log_return(( "" ));
}
