Changeset 38 for branches/slurm-drmaa-1.2.0/slurm_drmaa

Timestamp:

07/22/14 11:17:42 (11 years ago)

Author:

pkopta

Message:

 

Location:

branches/slurm-drmaa-1.2.0/slurm_drmaa

Files:

• job.c (modified) (4 diffs)
• job.h (modified) (1 diff)
• session.c (modified) (3 diffs)
• util.c (modified) (1 diff)

branches/slurm-drmaa-1.2.0/slurm_drmaa/job.c

@@ -261 +261 @@
                 const fsd_template_t *jt,
                 fsd_environ_t **envp,
-                job_desc_msg_t * job_desc,
-                int n_job /* ~job_step */
-                )
+                job_desc_msg_t * job_desc)
 {
         fsd_expand_drmaa_ph_t *volatile expand = NULL;
@@ -269 +267 @@
         TRY
          {
-                expand = fsd_expand_drmaa_ph_new( NULL, NULL, fsd_asprintf("%d",n_job) );
-                slurmdrmaa_job_create( session, jt, envp, expand, job_desc, n_job);
+                expand = fsd_expand_drmaa_ph_new( NULL, NULL, fsd_strdup("%a") );
+                slurmdrmaa_job_create( session, jt, envp, expand, job_desc );
          }
         EXCEPT_DEFAULT
@@ -308 +306 @@
                 fsd_environ_t **envp,
                 fsd_expand_drmaa_ph_t *expand, 
-                job_desc_msg_t * job_desc,
-                int n_job
+                job_desc_msg_t * job_desc
                 )
 {
@@ -326 +323 @@
         const char *job_category = "default";
         
-        slurmdrmaa_init_job_desc( job_desc );
-
-        slurm_init_job_desc_msg( job_desc );
-        
         job_desc->user_id = getuid();
         job_desc->group_id = getgid();

branches/slurm-drmaa-1.2.0/slurm_drmaa/job.h

@@ -43 +43 @@
 };
 
-void slurmdrmaa_job_create_req(fsd_drmaa_session_t *session, const fsd_template_t *jt, fsd_environ_t **envp, job_desc_msg_t * job_desc, int n_job );
-void slurmdrmaa_job_create(fsd_drmaa_session_t *session, const fsd_template_t *jt, fsd_environ_t **envp, fsd_expand_drmaa_ph_t *expand, job_desc_msg_t * job_desc, int n_job);
+void slurmdrmaa_job_create_req(fsd_drmaa_session_t *session, const fsd_template_t *jt, fsd_environ_t **envp, job_desc_msg_t * job_desc );
+void slurmdrmaa_job_create(fsd_drmaa_session_t *session, const fsd_template_t *jt, fsd_environ_t **envp, fsd_expand_drmaa_ph_t *expand, job_desc_msg_t * job_desc );
 
 #endif /* __SLURM_DRMAA__JOB_H */

branches/slurm-drmaa-1.2.0/slurm_drmaa/session.c

@@ -98 +98 @@
         job_desc_msg_t job_desc;
         submit_response_msg_t *submit_response = NULL;
+        job_info_msg_t *job_info = NULL;
+
+    /* zero out the struct, and set default vaules */
+        slurm_init_job_desc_msg( &job_desc );
         
         TRY
          {
-
+        unsigned i;
                 if( start != end )
                  {
-                        unsigned idx, i;
-
                         n_jobs = (end - start) / incr + 1;
 
                         fsd_calloc( job_ids, n_jobs+1, char* );
+                        job_desc.array_inx = fsd_asprintf( "%d-%d:%d", start, end, incr );
+         }
 
-                        for( idx = start, i = 0;  idx <= (unsigned)end;  idx += incr, i++ )
-                         {
-                                connection_lock = fsd_mutex_lock( &self->drm_connection_mutex );
-                                slurmdrmaa_job_create_req( self, jt, (fsd_environ_t**)&env , &job_desc, idx);
-                                if(slurm_submit_batch_job(&job_desc,&submit_response)){
-                                        fsd_exc_raise_fmt(
-                                                FSD_ERRNO_INTERNAL_ERROR,"slurm_submit_batch_job: %s",slurm_strerror(slurm_get_errno()));
+                connection_lock = fsd_mutex_lock( &self->drm_connection_mutex );
+                slurmdrmaa_job_create_req( self, jt, (fsd_environ_t**)&env , &job_desc );
+                if(slurm_submit_batch_job(&job_desc,&submit_response)){
+                        fsd_exc_raise_fmt(
+                                FSD_ERRNO_INTERNAL_ERROR,"slurm_submit_batch_job: %s",slurm_strerror(slurm_get_errno()));
+                }
+
+                connection_lock = fsd_mutex_unlock( &self->drm_connection_mutex );
+
+                fsd_log_debug(("job %u submitted", submit_response->job_id));
+
+                if( start != end )
+                {
+                        if ( SLURM_SUCCESS == slurm_load_job( &job_info, submit_response->job_id, 0) )
+                        {
+                                fsd_assert(  job_info->record_count == n_jobs );
+                                for(i=0; i < job_info->record_count; i++)
+                                {
+                                        job_ids[i] = fsd_asprintf( "%d", job_info->job_array[i].job_id);
+
+                                        job = slurmdrmaa_job_new( fsd_strdup(job_ids[i]) );
+                                        job->session = self;
+                                        job->submit_time = time(NULL);
+                                        self->jobs->add( self->jobs, job );
+                                        job->release( job );
+                                        job = NULL;
                                 }
-
-                                fsd_log_debug(("job %u submitted", submit_response->job_id));         
-                                connection_lock = fsd_mutex_unlock( &self->drm_connection_mutex );
-
-                                job_ids[i] = fsd_asprintf("%d",submit_response->job_id); /*TODO  */
-
-                                job = slurmdrmaa_job_new( fsd_strdup(job_ids[i]) ); 
-                                job->session = self;
-                                job->submit_time = time(NULL);
-                                self->jobs->add( self->jobs, job );
-                                job->release( job );  
-                                job = NULL;
-                         }
-                        fsd_assert( i == n_jobs );
-                 }
-                else /* ! bulk */
-                 {
+                        } else  {
+                                fsd_exc_raise_fmt( FSD_ERRNO_INTERNAL_ERROR,"slurm_load_job: %s",slurm_strerror(slurm_get_errno()));
+                        }
+                } else {
                         fsd_calloc( job_ids, n_jobs+1, char* );
 
-                        connection_lock = fsd_mutex_lock( &self->drm_connection_mutex );
-                        slurmdrmaa_job_create_req( self, jt, (fsd_environ_t**)&env , &job_desc, 0);
-                        if(slurm_submit_batch_job(&job_desc,&submit_response)){
-                                fsd_exc_raise_fmt(
-                                        FSD_ERRNO_INTERNAL_ERROR,"slurm_submit_batch_job: %s",slurm_strerror(slurm_get_errno()));
-                        }
-
-                        connection_lock = fsd_mutex_unlock( &self->drm_connection_mutex );
-
-                        fsd_log_debug(("job %u submitted", submit_response->job_id));         
-                        
                         job_ids[0] = fsd_asprintf( "%d", submit_response->job_id); /* .0*/
 
@@ -156 +155 @@
                         job->release( job );
                         job = NULL;
-                 }
+                }
          }
          ELSE
@@ -194 +193 @@
         return job;
 }
-

branches/slurm-drmaa-1.2.0/slurm_drmaa/util.c

@@ -168 +168 @@
         fsd_free(job_desc->exc_nodes);
         fsd_free(job_desc->gres);
+        fsd_free(job_desc->array_inx);
         
         fsd_log_return(( "" ));