static int submit_by_project(const char *project)
{
   drmaa_job_template_t *jt = NULL; 
   char diagnosis[DRMAA_ERROR_STRING_BUFFER];
   char jobid[100];
   int drmaa_errno, i;

   for (i=1; i<=40; i++) {
      if (!(jt = create_job_template(job_path, project, i))) {
         fprintf(stderr, "create_sleeper_job_template() failed\n");
         return 1;
      }

      while ((drmaa_errno=drmaa_run_job(jobid, sizeof(jobid)-1, jt, diagnosis,
               sizeof(diagnosis)-1)) == DRMAA_ERRNO_DRM_COMMUNICATION_FAILURE) {
         fprintf(stderr, "drmaa_run_job() failed - retry: %s\n", diagnosis);
         sleep(1);
      }
      if (drmaa_errno != DRMAA_ERRNO_SUCCESS) {
         fprintf(stderr, "drmaa_run_job() failed: %s\n", diagnosis);
         return 1;
      }
      if (!quiet)
         printf("\t \"%s\"\n", jobid);
      drmaa_delete_job_template(jt, NULL, 0);
   }

   return 0;
}
Exemplo n.º 2
0
int main(int argc, char *argv[])
{
   char diagnosis[DRMAA_ERROR_STRING_BUFFER + 1];
   char jobid[DRMAA_JOBNAME_BUFFER + 1];
   int drmaa_errno = DRMAA_ERRNO_SUCCESS;
   const char *job_path = NULL;
   const char *job_cat = NULL;
   void *buffer = NULL;
   size_t buffer_size = 0;
   drmaa_job_template_t *jt = NULL;
   int exit_code = 0;
   
   job_path = argv[1];
   job_cat = argv[2];

   if (drmaa_init("", diagnosis, DRMAA_ERROR_STRING_BUFFER) != DRMAA_ERRNO_SUCCESS) {
      fprintf(stderr, "drmaa_init() failed: %s\n", diagnosis);
      return 1;
   }

   /* Submit a job to find out what the current job id is. */
   jt = create_job_template(job_path, job_cat);
   
   if (jt == NULL) {
      fprintf(stderr, "create_job_template() failed\n");
      exit_code = 1;
      goto error;
   }
   
   /* Make sure the next available block of memory contains something other than
    * NULL. */
   buffer_size = 256 * sizeof(void*);
   buffer = malloc(buffer_size);
   memset(buffer, 255, buffer_size);
   sge_free(&buffer);
   
   drmaa_errno = drmaa_run_job(jobid, DRMAA_JOBNAME_BUFFER, jt, diagnosis,
                               DRMAA_ERROR_STRING_BUFFER);
   
   if (drmaa_errno != DRMAA_ERRNO_SUCCESS) {
      fprintf(stderr, "drmaa_run_job() failed: %s %s\n", diagnosis,
              drmaa_strerror(drmaa_errno));
      exit_code = 1;
      goto error;
   }
   
   drmaa_delete_job_template(jt, diagnosis, DRMAA_ERROR_STRING_BUFFER);

error:
   if (drmaa_exit(diagnosis, DRMAA_ERROR_STRING_BUFFER) != DRMAA_ERRNO_SUCCESS) {
      fprintf(stderr, "drmaa_exit() failed: %s\n", diagnosis);
      exit_code = 1;
   }
   
   return exit_code;
}
Exemplo n.º 3
0
int main(int argc, char **argv)
{
   drmaa_job_template_t *jt_a = NULL;
   drmaa_job_template_t *jt_b = NULL;
   drmaa_job_ids_t *jobids_a = NULL;
   drmaa_job_ids_t *jobids_b = NULL;
   const char *base_name = "Job";
   char name_a[512], name_b[512];

   struct chunking_t {
      int a;
      int b;
   } chunking[] = {
      { 1, 1 },
      { 1, 3 },
      { 1, 4 },
      { 1, 12 },
      { 3, 1 },
      { 3, 3 },
      { 3, 4 },
      { 3, 12 },
      { 4, 1 },
      { 4, 3 },
      { 4, 4 },
      { 4, 12 },
      { 12, 1 },
      { 12, 3 },
      { 12, 4 },
      { 12, 12 } 
   };
   int samples, i, drmaa_errno = 0;

   while (argc > 2) {
      if (!strcmp(argv[1], "-ps")) {
         fprintf(stderr, "use drmaa_job_ps() for progress monitoring\n");
         do_ps = 1;
         argc--;
         argv++;
      }
      if (!strcmp(argv[1], "-1st")) {
         fprintf(stderr, "first sample only\n");
         do_1st = 1;
         argc--;
         argv++;
      }
      if (!strcmp(argv[1], "-exit_unknown")) {
         fprintf(stderr, "first sample only\n");
         do_exit = 0;
         argc--;
         argv++;
      }

      if (!strcmp(argv[1], "-nm")) {
         argc--;
         argv++;
         base_name = argv[1];
         argc--;
         argv++;
         fprintf(stderr, "use \"%s\" as base name\n", base_name);
      }
   }

   if (argc != 2) {
      printf("Usage: %s path_to_sleeper_script\n", argv[0]);
      exit(1);
   }

   if (strstr(argv[1], "sleeper.sh") == NULL) {
      printf("Usage: %s path_to_sleeper_script\n", argv[0]);
      exit(1);
   }

   if (do_1st)
      samples = 1;
   else
      samples = sizeof(chunking)/sizeof(void *);

   while ((drmaa_errno=drmaa_init(NULL, errorbuf, sizeof(errorbuf)-1) == DRMAA_ERRNO_DRM_COMMUNICATION_FAILURE))
      sleep(1);
   if (drmaa_errno != DRMAA_ERRNO_SUCCESS) {
      fprintf(stderr, "drmaa_init failed: %s\n", errorbuf);
      exit(EXIT_FAILURE);
   }

   strcpy(name_a, base_name);
   strcat(name_a, "A");
   strcpy(name_b, base_name);
   strcat(name_b, "B");

   for (i=0; i< samples; i++) {

      jt_a = create_job_template(argv[1], name_a, NULL, chunking[i].a, 1);
      jt_b = create_job_template(argv[1], name_b, name_a, chunking[i].b, 0);
      
      while ((drmaa_errno = drmaa_run_bulk_jobs(&jobids_a, jt_a, 1, BULK_SIZE, chunking[i].a, 
            errorbuf, sizeof(errorbuf)-1))==DRMAA_ERRNO_DRM_COMMUNICATION_FAILURE)
         sleep(1);
      if(drmaa_errno != DRMAA_ERRNO_SUCCESS) {
         fprintf(stderr, "drmaa_run_bulk_jobs(%d-%d:%d) failed: %s\n", 1, BULK_SIZE, chunking[i].a, errorbuf);
         return 1;
      }

      while ((drmaa_errno = drmaa_run_bulk_jobs(&jobids_b, jt_b, 1, BULK_SIZE, chunking[i].b, 
            errorbuf, sizeof(errorbuf)-1))==DRMAA_ERRNO_DRM_COMMUNICATION_FAILURE)
         sleep(1);
      if(drmaa_errno != DRMAA_ERRNO_SUCCESS) {
         fprintf(stderr, "drmaa_run_bulk_jobs(%d-%d:%d) failed: %s\n", 1, BULK_SIZE, chunking[i].b, errorbuf);
         return 1;
      }

      /* sleeper will sleep 60s by default, so this test should be okay */
      validate_jobs(jobids_a, chunking[i].a, jobids_b, chunking[i].b);

      drmaa_release_job_ids(jobids_a);
      drmaa_release_job_ids(jobids_b);
      
      drmaa_errno = drmaa_delete_job_template(jt_a, errorbuf, sizeof(errorbuf)-1);
      if(drmaa_errno != DRMAA_ERRNO_SUCCESS) {
         fprintf(stderr, "drmaa_delete_job_template failed: %s\n", errorbuf);
         return 1;
      }

      drmaa_errno = drmaa_delete_job_template(jt_b, errorbuf, sizeof(errorbuf)-1);
      if(drmaa_errno != DRMAA_ERRNO_SUCCESS) {
         fprintf(stderr, "drmaa_delete_job_template failed: %s\n", errorbuf);
         return 1;
      }
   }

   while ((drmaa_errno = drmaa_exit(errorbuf, sizeof(errorbuf)-1)) == DRMAA_ERRNO_DRM_COMMUNICATION_FAILURE)
      sleep(1);
      
   if(drmaa_errno != DRMAA_ERRNO_SUCCESS) {
      fprintf(stderr, "drmaa_exit failed: %s\n", errorbuf);
      return 1;
   }
    
   printf ("OK\n");

   return 0;
}
Exemplo n.º 4
0
int main(int argc, char *argv[])
{
   char diagnosis[DRMAA_ERROR_STRING_BUFFER];
   const char *all_jobids[NBULKS*JOB_CHUNK + JOB_CHUNK+1];
   char jobid[100];
   int drmaa_errno, i, pos = 0;
   const char *job_path;
   drmaa_job_template_t *jt;
  
   if (argc<2) {
      fprintf(stderr, "usage: example <path-to-job>\n");
      return 1;
   }
   job_path = argv[1];

   if (drmaa_init(NULL, diagnosis, sizeof(diagnosis)-1) != DRMAA_ERRNO_SUCCESS) {
      fprintf(stderr, "drmaa_init() failed: %s\n", diagnosis);
      return 1;
   }

   /*
    *   submit some bulk jobs
    */
   if (!(jt = create_job_template(job_path, 5, 1))) {
      fprintf(stderr, "create_job_template() failed\n");
      return 1;
   }
   for (i=0; i<NBULKS; i++) {
      drmaa_job_ids_t *jobids;
      int j;

      while ((drmaa_errno=drmaa_run_bulk_jobs(&jobids, jt, 1, JOB_CHUNK, 1, diagnosis,
                 sizeof(diagnosis)-1))==DRMAA_ERRNO_DRM_COMMUNICATION_FAILURE) {
         fprintf(stderr, "drmaa_run_bulk_jobs() failed - retry: %s %s\n", diagnosis, drmaa_strerror(drmaa_errno));
         sleep(1);
      } 
      if (drmaa_errno != DRMAA_ERRNO_SUCCESS) {
         fprintf(stderr, "drmaa_run_bulk_jobs() failed: %s %s\n", diagnosis, drmaa_strerror(drmaa_errno));
         return 1;
      }

      printf("submitted bulk job with jobids:\n");
      for (j=0; j<JOB_CHUNK; j++) {
         drmaa_get_next_job_id(jobids, jobid, sizeof(jobid)-1);
         all_jobids[pos++] = strdup(jobid);
         printf("\t \"%s\"\n", jobid);
      } 
      drmaa_release_job_ids(jobids);
   }
   drmaa_delete_job_template(jt, NULL, 0);

   /*
    *   submit some sequential jobs
    */
   if (!(jt = create_job_template(job_path, 5, 0))) {
      fprintf(stderr, "create_sleeper_job_template() failed\n");
      return 1;
   }
   for (i=0; i<JOB_CHUNK; i++) {
      while ((drmaa_errno=drmaa_run_job(jobid, sizeof(jobid)-1, jt, diagnosis,
               sizeof(diagnosis)-1)) == DRMAA_ERRNO_DRM_COMMUNICATION_FAILURE) {
         fprintf(stderr, "drmaa_run_job() failed - retry: %s\n", diagnosis);
         sleep(1);
      }
      if (drmaa_errno != DRMAA_ERRNO_SUCCESS) {
         fprintf(stderr, "drmaa_run_job() failed: %s\n", diagnosis);
         return 1;
      }
      printf("\t \"%s\"\n", jobid);
      all_jobids[pos++] = strdup(jobid);
   }

   /* set string array end mark */
   all_jobids[pos] = NULL;

   drmaa_delete_job_template(jt, NULL, 0);

   /*
    *   synchronize with all jobs
    */
   drmaa_errno = drmaa_synchronize(all_jobids, DRMAA_TIMEOUT_WAIT_FOREVER, 0, diagnosis, sizeof(diagnosis)-1);
   if (drmaa_errno != DRMAA_ERRNO_SUCCESS) {
      fprintf(stderr, "drmaa_synchronize(DRMAA_JOB_IDS_SESSION_ALL, dispose) failed: %s\n", diagnosis);
      return 1;
   }
   printf("synchronized with all jobs\n");

   /*
    *   wait all those jobs
    */
   for (pos=0; pos<NBULKS*JOB_CHUNK + JOB_CHUNK; pos++) {
      int stat;
      int aborted, exited, exit_status, signaled;

      drmaa_errno = drmaa_wait(all_jobids[pos], jobid, sizeof(jobid)-1, 
         &stat, DRMAA_TIMEOUT_WAIT_FOREVER, NULL, diagnosis, sizeof(diagnosis)-1);

      if (drmaa_errno != DRMAA_ERRNO_SUCCESS) {
         fprintf(stderr, "drmaa_wait(%s) failed: %s\n", all_jobids[pos], diagnosis);
         return 1;
      }

      /*
       * report how job finished 
       */
      drmaa_wifaborted(&aborted, stat, NULL, 0);
      if (aborted)
         printf("job \"%s\" never ran\n", all_jobids[pos]);
      else {
         drmaa_wifexited(&exited, stat, NULL, 0);
         if (exited) {
            drmaa_wexitstatus(&exit_status, stat, NULL, 0);
            printf("job \"%s\" finished regularly with exit status %d\n", 
                  all_jobids[pos], exit_status);
         } else {
            drmaa_wifsignaled(&signaled, stat, NULL, 0);
            if (signaled) {
               char termsig[DRMAA_SIGNAL_BUFFER+1];
               drmaa_wtermsig(termsig, DRMAA_SIGNAL_BUFFER, stat, NULL, 0);
               printf("job \"%s\" finished due to signal %s\n", 
                  all_jobids[pos], termsig);
            } else
               printf("job \"%s\" finished with unclear conditions\n", 
                  all_jobids[pos]);
         }
      }
   }

   if (drmaa_exit(diagnosis, sizeof(diagnosis)-1) != DRMAA_ERRNO_SUCCESS) {
      fprintf(stderr, "drmaa_exit() failed: %s\n", diagnosis);
      return 1;
   }
   
  return 0;
}
int main(int argc, char *argv[])
{
   char diagnosis[DRMAA_ERROR_STRING_BUFFER];
   char *s, jobid[100];
   int drmaa_errno, i;
   int ret = 0;
   struct timeval start_s, finish_s, wait_s;
  
   if (argc<2) {
      usage();
      return 1;
   }

   i = 1;
   do {
      if (!strcmp("-help", argv[i]) ||
          !strcmp("-h", argv[i])) {
         usage();
         return 0;

      } else if (!strcmp("-jobs", argv[i])) {
         i++; 
         if (argc < i+1) {
            usage();
            return 1;
         }
         njobs = atoi(argv[i]);
         i++; 

      } else if (!strcmp("-native", argv[i])) {
         i++; 
         if (argc < i+1) {
            usage();
            return 1;
         }
         native_spec = argv[i];
         i++; 

      } else if (!strcmp("-threads", argv[i])) {
         if (argc < i+1) {
            usage();
            return 1;
         }
         i++; 
         nthreads = atoi(argv[i]);
         i++; 

      } else if (!strcmp("-quiet", argv[i])) {
         i++; 
         if (argc < i+1) {
            usage();
            return 1;
         }
         if (!strcmp("yes", argv[i]) || !strcmp("y", argv[i])) 
            quiet = 1;
         else if (!strcmp("no", argv[i]) || !strcmp("n", argv[i])) 
            quiet = 0;
         else {
            usage();
            return 1;
         }
         i++; 

      } else if (!strcmp("-wait", argv[i])) {
         i++; 
         if (argc < i+1) {
            usage();
            return 1;
         }
         if (!strcmp("yes", argv[i]) || !strcmp("y", argv[i])) 
            dowait = 1;
         else if (!strcmp("no", argv[i]) || !strcmp("n", argv[i])) 
            dowait = 0;
         else {
            usage();
            return 1;
         }
         i++; 

      } else if (!strcmp("-scenario", argv[i])) {
         i++; 
         if (argc < i+1) {
            usage();
            return 1;
         }
         s = strchr(argv[i], '.');
         *s = '\0';
         if (strcmp("queue", argv[i]) && strcmp("type", argv[i]) && 
             strcmp("number", argv[i]) && strcmp("pe", argv[i])) {
            usage();
            return 1;
         }
         scenario = strdup(argv[i]);
        
         s++;
         if (strcmp("hostgroup", s) && strcmp("resource", s) && 
             strcmp("none", s) && strcmp("softresource", s) && 
             strcmp("softhostgroup", s)) {
            usage();
            return 1;
         }
         site_b = strdup(s);
         i++; 

      } else {
         job_path = argv[i];
         i++; 
         if (job_path[0]=='-') {
            usage();
            return 1;
         }

         if (argv[i]) {
            job_args = &argv[i];
         }
      }
   } while (i < argc && !job_path);

   if (!job_path) {
      usage();
      return 1;
   }

#if 0
   printf("job_path: \"%s\"\n", job_path);
   printf("njobs:    %d\n", njobs);
   printf("nthreads: %d\n", nthreads);
   printf("native:   %s\n", native_spec);
   printf("dowait:   %s\n", dowait?"yes":"no");
   printf("quiet:    %s\n", quiet?"yes":"no");
   printf("scenario: %s\n", scenario?scenario:"<no such>");
   printf("site_b:   %s\n", site_b?site_b:"<no such>");
   printf("1st arg:  %s\n", job_args?job_args[0]:"<noargs>");
#endif

   if (drmaa_init(NULL, diagnosis, sizeof(diagnosis)-1) != DRMAA_ERRNO_SUCCESS) {
      fprintf(stderr, "drmaa_init() failed: %s\n", diagnosis);
      return 1;
   }

   get_gmt(&start_s);

   if (!scenario) {
      if (!getuid()) {
         fprintf(stderr, "switching to ah114088:gridware\n");
         setegid(339);
         seteuid(115088);
      }

      if (!(jt = create_job_template(job_path, NULL, 0))) {
         fprintf(stderr, "create_sleeper_job_template() failed\n");
         return 1;
      }

      if (nthreads==1) {
         if (submit_jobs(&argv[i]))
             return 1;
      } else {
         pthread_t *ids = NULL;
         ids = (pthread_t *)malloc(sizeof (pthread_t) * nthreads);

         for (i = 0; i < nthreads; i++) {
            if (pthread_create(&ids[i], NULL, submit_jobs, NULL)) {
               fprintf(stderr, "pthread_create() failed: %s\n", strerror(errno));
	       free(ids);
               return 1;
            }
         }

         for (i = 0; i < nthreads; i++) {
            pthread_join(ids[i], NULL);
         }
      }
   
      drmaa_delete_job_template(jt, NULL, 0);

      if (!getuid()) {
         fprintf(stderr, "switching to root:root\n");
         seteuid(0);
         setegid(0);
      }

   } else {
      if (submit_by_project("project1") || submit_by_project("project2") ||
          submit_by_project("project3") || submit_by_project("project4"))
            return 1;
   }

   get_gmt(&finish_s);
   printf("submission took %8.3f seconds\n", DELTA_SECONDS(start_s, finish_s)); 

   if (dowait) {
      int success = 1;

      for (i=0; i<njobs * nthreads; i++) {
         int stat;
         int aborted, exited, exit_status, signaled;

         drmaa_errno = drmaa_wait(DRMAA_JOB_IDS_SESSION_ANY, jobid, sizeof(jobid)-1, 
            &stat, DRMAA_TIMEOUT_WAIT_FOREVER, NULL, diagnosis, sizeof(diagnosis)-1);

         if (drmaa_errno != DRMAA_ERRNO_SUCCESS) {
            fprintf(stderr, "drmaa_wait() failed: %s\n", diagnosis);
            return 1;
         }

         /*
          * report how job finished 
          */
         drmaa_wifaborted(&aborted, stat, NULL, 0);
         if (aborted) {
            printf("job \"%s\" never ran\n", jobid);
            success = 0;
         } else {
            drmaa_wifexited(&exited, stat, NULL, 0);
            if (exited) {
               drmaa_wexitstatus(&exit_status, stat, NULL, 0);
               if (exit_status != 0) {
                  success = 0;
                  printf("job \"%s\" with exit status %d\n", jobid, exit_status);
               } else {
                  if (!quiet)
                     printf("job \"%s\" finished regularly\n", jobid);
               }
            } else {
               success = 0;
               drmaa_wifsignaled(&signaled, stat, NULL, 0);
               if (signaled) {
                  char termsig[DRMAA_SIGNAL_BUFFER+1];
                  drmaa_wtermsig(termsig, DRMAA_SIGNAL_BUFFER, stat, NULL, 0);
                  printf("job \"%s\" finished due to signal %s\n", jobid, termsig);
               } else
                  printf("job \"%s\" finished with unclear conditions\n", jobid);
            }
         }
      }

      if (!success)
         ret = 1;

      get_gmt(&wait_s);
      printf("wait took %8.3f seconds\n", DELTA_SECONDS(finish_s, wait_s)); 
      printf("jobs took %8.3f seconds\n", DELTA_SECONDS(start_s, wait_s)); 
   }

   if (drmaa_exit(diagnosis, sizeof(diagnosis)-1) != DRMAA_ERRNO_SUCCESS) {
      fprintf(stderr, "drmaa_exit() failed: %s\n", diagnosis);
      return 1;
   }
   
  return ret;
}