Example #1
0
/*
 * slurm_sprint_job_info - output information about a specific Slurm
 *	job based upon message as loaded using slurm_load_jobs
 * IN job_ptr - an individual job information record pointer
 * IN one_liner - print as a single line if true
 * RET out - char * containing formatted output (must be freed after call)
 *           NULL is returned on failure.
 */
extern char *
slurm_sprint_job_info ( job_info_t * job_ptr, int one_liner )
{
	int i, j, k;
	char time_str[32], *group_name, *user_name;
	char *gres_last = "", tmp1[128], tmp2[128];
	char *tmp6_ptr;
	char tmp_line[1024 * 128];
	char tmp_path[MAXPATHLEN];
	char *ionodes = NULL;
	uint16_t exit_status = 0, term_sig = 0;
	job_resources_t *job_resrcs = job_ptr->job_resrcs;
	char *out = NULL;
	time_t run_time;
	uint32_t min_nodes, max_nodes = 0;
	char *nodelist = "NodeList";
	bitstr_t *cpu_bitmap;
	char *host;
	int sock_inx, sock_reps, last;
	int abs_node_inx, rel_node_inx;
	int64_t nice;
	int bit_inx, bit_reps;
	uint64_t *last_mem_alloc_ptr = NULL;
	uint64_t last_mem_alloc = NO_VAL64;
	char *last_hosts;
	hostlist_t hl, hl_last;
	char select_buf[122];
	uint32_t cluster_flags = slurmdb_setup_cluster_flags();
	uint32_t threads;
	char *line_end = (one_liner) ? " " : "\n   ";

	if (cluster_flags & CLUSTER_FLAG_BG) {
		nodelist = "MidplaneList";
		select_g_select_jobinfo_get(job_ptr->select_jobinfo,
					    SELECT_JOBDATA_IONODES,
					    &ionodes);
	}

	/****** Line 1 ******/
	xstrfmtcat(out, "JobId=%u ", job_ptr->job_id);

	if (job_ptr->array_job_id) {
		if (job_ptr->array_task_str) {
			xstrfmtcat(out, "ArrayJobId=%u ArrayTaskId=%s ",
				   job_ptr->array_job_id,
				   job_ptr->array_task_str);
		} else {
			xstrfmtcat(out, "ArrayJobId=%u ArrayTaskId=%u ",
				   job_ptr->array_job_id,
				   job_ptr->array_task_id);
		}
	}
	xstrfmtcat(out, "JobName=%s", job_ptr->name);
	xstrcat(out, line_end);

	/****** Line 2 ******/
	user_name = uid_to_string((uid_t) job_ptr->user_id);
	group_name = gid_to_string((gid_t) job_ptr->group_id);
	xstrfmtcat(out, "UserId=%s(%u) GroupId=%s(%u) MCS_label=%s",
		   user_name, job_ptr->user_id, group_name, job_ptr->group_id,
		   (job_ptr->mcs_label==NULL) ? "N/A" : job_ptr->mcs_label);
	xfree(user_name);
	xfree(group_name);
	xstrcat(out, line_end);

	/****** Line 3 ******/
	nice = ((int64_t)job_ptr->nice) - NICE_OFFSET;
	xstrfmtcat(out, "Priority=%u Nice=%"PRIi64" Account=%s QOS=%s",
		   job_ptr->priority, nice, job_ptr->account, job_ptr->qos);
	if (slurm_get_track_wckey())
		xstrfmtcat(out, " WCKey=%s", job_ptr->wckey);
	xstrcat(out, line_end);

	/****** Line 4 ******/
	xstrfmtcat(out, "JobState=%s ", job_state_string(job_ptr->job_state));

	if (job_ptr->state_desc) {
		/* Replace white space with underscore for easier parsing */
		for (j=0; job_ptr->state_desc[j]; j++) {
			if (isspace((int)job_ptr->state_desc[j]))
				job_ptr->state_desc[j] = '_';
		}
		xstrfmtcat(out, "Reason=%s ", job_ptr->state_desc);
	} else
		xstrfmtcat(out, "Reason=%s ", job_reason_string(job_ptr->state_reason));

	xstrfmtcat(out, "Dependency=%s", job_ptr->dependency);
	xstrcat(out, line_end);

	/****** Line 5 ******/
	xstrfmtcat(out, "Requeue=%u Restarts=%u BatchFlag=%u Reboot=%u ",
		 job_ptr->requeue, job_ptr->restart_cnt, job_ptr->batch_flag,
		 job_ptr->reboot);
	if (WIFSIGNALED(job_ptr->exit_code))
		term_sig = WTERMSIG(job_ptr->exit_code);
	exit_status = WEXITSTATUS(job_ptr->exit_code);
	xstrfmtcat(out, "ExitCode=%u:%u", exit_status, term_sig);
	xstrcat(out, line_end);

	/****** Line 5a (optional) ******/
	if (job_ptr->show_flags & SHOW_DETAIL) {
		if (WIFSIGNALED(job_ptr->derived_ec))
			term_sig = WTERMSIG(job_ptr->derived_ec);
		else
			term_sig = 0;
		exit_status = WEXITSTATUS(job_ptr->derived_ec);
		xstrfmtcat(out, "DerivedExitCode=%u:%u", exit_status, term_sig);
		xstrcat(out, line_end);
	}

	/****** Line 6 ******/
	if (IS_JOB_PENDING(job_ptr))
		run_time = 0;
	else if (IS_JOB_SUSPENDED(job_ptr))
		run_time = job_ptr->pre_sus_time;
	else {
		time_t end_time;
		if (IS_JOB_RUNNING(job_ptr) || (job_ptr->end_time == 0))
			end_time = time(NULL);
		else
			end_time = job_ptr->end_time;
		if (job_ptr->suspend_time) {
			run_time = (time_t)
				(difftime(end_time, job_ptr->suspend_time)
				 + job_ptr->pre_sus_time);
		} else
			run_time = (time_t)
				difftime(end_time, job_ptr->start_time);
	}
	secs2time_str(run_time, time_str, sizeof(time_str));
	xstrfmtcat(out, "RunTime=%s ", time_str);

	if (job_ptr->time_limit == NO_VAL)
		xstrcat(out, "TimeLimit=Partition_Limit ");
	else {
		mins2time_str(job_ptr->time_limit, time_str, sizeof(time_str));
		xstrfmtcat(out, "TimeLimit=%s ", time_str);
	}

	if (job_ptr->time_min == 0)
		xstrcat(out, "TimeMin=N/A");
	else {
		mins2time_str(job_ptr->time_min, time_str, sizeof(time_str));
		xstrfmtcat(out, "TimeMin=%s", time_str);
	}
	xstrcat(out, line_end);

	/****** Line 7 ******/
	slurm_make_time_str(&job_ptr->submit_time, time_str, sizeof(time_str));
	xstrfmtcat(out, "SubmitTime=%s ", time_str);

	slurm_make_time_str(&job_ptr->eligible_time, time_str, sizeof(time_str));
	xstrfmtcat(out, "EligibleTime=%s", time_str);

	xstrcat(out, line_end);

	/****** Line 8 (optional) ******/
	if (job_ptr->resize_time) {
		slurm_make_time_str(&job_ptr->resize_time, time_str, sizeof(time_str));
		xstrfmtcat(out, "ResizeTime=%s", time_str);
		xstrcat(out, line_end);
	}

	/****** Line 9 ******/
	slurm_make_time_str(&job_ptr->start_time, time_str, sizeof(time_str));
	xstrfmtcat(out, "StartTime=%s ", time_str);

	if ((job_ptr->time_limit == INFINITE) &&
	    (job_ptr->end_time > time(NULL)))
		xstrcat(out, "EndTime=Unknown ");
	else {
		slurm_make_time_str(&job_ptr->end_time, time_str, sizeof(time_str));
		xstrfmtcat(out, "EndTime=%s ", time_str);
	}

	if (job_ptr->deadline) {
		slurm_make_time_str(&job_ptr->deadline, time_str, sizeof(time_str));
		xstrfmtcat(out, "Deadline=%s", time_str);
	} else {
		xstrcat(out, "Deadline=N/A");
	}

	xstrcat(out, line_end);

	/****** Line 10 ******/
	if (job_ptr->preempt_time == 0)
		xstrcat(out, "PreemptTime=None ");
	else {
		slurm_make_time_str(&job_ptr->preempt_time, time_str, sizeof(time_str));
		xstrfmtcat(out, "PreemptTime=%s ", time_str);
	}

	if (job_ptr->suspend_time) {
		slurm_make_time_str(&job_ptr->suspend_time, time_str, sizeof(time_str));
		xstrfmtcat(out, "SuspendTime=%s ", time_str);
	} else
		xstrcat(out, "SuspendTime=None ");

	xstrfmtcat(out, "SecsPreSuspend=%ld", (long int)job_ptr->pre_sus_time);
	xstrcat(out, line_end);

	/****** Line 11 ******/
	xstrfmtcat(out, "Partition=%s AllocNode:Sid=%s:%u",
		   job_ptr->partition, job_ptr->alloc_node, job_ptr->alloc_sid);
	xstrcat(out, line_end);

	/****** Line 12 ******/
	xstrfmtcat(out, "Req%s=%s Exc%s=%s", nodelist, job_ptr->req_nodes,
		   nodelist, job_ptr->exc_nodes);
	xstrcat(out, line_end);

	/****** Line 13 ******/
	xstrfmtcat(out, "%s=%s", nodelist, job_ptr->nodes);
	if (job_ptr->nodes && ionodes) {
		xstrfmtcat(out, "[%s]", ionodes);
		xfree(ionodes);
	}
	if (job_ptr->sched_nodes)
		xstrfmtcat(out, " Sched%s=%s", nodelist, job_ptr->sched_nodes);

	xstrcat(out, line_end);

	/****** Line 14 (optional) ******/
	if (job_ptr->batch_host) {
		xstrfmtcat(out, "BatchHost=%s", job_ptr->batch_host);
		xstrcat(out, line_end);
	}

	/****** Line 14a (optional) ******/
	if (job_ptr->fed_siblings) {
		xstrfmtcat(out, "FedOrigin=%s FedSiblings=%s",
			   job_ptr->fed_origin_str, job_ptr->fed_siblings_str);
		xstrcat(out, line_end);
	}

	/****** Line 15 ******/
	if (cluster_flags & CLUSTER_FLAG_BG) {
		select_g_select_jobinfo_get(job_ptr->select_jobinfo,
					    SELECT_JOBDATA_NODE_CNT,
					    &min_nodes);
		if ((min_nodes == 0) || (min_nodes == NO_VAL)) {
			min_nodes = job_ptr->num_nodes;
			max_nodes = job_ptr->max_nodes;
		} else if (job_ptr->max_nodes)
			max_nodes = min_nodes;
	} else if (IS_JOB_PENDING(job_ptr)) {
		min_nodes = job_ptr->num_nodes;
		max_nodes = job_ptr->max_nodes;
		if (max_nodes && (max_nodes < min_nodes))
			min_nodes = max_nodes;
	} else {
		min_nodes = job_ptr->num_nodes;
		max_nodes = 0;
	}

	_sprint_range(tmp_line, sizeof(tmp_line), min_nodes, max_nodes);
	xstrfmtcat(out, "NumNodes=%s ", tmp_line);
	_sprint_range(tmp_line, sizeof(tmp_line), job_ptr->num_cpus, job_ptr->max_cpus);
	xstrfmtcat(out, "NumCPUs=%s ", tmp_line);

	xstrfmtcat(out, "NumTasks=%u ", job_ptr->num_tasks);
	xstrfmtcat(out, "CPUs/Task=%u ", job_ptr->cpus_per_task);

	if (job_ptr->boards_per_node == (uint16_t) NO_VAL)
		xstrcat(out, "ReqB:S:C:T=*:");
	else
		xstrfmtcat(out, "ReqB:S:C:T=%u:", job_ptr->boards_per_node);

	if (job_ptr->sockets_per_board == (uint16_t) NO_VAL)
		xstrcat(out, "*:");
	else
		xstrfmtcat(out, "%u:", job_ptr->sockets_per_board);

	if (job_ptr->cores_per_socket == (uint16_t) NO_VAL)
		xstrcat(out, "*:");
	else
		xstrfmtcat(out, "%u:", job_ptr->cores_per_socket);

	if (job_ptr->threads_per_core == (uint16_t) NO_VAL)
		xstrcat(out, "*");
	else
		xstrfmtcat(out, "%u", job_ptr->threads_per_core);

	xstrcat(out, line_end);

	/****** Line 16 ******/
	/* Tres should already of been converted at this point from simple */
	xstrfmtcat(out, "TRES=%s",
		   job_ptr->tres_alloc_str ? job_ptr->tres_alloc_str
					   : job_ptr->tres_req_str);
	xstrcat(out, line_end);

	/****** Line 17 ******/
	if (job_ptr->sockets_per_node == (uint16_t) NO_VAL)
		xstrcat(out, "Socks/Node=* ");
	else
		xstrfmtcat(out, "Socks/Node=%u ", job_ptr->sockets_per_node);

	if (job_ptr->ntasks_per_node == (uint16_t) NO_VAL)
		xstrcat(out, "NtasksPerN:B:S:C=*:");
	else
		xstrfmtcat(out, "NtasksPerN:B:S:C=%u:", job_ptr->ntasks_per_node);

	if (job_ptr->ntasks_per_board == (uint16_t) NO_VAL)
		xstrcat(out, "*:");
	else
		xstrfmtcat(out, "%u:", job_ptr->ntasks_per_board);

	if ((job_ptr->ntasks_per_socket == (uint16_t) NO_VAL) ||
	    (job_ptr->ntasks_per_socket == (uint16_t) INFINITE))
		xstrcat(out, "*:");
	else
		xstrfmtcat(out, "%u:", job_ptr->ntasks_per_socket);

	if ((job_ptr->ntasks_per_core == (uint16_t) NO_VAL) ||
	    (job_ptr->ntasks_per_core == (uint16_t) INFINITE))
		xstrcat(out, "* ");
	else
		xstrfmtcat(out, "%u ", job_ptr->ntasks_per_core);

	if (job_ptr->core_spec == (uint16_t) NO_VAL)
		xstrcat(out, "CoreSpec=*");
	else if (job_ptr->core_spec & CORE_SPEC_THREAD)
		xstrfmtcat(out, "ThreadSpec=%d",
			   (job_ptr->core_spec & (~CORE_SPEC_THREAD)));
	else
		xstrfmtcat(out, "CoreSpec=%u", job_ptr->core_spec);

	xstrcat(out, line_end);

	if (job_resrcs && cluster_flags & CLUSTER_FLAG_BG) {
		if ((job_resrcs->cpu_array_cnt > 0) &&
		    (job_resrcs->cpu_array_value) &&
		    (job_resrcs->cpu_array_reps)) {
			int length = 0;
			xstrcat(out, "CPUs=");
			for (i = 0; i < job_resrcs->cpu_array_cnt; i++) {
				/* only print 60 characters worth of this record */
				if (length > 60) {
					/* skip to last CPU group entry */
					if (i < job_resrcs->cpu_array_cnt - 1) {
						continue;
					}
					/* add ellipsis before last entry */
					xstrcat(out, "...,");
				}

				length += xstrfmtcat(out, "%d", job_resrcs->cpus[i]);
				if (job_resrcs->cpu_array_reps[i] > 1) {
					length += xstrfmtcat(out, "*%d",
							     job_resrcs->cpu_array_reps[i]);
				}
				if (i < job_resrcs->cpu_array_cnt - 1) {
					xstrcat(out, ",");
					length++;
				}
			}
			xstrcat(out, line_end);
		}
	} else if (job_resrcs && job_resrcs->core_bitmap &&
		   ((last = bit_fls(job_resrcs->core_bitmap)) != -1)) {
		hl = hostlist_create(job_resrcs->nodes);
		if (!hl) {
			error("slurm_sprint_job_info: hostlist_create: %s",
			      job_resrcs->nodes);
			return NULL;
		}
		hl_last = hostlist_create(NULL);
		if (!hl_last) {
			error("slurm_sprint_job_info: hostlist_create: NULL");
			hostlist_destroy(hl);
			return NULL;
		}

		bit_inx = 0;
		i = sock_inx = sock_reps = 0;
		abs_node_inx = job_ptr->node_inx[i];

		gres_last = "";
		/* tmp1[] stores the current cpu(s) allocated */
		tmp2[0] = '\0';	/* stores last cpu(s) allocated */
		for (rel_node_inx=0; rel_node_inx < job_resrcs->nhosts;
		     rel_node_inx++) {

			if (sock_reps >=
			    job_resrcs->sock_core_rep_count[sock_inx]) {
				sock_inx++;
				sock_reps = 0;
			}
			sock_reps++;

			bit_reps = job_resrcs->sockets_per_node[sock_inx] *
				   job_resrcs->cores_per_socket[sock_inx];
			host = hostlist_shift(hl);
			threads = _threads_per_core(host);
			cpu_bitmap = bit_alloc(bit_reps * threads);
			for (j = 0; j < bit_reps; j++) {
				if (bit_test(job_resrcs->core_bitmap, bit_inx)){
					for (k = 0; k < threads; k++)
						bit_set(cpu_bitmap,
							(j * threads) + k);
				}
				bit_inx++;
			}
			bit_fmt(tmp1, sizeof(tmp1), cpu_bitmap);
			FREE_NULL_BITMAP(cpu_bitmap);
			/*
			 * If the allocation values for this host are not the
			 * same as the last host, print the report of the last
			 * group of hosts that had identical allocation values.
			 */
			if (xstrcmp(tmp1, tmp2) ||
			    ((rel_node_inx < job_ptr->gres_detail_cnt) &&
			     xstrcmp(job_ptr->gres_detail_str[rel_node_inx],
				     gres_last)) ||
			    (last_mem_alloc_ptr != job_resrcs->memory_allocated) ||
			    (job_resrcs->memory_allocated &&
			     (last_mem_alloc !=
			      job_resrcs->memory_allocated[rel_node_inx]))) {
				if (hostlist_count(hl_last)) {
					last_hosts =
						hostlist_ranged_string_xmalloc(
						hl_last);
					xstrfmtcat(out,
						   "  Nodes=%s CPU_IDs=%s "
						   "Mem=%"PRIu64" GRES_IDX=%s",
						   last_hosts, tmp2,
						   last_mem_alloc_ptr ?
						   last_mem_alloc : 0,
						    gres_last);
					xfree(last_hosts);
					xstrcat(out, line_end);

					hostlist_destroy(hl_last);
					hl_last = hostlist_create(NULL);
				}
				strcpy(tmp2, tmp1);
				if (rel_node_inx < job_ptr->gres_detail_cnt) {
					gres_last = job_ptr->
						    gres_detail_str[rel_node_inx];
				} else {
					gres_last = "";
				}
				last_mem_alloc_ptr = job_resrcs->memory_allocated;
				if (last_mem_alloc_ptr)
					last_mem_alloc = job_resrcs->
						memory_allocated[rel_node_inx];
				else
					last_mem_alloc = NO_VAL64;
			}
			hostlist_push_host(hl_last, host);
			free(host);

			if (bit_inx > last)
				break;

			if (abs_node_inx > job_ptr->node_inx[i+1]) {
				i += 2;
				abs_node_inx = job_ptr->node_inx[i];
			} else {
				abs_node_inx++;
			}
		}

		if (hostlist_count(hl_last)) {
			last_hosts = hostlist_ranged_string_xmalloc(hl_last);
			xstrfmtcat(out, "  Nodes=%s CPU_IDs=%s Mem=%"PRIu64" GRES_IDX=%s",
				 last_hosts, tmp2,
				 last_mem_alloc_ptr ? last_mem_alloc : 0,
				 gres_last);
			xfree(last_hosts);
			xstrcat(out, line_end);
		}
		hostlist_destroy(hl);
		hostlist_destroy(hl_last);
	}
	/****** Line 18 ******/
	if (job_ptr->pn_min_memory & MEM_PER_CPU) {
		job_ptr->pn_min_memory &= (~MEM_PER_CPU);
		tmp6_ptr = "CPU";
	} else
		tmp6_ptr = "Node";

	if (cluster_flags & CLUSTER_FLAG_BG) {
		convert_num_unit((float)job_ptr->pn_min_cpus, tmp1,
				 sizeof(tmp1), UNIT_NONE, NO_VAL,
				 CONVERT_NUM_UNIT_EXACT);
		xstrfmtcat(out, "MinCPUsNode=%s ", tmp1);
	} else {
		xstrfmtcat(out, "MinCPUsNode=%u ", job_ptr->pn_min_cpus);
	}

	convert_num_unit((float)job_ptr->pn_min_memory, tmp1, sizeof(tmp1),
			 UNIT_MEGA, NO_VAL, CONVERT_NUM_UNIT_EXACT);
	convert_num_unit((float)job_ptr->pn_min_tmp_disk, tmp2, sizeof(tmp2),
			 UNIT_MEGA, NO_VAL, CONVERT_NUM_UNIT_EXACT);
	xstrfmtcat(out, "MinMemory%s=%s MinTmpDiskNode=%s", tmp6_ptr, tmp1, tmp2);
	xstrcat(out, line_end);

	/****** Line ******/
	secs2time_str((time_t)job_ptr->delay_boot, tmp1, sizeof(tmp1));
	xstrfmtcat(out, "Features=%s DelayBoot=%s", job_ptr->features, tmp1);
	xstrcat(out, line_end);

	/****** Line ******/
	xstrfmtcat(out, "Gres=%s Reservation=%s",
		   job_ptr->gres, job_ptr->resv_name);
	xstrcat(out, line_end);

	/****** Line 20 ******/
	xstrfmtcat(out, "OverSubscribe=%s Contiguous=%d Licenses=%s Network=%s",
		   job_share_string(job_ptr->shared), job_ptr->contiguous,
		   job_ptr->licenses, job_ptr->network);
	xstrcat(out, line_end);

	/****** Line 21 ******/
	xstrfmtcat(out, "Command=%s", job_ptr->command);
	xstrcat(out, line_end);

	/****** Line 22 ******/
	xstrfmtcat(out, "WorkDir=%s", job_ptr->work_dir);

	if (cluster_flags & CLUSTER_FLAG_BG) {
		/****** Line 23 (optional) ******/
		select_g_select_jobinfo_sprint(job_ptr->select_jobinfo,
					       select_buf, sizeof(select_buf),
					       SELECT_PRINT_BG_ID);
		if (select_buf[0] != '\0') {
			xstrcat(out, line_end);
			xstrfmtcat(out, "Block_ID=%s", select_buf);
		}

		/****** Line 24 (optional) ******/
		select_g_select_jobinfo_sprint(job_ptr->select_jobinfo,
					       select_buf, sizeof(select_buf),
					       SELECT_PRINT_MIXED_SHORT);
		if (select_buf[0] != '\0') {
			xstrcat(out, line_end);
			xstrcat(out, select_buf);
		}

		/****** Line 26 (optional) ******/
		select_g_select_jobinfo_sprint(job_ptr->select_jobinfo,
					       select_buf, sizeof(select_buf),
					       SELECT_PRINT_LINUX_IMAGE);
		if (select_buf[0] != '\0') {
			xstrcat(out, line_end);
			xstrfmtcat(out, "CnloadImage=%s", select_buf);
		}
		/****** Line 27 (optional) ******/
		select_g_select_jobinfo_sprint(job_ptr->select_jobinfo,
					       select_buf, sizeof(select_buf),
					       SELECT_PRINT_MLOADER_IMAGE);
		if (select_buf[0] != '\0') {
			xstrcat(out, line_end);
			xstrfmtcat(out, "MloaderImage=%s", select_buf);
		}
		/****** Line 28 (optional) ******/
		select_g_select_jobinfo_sprint(job_ptr->select_jobinfo,
					       select_buf, sizeof(select_buf),
					       SELECT_PRINT_RAMDISK_IMAGE);
		if (select_buf[0] != '\0') {
			xstrcat(out, line_end);
			xstrfmtcat(out, "IoloadImage=%s", select_buf);
		}
	}

	/****** Line (optional) ******/
	if (job_ptr->admin_comment) {
		xstrcat(out, line_end);
		xstrfmtcat(out, "AdminComment=%s ", job_ptr->admin_comment);
	}

	/****** Line (optional) ******/
	if (job_ptr->comment) {
		xstrcat(out, line_end);
		xstrfmtcat(out, "Comment=%s ", job_ptr->comment);
	}

	/****** Line 30 (optional) ******/
	if (job_ptr->batch_flag) {
		xstrcat(out, line_end);
		slurm_get_job_stderr(tmp_path, sizeof(tmp_path), job_ptr);
		xstrfmtcat(out, "StdErr=%s", tmp_path);
	}

	/****** Line 31 (optional) ******/
	if (job_ptr->batch_flag) {
		xstrcat(out, line_end);
		slurm_get_job_stdin(tmp_path, sizeof(tmp_path), job_ptr);
		xstrfmtcat(out, "StdIn=%s", tmp_path);
	}

	/****** Line 32 (optional) ******/
	if (job_ptr->batch_flag) {
		xstrcat(out, line_end);
		slurm_get_job_stdout(tmp_path, sizeof(tmp_path), job_ptr);
		xstrfmtcat(out, "StdOut=%s", tmp_path);
	}

	/****** Line 33 (optional) ******/
	if (job_ptr->batch_script) {
		xstrcat(out, line_end);
		xstrcat(out, "BatchScript=\n");
		xstrcat(out, job_ptr->batch_script);
	}

	/****** Line 34 (optional) ******/
	if (job_ptr->req_switch) {
		char time_buf[32];
		xstrcat(out, line_end);
		secs2time_str((time_t) job_ptr->wait4switch, time_buf,
			      sizeof(time_buf));
		xstrfmtcat(out, "Switches=%u@%s\n", job_ptr->req_switch, time_buf);
	}

	/****** Line 35 (optional) ******/
	if (job_ptr->burst_buffer) {
		xstrcat(out, line_end);
		xstrfmtcat(out, "BurstBuffer=%s", job_ptr->burst_buffer);
	}

	/****** Line (optional) ******/
	if (job_ptr->burst_buffer_state) {
		xstrcat(out, line_end);
		xstrfmtcat(out, "BurstBufferState=%s",
			   job_ptr->burst_buffer_state);
	}

	/****** Line 36 (optional) ******/
	if (cpu_freq_debug(NULL, NULL, tmp1, sizeof(tmp1),
			   job_ptr->cpu_freq_gov, job_ptr->cpu_freq_min,
			   job_ptr->cpu_freq_max, NO_VAL) != 0) {
		xstrcat(out, line_end);
		xstrcat(out, tmp1);
	}

	/****** Line 37 ******/
	xstrcat(out, line_end);
	xstrfmtcat(out, "Power=%s", power_flags_str(job_ptr->power_flags));

	/****** Line 38 (optional) ******/
	if (job_ptr->bitflags) {
		xstrcat(out, line_end);
		if (job_ptr->bitflags & GRES_ENFORCE_BIND)
			xstrcat(out, "GresEnforceBind=Yes");
		if (job_ptr->bitflags & KILL_INV_DEP)
			xstrcat(out, "KillOInInvalidDependent=Yes");
		if (job_ptr->bitflags & NO_KILL_INV_DEP)
			xstrcat(out, "KillOInInvalidDependent=No");
		if (job_ptr->bitflags & SPREAD_JOB)
			xstrcat(out, "SpreadJob=Yes");
	}

	/****** END OF JOB RECORD ******/
	if (one_liner)
		xstrcat(out, "\n");
	else
		xstrcat(out, "\n\n");

	return out;
}
Example #2
0
extern int slurm_jobcomp_log_record ( struct job_record *job_ptr )
{
	int rc = SLURM_SUCCESS;
	char job_rec[1024];
	char usr_str[32], grp_str[32], start_str[32], end_str[32], lim_str[32];
	char select_buf[128], *state_string, *work_dir;
	size_t offset = 0, tot_size, wrote;
	uint32_t job_state;
	uint32_t time_limit;

	if ((log_name == NULL) || (job_comp_fd < 0)) {
		error("JobCompLoc log file %s not open", log_name);
		return SLURM_ERROR;
	}

	slurm_mutex_lock( &file_lock );
	_get_user_name(job_ptr->user_id, usr_str, sizeof(usr_str));
	_get_group_name(job_ptr->group_id, grp_str, sizeof(grp_str));

	if ((job_ptr->time_limit == NO_VAL) && job_ptr->part_ptr)
		time_limit = job_ptr->part_ptr->max_time;
	else
		time_limit = job_ptr->time_limit;
	if (time_limit == INFINITE)
		strcpy(lim_str, "UNLIMITED");
	else {
		snprintf(lim_str, sizeof(lim_str), "%lu",
			 (unsigned long) time_limit);
	}

	if (job_ptr->job_state & JOB_RESIZING) {
		time_t now = time(NULL);
		state_string = job_state_string(job_ptr->job_state);
		if (job_ptr->resize_time) {
			_make_time_str(&job_ptr->resize_time, start_str,
				       sizeof(start_str));
		} else {
			_make_time_str(&job_ptr->start_time, start_str,
				       sizeof(start_str));
		}
		_make_time_str(&now, end_str, sizeof(end_str));
	} else {
		/* Job state will typically have JOB_COMPLETING or JOB_RESIZING
		 * flag set when called. We remove the flags to get the eventual
		 * completion state: JOB_FAILED, JOB_TIMEOUT, etc. */
		job_state = job_ptr->job_state & JOB_STATE_BASE;
		state_string = job_state_string(job_state);
		if (job_ptr->resize_time) {
			_make_time_str(&job_ptr->resize_time, start_str,
				       sizeof(start_str));
		} else if (job_ptr->start_time > job_ptr->end_time) {
			/* Job cancelled while pending and
			 * expected start time is in the future. */
			snprintf(start_str, sizeof(start_str), "Unknown");
		} else {
			_make_time_str(&job_ptr->start_time, start_str,
				       sizeof(start_str));
		}
		_make_time_str(&job_ptr->end_time, end_str, sizeof(end_str));
	}

	if (job_ptr->details && job_ptr->details->work_dir)
		work_dir = job_ptr->details->work_dir;
	else
		work_dir = "unknown";

	select_g_select_jobinfo_sprint(job_ptr->select_jobinfo,
		select_buf, sizeof(select_buf), SELECT_PRINT_MIXED);

	snprintf(job_rec, sizeof(job_rec), JOB_FORMAT,
		 (unsigned long) job_ptr->job_id, usr_str,
		 (unsigned long) job_ptr->user_id, grp_str,
		 (unsigned long) job_ptr->group_id, job_ptr->name,
		 state_string, job_ptr->partition, lim_str, start_str,
		 end_str, job_ptr->nodes, job_ptr->node_cnt,
		 job_ptr->total_cpus, work_dir,
		 select_buf);
	tot_size = strlen(job_rec);

	while ( offset < tot_size ) {
		wrote = write(job_comp_fd, job_rec + offset,
			tot_size - offset);
		if (wrote == -1) {
			if (errno == EAGAIN)
				continue;
			else {
				plugin_errno = errno;
				rc = SLURM_ERROR;
				break;
			}
		}
		offset += wrote;
	}
	slurm_mutex_unlock( &file_lock );
	return rc;
}
Example #3
0
/*
 * slurm_sprint_job_info - output information about a specific Slurm
 *	job based upon message as loaded using slurm_load_jobs
 * IN job_ptr - an individual job information record pointer
 * IN one_liner - print as a single line if true
 * RET out - char * containing formatted output (must be freed after call)
 *           NULL is returned on failure.
 */
extern char *
slurm_sprint_job_info ( job_info_t * job_ptr, int one_liner )
{
	int i, j;
	char time_str[32], *group_name, *user_name;
	char tmp1[128], tmp2[128], tmp3[128], tmp4[128], tmp5[128], tmp6[128];
	char *tmp6_ptr;
	char tmp_line[512];
	char *ionodes = NULL;
	uint16_t exit_status = 0, term_sig = 0;
	job_resources_t *job_resrcs = job_ptr->job_resrcs;
	char *out = NULL;
	time_t run_time;
	uint32_t min_nodes, max_nodes = 0;
	char *nodelist = "NodeList";
	bitstr_t *core_bitmap;
	char *host;
	int sock_inx, sock_reps, last;
	int abs_node_inx, rel_node_inx;
	int nice;
	int bit_inx, bit_reps;
	uint32_t *last_mem_alloc_ptr = NULL;
	uint32_t last_mem_alloc = NO_VAL;
	char *last_hosts;
	hostlist_t hl, hl_last;
	char select_buf[122];
	uint32_t cluster_flags = slurmdb_setup_cluster_flags();

	if (cluster_flags & CLUSTER_FLAG_BG) {
		nodelist = "MidplaneList";
		select_g_select_jobinfo_get(job_ptr->select_jobinfo,
					    SELECT_JOBDATA_IONODES,
					    &ionodes);
	}

	/****** Line 1 ******/
	snprintf(tmp_line, sizeof(tmp_line), "JobId=%u ", job_ptr->job_id);
	out = xstrdup(tmp_line);
	if (job_ptr->array_job_id) {
		snprintf(tmp_line, sizeof(tmp_line),
			 "ArrayJobId=%u ArrayTaskId=%u ",
			 job_ptr->array_job_id, job_ptr->array_task_id);
		xstrcat(out, tmp_line);
	}
	snprintf(tmp_line, sizeof(tmp_line), "Name=%s", job_ptr->name);
	xstrcat(out, tmp_line);
	if (one_liner)
		xstrcat(out, " ");
	else
		xstrcat(out, "\n   ");

	/****** Line 2 ******/
	user_name = uid_to_string((uid_t) job_ptr->user_id);
	group_name = gid_to_string((gid_t) job_ptr->group_id);
	snprintf(tmp_line, sizeof(tmp_line),
		 "UserId=%s(%u) GroupId=%s(%u)",
		 user_name, job_ptr->user_id, group_name, job_ptr->group_id);
	xfree(user_name);
	xfree(group_name);
	xstrcat(out, tmp_line);
	if (one_liner)
		xstrcat(out, " ");
	else
		xstrcat(out, "\n   ");

	/****** Line 3 ******/
	nice  = job_ptr->nice;
	nice -= NICE_OFFSET;
	snprintf(tmp_line, sizeof(tmp_line),
		 "Priority=%u Nice=%d Account=%s QOS=%s",
		 job_ptr->priority, nice, job_ptr->account, job_ptr->qos);
	xstrcat(out, tmp_line);
	if (slurm_get_track_wckey()) {
		snprintf(tmp_line, sizeof(tmp_line),
			 " WCKey=%s", job_ptr->wckey);
		xstrcat(out, tmp_line);
	}
	if (one_liner)
		xstrcat(out, " ");
	else
		xstrcat(out, "\n   ");

	/****** Line 4 ******/
	if (job_ptr->state_desc) {
		/* Replace white space with underscore for easier parsing */
		for (j=0; job_ptr->state_desc[j]; j++) {
			if (isspace((int)job_ptr->state_desc[j]))
				job_ptr->state_desc[j] = '_';
		}
		tmp6_ptr = job_ptr->state_desc;
	} else
		tmp6_ptr = job_reason_string(job_ptr->state_reason);
	snprintf(tmp_line, sizeof(tmp_line),
		 "JobState=%s Reason=%s Dependency=%s",
		 job_state_string(job_ptr->job_state), tmp6_ptr,
		 job_ptr->dependency);
	xstrcat(out, tmp_line);
	if (one_liner)
		xstrcat(out, " ");
	else
		xstrcat(out, "\n   ");

	/****** Line 5 ******/
	snprintf(tmp_line, sizeof(tmp_line),
		 "Requeue=%u Restarts=%u BatchFlag=%u ",
		 job_ptr->requeue, job_ptr->restart_cnt, job_ptr->batch_flag);
	xstrcat(out, tmp_line);
	if (WIFSIGNALED(job_ptr->exit_code))
		term_sig = WTERMSIG(job_ptr->exit_code);
	exit_status = WEXITSTATUS(job_ptr->exit_code);
	snprintf(tmp_line, sizeof(tmp_line),
		 "ExitCode=%u:%u", exit_status, term_sig);
	xstrcat(out, tmp_line);
	if (one_liner)
		xstrcat(out, " ");
	else
		xstrcat(out, "\n   ");

	/****** Line 5a (optional) ******/
	if (!(job_ptr->show_flags & SHOW_DETAIL))
		goto line6;
	if (WIFSIGNALED(job_ptr->derived_ec))
		term_sig = WTERMSIG(job_ptr->derived_ec);
	else
		term_sig = 0;
	exit_status = WEXITSTATUS(job_ptr->derived_ec);
	snprintf(tmp_line, sizeof(tmp_line),
		 "DerivedExitCode=%u:%u", exit_status, term_sig);
	xstrcat(out, tmp_line);
	if (one_liner)
		xstrcat(out, " ");
	else
		xstrcat(out, "\n   ");

	/****** Line 6 ******/
line6:
	snprintf(tmp_line, sizeof(tmp_line), "RunTime=");
	xstrcat(out, tmp_line);
	if (IS_JOB_PENDING(job_ptr))
		run_time = 0;
	else if (IS_JOB_SUSPENDED(job_ptr))
		run_time = job_ptr->pre_sus_time;
	else {
		time_t end_time;
		if (IS_JOB_RUNNING(job_ptr) || (job_ptr->end_time == 0))
			end_time = time(NULL);
		else
			end_time = job_ptr->end_time;
		if (job_ptr->suspend_time) {
			run_time = (time_t)
				(difftime(end_time, job_ptr->suspend_time)
				 + job_ptr->pre_sus_time);
		} else
			run_time = (time_t)
				difftime(end_time, job_ptr->start_time);
	}
	secs2time_str(run_time, tmp1, sizeof(tmp1));
	sprintf(tmp_line, "%s ", tmp1);
	xstrcat(out, tmp_line);

	snprintf(tmp_line, sizeof(tmp_line), "TimeLimit=");
	xstrcat(out, tmp_line);
	if (job_ptr->time_limit == NO_VAL)
		sprintf(tmp_line, "Partition_Limit");
	else {
		mins2time_str(job_ptr->time_limit, tmp_line,
			      sizeof(tmp_line));
	}
	xstrcat(out, tmp_line);
	snprintf(tmp_line, sizeof(tmp_line), " TimeMin=");
	xstrcat(out, tmp_line);
	if (job_ptr->time_min == 0)
		sprintf(tmp_line, "N/A");
	else {
		mins2time_str(job_ptr->time_min, tmp_line,
			      sizeof(tmp_line));
	}
	xstrcat(out, tmp_line);
	if (one_liner)
		xstrcat(out, " ");
	else
		xstrcat(out, "\n   ");

	/****** Line 7 ******/
	slurm_make_time_str((time_t *)&job_ptr->submit_time, time_str,
			    sizeof(time_str));
	snprintf(tmp_line, sizeof(tmp_line), "SubmitTime=%s ", time_str);
	xstrcat(out, tmp_line);

	slurm_make_time_str((time_t *)&job_ptr->eligible_time, time_str,
			    sizeof(time_str));
	snprintf(tmp_line, sizeof(tmp_line), "EligibleTime=%s", time_str);
	xstrcat(out, tmp_line);
	if (one_liner)
		xstrcat(out, " ");
	else
		xstrcat(out, "\n   ");

	/****** Line 8 (optional) ******/
	if (job_ptr->resize_time) {
		slurm_make_time_str((time_t *)&job_ptr->resize_time, time_str,
				    sizeof(time_str));
		snprintf(tmp_line, sizeof(tmp_line), "ResizeTime=%s", time_str);
		xstrcat(out, tmp_line);
		if (one_liner)
			xstrcat(out, " ");
		else
			xstrcat(out, "\n   ");
	}

	/****** Line 9 ******/
	slurm_make_time_str((time_t *)&job_ptr->start_time, time_str,
			    sizeof(time_str));
	snprintf(tmp_line, sizeof(tmp_line), "StartTime=%s ", time_str);
	xstrcat(out, tmp_line);

	snprintf(tmp_line, sizeof(tmp_line), "EndTime=");
	xstrcat(out, tmp_line);
	if ((job_ptr->time_limit == INFINITE) &&
	    (job_ptr->end_time > time(NULL)))
		sprintf(tmp_line, "Unknown");
	else {
		slurm_make_time_str ((time_t *)&job_ptr->end_time, time_str,
				     sizeof(time_str));
		sprintf(tmp_line, "%s", time_str);
	}
	xstrcat(out, tmp_line);
	if (one_liner)
		xstrcat(out, " ");
	else
		xstrcat(out, "\n   ");

	/****** Line 10 ******/
	if (job_ptr->preempt_time == 0)
		sprintf(tmp_line, "PreemptTime=None ");
	else {
		slurm_make_time_str((time_t *)&job_ptr->preempt_time,
				    time_str, sizeof(time_str));
		snprintf(tmp_line, sizeof(tmp_line), "PreemptTime=%s ",
			 time_str);
	}
	xstrcat(out, tmp_line);
	if (job_ptr->suspend_time) {
		slurm_make_time_str ((time_t *)&job_ptr->suspend_time,
				     time_str, sizeof(time_str));
	} else {
		strncpy(time_str, "None", sizeof(time_str));
	}
	snprintf(tmp_line, sizeof(tmp_line),
		 "SuspendTime=%s SecsPreSuspend=%ld",
		 time_str, (long int)job_ptr->pre_sus_time);
	xstrcat(out, tmp_line);
	if (one_liner)
		xstrcat(out, " ");
	else
		xstrcat(out, "\n   ");

	/****** Line 11 ******/
	snprintf(tmp_line, sizeof(tmp_line),
		 "Partition=%s AllocNode:Sid=%s:%u",
		 job_ptr->partition, job_ptr->alloc_node, job_ptr->alloc_sid);
	xstrcat(out, tmp_line);
	if (one_liner)
		xstrcat(out, " ");
	else
		xstrcat(out, "\n   ");

	/****** Line 12 ******/
	snprintf(tmp_line, sizeof(tmp_line), "Req%s=%s Exc%s=%s",
		 nodelist, job_ptr->req_nodes, nodelist, job_ptr->exc_nodes);
	xstrcat(out, tmp_line);
	if (one_liner)
		xstrcat(out, " ");
	else
		xstrcat(out, "\n   ");

	/****** Line 13 ******/
	xstrfmtcat(out, "%s=", nodelist);
	xstrcat(out, job_ptr->nodes);
	if (job_ptr->nodes && ionodes) {
		snprintf(tmp_line, sizeof(tmp_line), "[%s]", ionodes);
		xstrcat(out, tmp_line);
		xfree(ionodes);
	}
	if (one_liner)
		xstrcat(out, " ");
	else
		xstrcat(out, "\n   ");

	/****** Line 14 (optional) ******/
	if (job_ptr->batch_host) {
		snprintf(tmp_line, sizeof(tmp_line), "BatchHost=%s",
			 job_ptr->batch_host);
		xstrcat(out, tmp_line);
		if (one_liner)
			xstrcat(out, " ");
		else
			xstrcat(out, "\n   ");
	}

	/****** Line 15 ******/
	if (cluster_flags & CLUSTER_FLAG_BG) {
		select_g_select_jobinfo_get(job_ptr->select_jobinfo,
					    SELECT_JOBDATA_NODE_CNT,
					    &min_nodes);
		if ((min_nodes == 0) || (min_nodes == NO_VAL)) {
			min_nodes = job_ptr->num_nodes;
			max_nodes = job_ptr->max_nodes;
		} else if (job_ptr->max_nodes)
			max_nodes = min_nodes;
	} else {
		min_nodes = job_ptr->num_nodes;
		if ((min_nodes == 1) && (job_ptr->num_cpus > 1)
		    && job_ptr->ntasks_per_node
		    && (job_ptr->ntasks_per_node != (uint16_t) NO_VAL)) {
			int num_tasks = job_ptr->num_cpus;
			if (job_ptr->cpus_per_task != (uint16_t) NO_VAL)
				num_tasks /= job_ptr->cpus_per_task;
			min_nodes = (num_tasks + 1) / job_ptr->ntasks_per_node;
			if (min_nodes > num_tasks)
				min_nodes = num_tasks;
			else if (!min_nodes)
				min_nodes = 1;
		}
		max_nodes = job_ptr->max_nodes;
	}

	_sprint_range(tmp1, sizeof(tmp1), job_ptr->num_cpus, job_ptr->max_cpus);
	_sprint_range(tmp2, sizeof(tmp2), min_nodes, max_nodes);
	if (job_ptr->boards_per_node == (uint16_t) NO_VAL)
		strcpy(tmp3, "*");
	else
		snprintf(tmp3, sizeof(tmp3), "%u", job_ptr->boards_per_node);
	if (job_ptr->sockets_per_board == (uint16_t) NO_VAL)
		strcpy(tmp4, "*");
	else
		snprintf(tmp4, sizeof(tmp4), "%u", job_ptr->sockets_per_board);
	if (job_ptr->cores_per_socket == (uint16_t) NO_VAL)
		strcpy(tmp5, "*");
	else
		snprintf(tmp5, sizeof(tmp5), "%u", job_ptr->cores_per_socket);
	if (job_ptr->threads_per_core == (uint16_t) NO_VAL)
		strcpy(tmp6, "*");
	else
		snprintf(tmp6, sizeof(tmp6), "%u", job_ptr->threads_per_core);
	snprintf(tmp_line, sizeof(tmp_line),
		 "NumNodes=%s NumCPUs=%s CPUs/Task=%u ReqB:S:C:T=%s:%s:%s:%s",
		 tmp2, tmp1, job_ptr->cpus_per_task, tmp3, tmp4, tmp5, tmp6);
	xstrcat(out, tmp_line);
	if (one_liner)
		xstrcat(out, " ");
	else
		xstrcat(out, "\n   ");

	/****** Line 16 ******/
	if (job_ptr->sockets_per_node == (uint16_t) NO_VAL)
		strcpy(tmp1, "*");
	else
		snprintf(tmp1, sizeof(tmp1), "%u", job_ptr->sockets_per_node);
	if (job_ptr->ntasks_per_node == (uint16_t) NO_VAL)
		strcpy(tmp2, "*");
	else
		snprintf(tmp2, sizeof(tmp2), "%u", job_ptr->ntasks_per_node);
	if (job_ptr->ntasks_per_board == (uint16_t) NO_VAL)
		strcpy(tmp3, "*");
	else
		snprintf(tmp3, sizeof(tmp3), "%u", job_ptr->ntasks_per_board);
	if ((job_ptr->ntasks_per_socket == (uint16_t) NO_VAL) ||
	    (job_ptr->ntasks_per_socket == (uint16_t) INFINITE))
		strcpy(tmp4, "*");
	else
		snprintf(tmp4, sizeof(tmp4), "%u", job_ptr->ntasks_per_socket);
	if ((job_ptr->ntasks_per_core == (uint16_t) NO_VAL) ||
	    (job_ptr->ntasks_per_core == (uint16_t) INFINITE))
		strcpy(tmp5, "*");
	else
		snprintf(tmp5, sizeof(tmp5), "%u", job_ptr->ntasks_per_core);
	snprintf(tmp_line, sizeof(tmp_line),
		 "Socks/Node=%s NtasksPerN:B:S:C=%s:%s:%s:%s CoreSpec=%u",
		 tmp1, tmp2, tmp3, tmp4, tmp5, job_ptr->core_spec);
	xstrcat(out, tmp_line);
	if (one_liner)
		xstrcat(out, " ");
	else
		xstrcat(out, "\n   ");

	if (!job_resrcs)
		goto line15;

	if (cluster_flags & CLUSTER_FLAG_BG) {
		if ((job_resrcs->cpu_array_cnt > 0) &&
		    (job_resrcs->cpu_array_value) &&
		    (job_resrcs->cpu_array_reps)) {
			int length = 0;
			xstrcat(out, "CPUs=");
			length += 10;
			for (i = 0; i < job_resrcs->cpu_array_cnt; i++) {
				if (length > 70) {
					/* skip to last CPU group entry */
					if (i < job_resrcs->cpu_array_cnt - 1) {
						continue;
					}
					/* add ellipsis before last entry */
					xstrcat(out, "...,");
					length += 4;
				}

				snprintf(tmp_line, sizeof(tmp_line), "%d",
					 job_resrcs->cpus[i]);
				xstrcat(out, tmp_line);
				length += strlen(tmp_line);
				if (job_resrcs->cpu_array_reps[i] > 1) {
					snprintf(tmp_line, sizeof(tmp_line),
						 "*%d",
						 job_resrcs->cpu_array_reps[i]);
					xstrcat(out, tmp_line);
					length += strlen(tmp_line);
				}
				if (i < job_resrcs->cpu_array_cnt - 1) {
					xstrcat(out, ",");
					length++;
				}
			}
			if (one_liner)
				xstrcat(out, " ");
			else
				xstrcat(out, "\n   ");
		}
	} else {
		if (!job_resrcs->core_bitmap)
			goto line15;

		last  = bit_fls(job_resrcs->core_bitmap);
		if (last == -1)
			goto line15;

		hl = hostlist_create(job_resrcs->nodes);
		if (!hl) {
			error("slurm_sprint_job_info: hostlist_create: %s",
			      job_resrcs->nodes);
			return NULL;
		}
		hl_last = hostlist_create(NULL);
		if (!hl_last) {
			error("slurm_sprint_job_info: hostlist_create: NULL");
			hostlist_destroy(hl);
			return NULL;
		}

		bit_inx = 0;
		i = sock_inx = sock_reps = 0;
		abs_node_inx = job_ptr->node_inx[i];

/*	tmp1[] stores the current cpu(s) allocated	*/
		tmp2[0] = '\0';	/* stores last cpu(s) allocated */
		for (rel_node_inx=0; rel_node_inx < job_resrcs->nhosts;
		     rel_node_inx++) {

			if (sock_reps >=
			    job_resrcs->sock_core_rep_count[sock_inx]) {
				sock_inx++;
				sock_reps = 0;
			}
			sock_reps++;

			bit_reps = job_resrcs->sockets_per_node[sock_inx] *
				job_resrcs->cores_per_socket[sock_inx];

			core_bitmap = bit_alloc(bit_reps);
			for (j=0; j < bit_reps; j++) {
				if (bit_test(job_resrcs->core_bitmap, bit_inx))
					bit_set(core_bitmap, j);
				bit_inx++;
			}

			bit_fmt(tmp1, sizeof(tmp1), core_bitmap);
			FREE_NULL_BITMAP(core_bitmap);
			host = hostlist_shift(hl);
/*
 *		If the allocation values for this host are not the same as the
 *		last host, print the report of the last group of hosts that had
 *		identical allocation values.
 */
			if (strcmp(tmp1, tmp2) ||
			    (last_mem_alloc_ptr != job_resrcs->memory_allocated) ||
			    (job_resrcs->memory_allocated &&
			     (last_mem_alloc !=
			      job_resrcs->memory_allocated[rel_node_inx]))) {
				if (hostlist_count(hl_last)) {
					last_hosts =
						hostlist_ranged_string_xmalloc(
						hl_last);
					snprintf(tmp_line, sizeof(tmp_line),
						 "  Nodes=%s CPU_IDs=%s Mem=%u",
						 last_hosts, tmp2,
						 last_mem_alloc_ptr ?
						 last_mem_alloc : 0);
					xfree(last_hosts);
					xstrcat(out, tmp_line);
					if (one_liner)
						xstrcat(out, " ");
					else
						xstrcat(out, "\n   ");

					hostlist_destroy(hl_last);
					hl_last = hostlist_create(NULL);
				}
				strcpy(tmp2, tmp1);
				last_mem_alloc_ptr = job_resrcs->memory_allocated;
				if (last_mem_alloc_ptr)
					last_mem_alloc = job_resrcs->
						memory_allocated[rel_node_inx];
				else
					last_mem_alloc = NO_VAL;
			}
			hostlist_push_host(hl_last, host);
			free(host);

			if (bit_inx > last)
				break;

			if (abs_node_inx > job_ptr->node_inx[i+1]) {
				i += 2;
				abs_node_inx = job_ptr->node_inx[i];
			} else {
				abs_node_inx++;
			}
		}

		if (hostlist_count(hl_last)) {
			last_hosts = hostlist_ranged_string_xmalloc(hl_last);
			snprintf(tmp_line, sizeof(tmp_line),
				 "  Nodes=%s CPU_IDs=%s Mem=%u",
				 last_hosts, tmp2,
				 last_mem_alloc_ptr ? last_mem_alloc : 0);
			xfree(last_hosts);
			xstrcat(out, tmp_line);
			if (one_liner)
				xstrcat(out, " ");
			else
				xstrcat(out, "\n   ");
		}
		hostlist_destroy(hl);
		hostlist_destroy(hl_last);
	}
	/****** Line 17 ******/
line15:
	if (job_ptr->pn_min_memory & MEM_PER_CPU) {
		job_ptr->pn_min_memory &= (~MEM_PER_CPU);
		tmp6_ptr = "CPU";
	} else
		tmp6_ptr = "Node";

	if (cluster_flags & CLUSTER_FLAG_BG) {
		convert_num_unit((float)job_ptr->pn_min_cpus,
				 tmp1, sizeof(tmp1), UNIT_NONE);
		snprintf(tmp_line, sizeof(tmp_line), "MinCPUsNode=%s",	tmp1);
	} else {
		snprintf(tmp_line, sizeof(tmp_line), "MinCPUsNode=%u",
			 job_ptr->pn_min_cpus);
	}

	xstrcat(out, tmp_line);
	convert_num_unit((float)job_ptr->pn_min_memory, tmp1, sizeof(tmp1),
			 UNIT_MEGA);
	convert_num_unit((float)job_ptr->pn_min_tmp_disk, tmp2, sizeof(tmp2),
			 UNIT_MEGA);
	snprintf(tmp_line, sizeof(tmp_line),
		 " MinMemory%s=%s MinTmpDiskNode=%s",
		 tmp6_ptr, tmp1, tmp2);
	xstrcat(out, tmp_line);
	if (one_liner)
		xstrcat(out, " ");
	else
		xstrcat(out, "\n   ");

	/****** Line 18 ******/
	snprintf(tmp_line, sizeof(tmp_line),
		 "Features=%s Gres=%s Reservation=%s",
		 job_ptr->features, job_ptr->gres, job_ptr->resv_name);
	xstrcat(out, tmp_line);
	if (one_liner)
		xstrcat(out, " ");
	else
		xstrcat(out, "\n   ");

	/****** Line 19 ******/
	snprintf(tmp_line, sizeof(tmp_line),
		 "Shared=%s Contiguous=%d Licenses=%s Network=%s",
		 (job_ptr->shared == 0 ? "0" :
		  job_ptr->shared == 1 ? "1" : "OK"),
		 job_ptr->contiguous, job_ptr->licenses, job_ptr->network);
	xstrcat(out, tmp_line);
	if (one_liner)
		xstrcat(out, " ");
	else
		xstrcat(out, "\n   ");

	/****** Line 20 ******/
	snprintf(tmp_line, sizeof(tmp_line), "Command=%s",
		 job_ptr->command);
	xstrcat(out, tmp_line);
	if (one_liner)
		xstrcat(out, " ");
	else
		xstrcat(out, "\n   ");

	/****** Line 21 ******/
	snprintf(tmp_line, sizeof(tmp_line), "WorkDir=%s",
		 job_ptr->work_dir);
	xstrcat(out, tmp_line);

	if (cluster_flags & CLUSTER_FLAG_BG) {
		/****** Line 22 (optional) ******/
		select_g_select_jobinfo_sprint(job_ptr->select_jobinfo,
					       select_buf, sizeof(select_buf),
					       SELECT_PRINT_BG_ID);
		if (select_buf[0] != '\0') {
			if (one_liner)
				xstrcat(out, " ");
			else
				xstrcat(out, "\n   ");
			snprintf(tmp_line, sizeof(tmp_line),
				 "Block_ID=%s", select_buf);
			xstrcat(out, tmp_line);
		}

		/****** Line 23 (optional) ******/
		select_g_select_jobinfo_sprint(job_ptr->select_jobinfo,
					       select_buf, sizeof(select_buf),
					       SELECT_PRINT_MIXED_SHORT);
		if (select_buf[0] != '\0') {
			if (one_liner)
				xstrcat(out, " ");
			else
				xstrcat(out, "\n   ");
			xstrcat(out, select_buf);
		}

		if (cluster_flags & CLUSTER_FLAG_BGL) {
			/****** Line 24 (optional) ******/
			select_g_select_jobinfo_sprint(
				job_ptr->select_jobinfo,
				select_buf, sizeof(select_buf),
				SELECT_PRINT_BLRTS_IMAGE);
			if (select_buf[0] != '\0') {
				if (one_liner)
					xstrcat(out, " ");
				else
					xstrcat(out, "\n   ");
				snprintf(tmp_line, sizeof(tmp_line),
					 "BlrtsImage=%s", select_buf);
				xstrcat(out, tmp_line);
			}
		}
		/****** Line 25 (optional) ******/
		select_g_select_jobinfo_sprint(job_ptr->select_jobinfo,
					       select_buf, sizeof(select_buf),
					       SELECT_PRINT_LINUX_IMAGE);
		if (select_buf[0] != '\0') {
			if (one_liner)
				xstrcat(out, " ");
			else
				xstrcat(out, "\n   ");
			if (cluster_flags & CLUSTER_FLAG_BGL)
				snprintf(tmp_line, sizeof(tmp_line),
					 "LinuxImage=%s", select_buf);
			else
				snprintf(tmp_line, sizeof(tmp_line),
					 "CnloadImage=%s", select_buf);

			xstrcat(out, tmp_line);
		}
		/****** Line 26 (optional) ******/
		select_g_select_jobinfo_sprint(job_ptr->select_jobinfo,
					       select_buf, sizeof(select_buf),
					       SELECT_PRINT_MLOADER_IMAGE);
		if (select_buf[0] != '\0') {
			if (one_liner)
				xstrcat(out, " ");
			else
				xstrcat(out, "\n   ");
			snprintf(tmp_line, sizeof(tmp_line),
				 "MloaderImage=%s", select_buf);
			xstrcat(out, tmp_line);
		}
		/****** Line 27 (optional) ******/
		select_g_select_jobinfo_sprint(job_ptr->select_jobinfo,
					       select_buf, sizeof(select_buf),
					       SELECT_PRINT_RAMDISK_IMAGE);
		if (select_buf[0] != '\0') {
			if (one_liner)
				xstrcat(out, " ");
			else
				xstrcat(out, "\n   ");
			if (cluster_flags & CLUSTER_FLAG_BGL)
				snprintf(tmp_line, sizeof(tmp_line),
					 "RamDiskImage=%s", select_buf);
			else
				snprintf(tmp_line, sizeof(tmp_line),
					 "IoloadImage=%s", select_buf);
			xstrcat(out, tmp_line);
		}
	}

	/****** Line 28 (optional) ******/
	if (job_ptr->comment) {
		if (one_liner)
			xstrcat(out, " ");
		else
			xstrcat(out, "\n   ");
		snprintf(tmp_line, sizeof(tmp_line), "Comment=%s ",
			 job_ptr->comment);
		xstrcat(out, tmp_line);
	}

	/****** Line 29 (optional) ******/
	if (job_ptr->batch_flag) {
		if (one_liner)
			xstrcat(out, " ");
		else
			xstrcat(out, "\n   ");
		slurm_get_job_stderr(tmp1, sizeof(tmp1), job_ptr);
		xstrfmtcat(out, "StdErr=%s", tmp1);
	}

	/****** Line 30 (optional) ******/
	if (job_ptr->batch_flag) {
		if (one_liner)
			xstrcat(out, " ");
		else
			xstrcat(out, "\n   ");
		slurm_get_job_stdin(tmp1, sizeof(tmp1), job_ptr);
		xstrfmtcat(out, "StdIn=%s", tmp1);
	}

	/****** Line 31 (optional) ******/
	if (job_ptr->batch_flag) {
		if (one_liner)
			xstrcat(out, " ");
		else
			xstrcat(out, "\n   ");
		slurm_get_job_stdout(tmp1, sizeof(tmp1), job_ptr);
		xstrfmtcat(out, "StdOut=%s", tmp1);
	}

	/****** Line 32 (optional) ******/
	if (job_ptr->batch_script) {
		if (one_liner)
			xstrcat(out, " ");
		else
			xstrcat(out, "\n   ");
		xstrcat(out, "BatchScript=\n");
		xstrcat(out, job_ptr->batch_script);
	}

	/****** Line 33 (optional) ******/
	if (job_ptr->req_switch) {
		char time_buf[32];
		if (one_liner)
			xstrcat(out, " ");
		else
			xstrcat(out, "\n   ");
		secs2time_str((time_t) job_ptr->wait4switch, time_buf,
			      sizeof(time_buf));
		snprintf(tmp_line, sizeof(tmp_line), "Switches=%u@%s\n",
			 job_ptr->req_switch, time_buf);
		xstrcat(out, tmp_line);
	}

	/****** Line 34 (optional) ******/
	if (one_liner)
		xstrcat(out, "\n");
	else
		xstrcat(out, "\n\n");

	return out;

}
Example #4
0
extern int slurm_jobcomp_log_record ( struct job_record *job_ptr )
{
	int rc = SLURM_SUCCESS;
	char job_rec[1024];
	char usr_str[32], grp_str[32], start_str[32], end_str[32], lim_str[32];
	char *resv_name, *gres, *account, *qos, *wckey, *cluster;
	char submit_time[32], eligible_time[32], array_id[64], pack_id[64];
	char select_buf[128], *state_string, *work_dir;
	size_t offset = 0, tot_size, wrote;
	uint32_t job_state;
	uint32_t time_limit;

	if ((log_name == NULL) || (job_comp_fd < 0)) {
		error("JobCompLoc log file %s not open", log_name);
		return SLURM_ERROR;
	}

	slurm_mutex_lock( &file_lock );
	_get_user_name(job_ptr->user_id, usr_str, sizeof(usr_str));
	_get_group_name(job_ptr->group_id, grp_str, sizeof(grp_str));

	if ((job_ptr->time_limit == NO_VAL) && job_ptr->part_ptr)
		time_limit = job_ptr->part_ptr->max_time;
	else
		time_limit = job_ptr->time_limit;
	if (time_limit == INFINITE)
		strcpy(lim_str, "UNLIMITED");
	else {
		snprintf(lim_str, sizeof(lim_str), "%lu",
			 (unsigned long) time_limit);
	}

	if (job_ptr->job_state & JOB_RESIZING) {
		time_t now = time(NULL);
		state_string = job_state_string(job_ptr->job_state);
		if (job_ptr->resize_time) {
			_make_time_str(&job_ptr->resize_time, start_str,
				       sizeof(start_str));
		} else {
			_make_time_str(&job_ptr->start_time, start_str,
				       sizeof(start_str));
		}
		_make_time_str(&now, end_str, sizeof(end_str));
	} else {
		/* Job state will typically have JOB_COMPLETING or JOB_RESIZING
		 * flag set when called. We remove the flags to get the eventual
		 * completion state: JOB_FAILED, JOB_TIMEOUT, etc. */
		job_state = job_ptr->job_state & JOB_STATE_BASE;
		state_string = job_state_string(job_state);
		if (job_ptr->resize_time) {
			_make_time_str(&job_ptr->resize_time, start_str,
				       sizeof(start_str));
		} else if (job_ptr->start_time > job_ptr->end_time) {
			/* Job cancelled while pending and
			 * expected start time is in the future. */
			snprintf(start_str, sizeof(start_str), "Unknown");
		} else {
			_make_time_str(&job_ptr->start_time, start_str,
				       sizeof(start_str));
		}
		_make_time_str(&job_ptr->end_time, end_str, sizeof(end_str));
	}

	if (job_ptr->details && job_ptr->details->work_dir)
		work_dir = job_ptr->details->work_dir;
	else
		work_dir = "unknown";

	if (job_ptr->resv_name && job_ptr->resv_name[0])
		resv_name = job_ptr->resv_name;
	else
		resv_name = "";

	if (job_ptr->gres_req && job_ptr->gres_req[0])
		gres = job_ptr->gres_req;
	else
		gres = "";

	if (job_ptr->account && job_ptr->account[0])
		account = job_ptr->account;
	else
		account = "";

	if (job_ptr->qos_ptr != NULL) {
		qos = job_ptr->qos_ptr->name;
	} else
		qos = "";

	if (job_ptr->wckey && job_ptr->wckey[0])
		wckey = job_ptr->wckey;
	else
		wckey = "";

	if (job_ptr->assoc_ptr != NULL)
		cluster = job_ptr->assoc_ptr->cluster;
	else
		cluster = "unknown";

	if (job_ptr->details && job_ptr->details->submit_time) {
		_make_time_str(&job_ptr->details->submit_time,
			       submit_time, sizeof(submit_time));
	} else {
		snprintf(submit_time, sizeof(submit_time), "unknown");
	}

	if (job_ptr->details && job_ptr->details->begin_time) {
		_make_time_str(&job_ptr->details->begin_time,
			       eligible_time, sizeof(eligible_time));
	} else {
		snprintf(eligible_time, sizeof(eligible_time), "unknown");
	}

	if (job_ptr->array_task_id != NO_VAL) {
		snprintf(array_id, sizeof(array_id),
			 " ArrayJobId=%u ArrayTaskId=%u",
			 job_ptr->array_job_id, job_ptr->array_task_id);
	} else {
		array_id[0] = '\0';
	}

	if (job_ptr->pack_job_id) {
		snprintf(pack_id, sizeof(pack_id),
			 " PackJobId=%u PackJobOffset=%u",
			 job_ptr->pack_job_id, job_ptr->pack_job_offset);
	} else {
		pack_id[0] = '\0';
	}

	select_g_select_jobinfo_sprint(job_ptr->select_jobinfo,
		select_buf, sizeof(select_buf), SELECT_PRINT_MIXED);

	snprintf(job_rec, sizeof(job_rec), JOB_FORMAT,
		 (unsigned long) job_ptr->job_id, usr_str,
		 (unsigned long) job_ptr->user_id, grp_str,
		 (unsigned long) job_ptr->group_id, job_ptr->name,
		 state_string, job_ptr->partition, lim_str, start_str,
		 end_str, job_ptr->nodes, job_ptr->node_cnt,
		 job_ptr->total_cpus, work_dir, resv_name, gres, account, qos,
		 wckey, cluster, submit_time, eligible_time, array_id, pack_id,
		 job_ptr->derived_ec, job_ptr->exit_code, select_buf);
	tot_size = strlen(job_rec);

	while (offset < tot_size) {
		wrote = write(job_comp_fd, job_rec + offset,
			tot_size - offset);
		if (wrote == -1) {
			if (errno == EAGAIN)
				continue;
			else {
				plugin_errno = errno;
				rc = SLURM_ERROR;
				break;
			}
		}
		offset += wrote;
	}
	slurm_mutex_unlock( &file_lock );
	return rc;
}