support for bluegene systems on sacct

src/sacct/options.c

@@ -128,7 +128,7 @@ void _dump_header(acct_header_t header)
 	       (int)header.timestamp,
 	       header.uid,
 	       header.gid,
-	       "-",	/* reserved 2 */
+	       header.blockid,	/* block id */
 	       "-");	/* reserved 1 */
 }
 /* _open_log_file() -- find the current or specified log file, and open it
@@ -1449,7 +1449,7 @@ void do_fdump(char* f[], int lc)
 			     "timestamp", /* F_TIMESTAMP */
 			     "uid",	 /* F_UIDGID */
 			     "gid",	 /* F_UIDGID */
-			     "reserved-1",/* F_RESERVED1 */
+			     "BlockID",  /* F_BLOCKID */
 			     "reserved-2",/* F_RESERVED1 */
 			     "recordType",/* F_RECTYPE */
 			     NULL};

src/sacct/print.c

@@ -644,6 +644,35 @@ void print_partition(type_t type, void *object)
 	} 
 }
 
+void print_blockid(type_t type, void *object)
+{ 
+	job_rec_t *job = (job_rec_t *)object;
+	step_rec_t *step = (step_rec_t *)object;
+
+	switch(type) {
+	case HEADLINE:
+		printf("%-16s", "BlockID");
+		break;
+	case UNDERSCORE:
+		printf("%-16s", "----------------");
+		break;
+	case JOB:
+		if(strlen(job->header.partition)<17)
+			printf("%-16s", job->header.partition);
+		else
+			printf("%-13.13s...", job->header.partition);
+		
+		break;
+	case JOBSTEP:
+		if(strlen(step->header.partition)<17)
+			printf("%-16s", step->header.blockid);
+		else
+			printf("%-13.13s...", step->header.blockid);
+	
+		break;
+	} 
+}
+
 void print_psize(type_t type, void *object)
 { 
 	job_rec_t *job = (job_rec_t *)object;

src/sacct/process.c

@@ -30,7 +30,8 @@
 
 job_rec_t *_find_job_record(acct_header_t header);
 step_rec_t *_find_step_record(job_rec_t *job, long jobstep);
-job_rec_t *_init_job_rec(acct_header_t header, int lc);
+job_rec_t *_init_job_rec(acct_header_t header);
+step_rec_t *_init_step_rec(acct_header_t header);
 int _parse_line(char *f[], void **data);
 
 job_rec_t *_find_job_record(acct_header_t header)
@@ -80,50 +81,55 @@ step_rec_t *_find_step_record(job_rec_t *job, long stepnum)
 	return step;
 }
 
-job_rec_t *_init_job_rec(acct_header_t header, int lc)
+job_rec_t *_init_job_rec(acct_header_t header)
 {
 	job_rec_t *job = xmalloc(sizeof(job_rec_t));
-
-	job->header.jobnum = header.jobnum;
-	job->header.partition = xstrdup(header.partition);
-	job->header.job_start = header.job_start;
-	job->header.timestamp = header.timestamp;
-	job->header.uid = header.uid;
-	job->header.gid = header.gid;
+	memcpy(&job->header, &header, sizeof(acct_header_t));
+	memset(&job->rusage, 0, sizeof(struct rusage));
 	job->job_start_seen = 0;
 	job->job_step_seen = 0;
 	job->job_terminated_seen = 0;
 	job->jobnum_superseded = 0;
 	job->jobname = xstrdup("(unknown)");
 	job->status = JOB_PENDING;
+	job->nodes = NULL;
+	job->jobname = NULL;
+	job->exitcode = 0;
+	job->priority = 0;
+	job->ntasks = 0;
+	job->ncpus = 0;
+	job->elapsed = 0;
 	job->tot_cpu_sec = 0;
 	job->tot_cpu_usec = 0;
-	job->rusage.ru_utime.tv_sec = 0;
-	job->rusage.ru_utime.tv_usec += 0;
-	job->rusage.ru_stime.tv_sec += 0;
-	job->rusage.ru_stime.tv_usec += 0;
-	job->rusage.ru_inblock += 0;
-	job->rusage.ru_oublock += 0;
-	job->rusage.ru_msgsnd += 0;
-	job->rusage.ru_msgrcv += 0;
-	job->rusage.ru_nsignals += 0;
-	job->rusage.ru_nvcsw += 0;
-	job->rusage.ru_nivcsw += 0;
-	job->rusage.ru_maxrss = 0;
-	job->rusage.ru_ixrss = 0;
-	job->rusage.ru_idrss = 0;
-	job->rusage.ru_isrss = 0;
-	job->rusage.ru_minflt = 0;
-	job->rusage.ru_majflt = 0;
-	job->rusage.ru_nswap = 0;
 	job->vsize = 0;
-	job->psize = 0;
-	job->exitcode = 0;
+	job->psize = 0;	
 	job->steps = list_create(destroy_step);
 	job->nodes = NULL;
+	job->track_steps = 0;
+
       	return job;
 }
 
+step_rec_t *_init_step_rec(acct_header_t header)
+{
+	step_rec_t *step = xmalloc(sizeof(job_rec_t));
+	memcpy(&step->header, &header, sizeof(acct_header_t));
+	memset(&step->rusage, 0, sizeof(struct rusage));
+	step->stepnum = (uint32_t)NO_VAL;
+	step->nodes = NULL;
+	step->stepname = NULL;
+	step->status = NO_VAL;
+	step->exitcode = NO_VAL;
+	step->ntasks = (uint32_t)NO_VAL;
+	step->ncpus = (uint32_t)NO_VAL;
+	step->elapsed = (uint32_t)NO_VAL;
+	step->tot_cpu_sec = (uint32_t)NO_VAL;
+	step->tot_cpu_usec = (uint32_t)NO_VAL;
+	step->vsize = (uint32_t)NO_VAL;
+	step->psize = (uint32_t)NO_VAL;
+
+	return step;
+}
 int _parse_header(char *f[], acct_header_t *header)
 {
 	header->jobnum = atoi(f[F_JOB]);
@@ -132,6 +138,7 @@ int _parse_header(char *f[], acct_header_t *header)
 	header->timestamp = atoi(f[F_TIMESTAMP]);
 	header->uid = atoi(f[F_UID]);
 	header->gid = atoi(f[F_GID]);
+	header->blockid = xstrdup(f[F_BLOCKID]);
 	return SLURM_SUCCESS;
 }
 
@@ -140,11 +147,12 @@ int _parse_line(char *f[], void **data)
 	int i = atoi(f[F_RECTYPE]);
 	job_rec_t **job = (job_rec_t **)data;
 	step_rec_t **step = (step_rec_t **)data;
-	
+	acct_header_t header;
+	_parse_header(f, &header);
+		
 	switch(i) {
 	case JOB_START:
-		*job = xmalloc(sizeof(job_rec_t));
-		_parse_header(f, &(*job)->header);
+		*job = _init_job_rec(header);
 		(*job)->jobname = xstrdup(f[F_JOBNAME]);
 		(*job)->track_steps = atoi(f[F_TRACK_STEPS]);
 		(*job)->priority = atoi(f[F_PRIORITY]);
@@ -159,8 +167,7 @@ int _parse_line(char *f[], void **data)
 		}
 		break;
 	case JOB_STEP:
-		*step = xmalloc(sizeof(step_rec_t));
-		_parse_header(f, &(*step)->header);
+		*step = _init_step_rec(header);
 		(*step)->stepnum = atoi(f[F_JOBSTEP]);
 		(*step)->status = atoi(f[F_STATUS]);
 		(*step)->exitcode = atoi(f[F_EXITCODE]);
@@ -191,11 +198,11 @@ int _parse_line(char *f[], void **data)
 		(*step)->psize = atoi(f[F_PSIZE]);
 		(*step)->stepname = xstrdup(f[F_STEPNAME]);
 		(*step)->nodes = xstrdup(f[F_STEPNODES]);
+				
 		break;
 	case JOB_SUSPEND:
 	case JOB_TERMINATED:
-		*job = xmalloc(sizeof(job_rec_t));
-		_parse_header(f, &(*job)->header);
+		*job = _init_job_rec(header);
 		(*job)->elapsed = atoi(f[F_TOT_ELAPSED]);
 		(*job)->status = atoi(f[F_STATUS]);		
 		break;
@@ -223,18 +230,11 @@ void process_start(char *f[], int lc)
 		return;
 	}
 	
-	job = _init_job_rec(temp->header, lc);
+	job = temp;
+	
 	list_append(jobs, job);
 	job->job_start_seen = 1;
-	job->header.uid = temp->header.uid;
-	job->header.gid = temp->header.gid;
-	xfree(job->jobname);
-	job->jobname = xstrdup(temp->jobname);
-	job->priority = temp->priority;
-	job->track_steps = temp->track_steps;
-	job->ncpus = temp->ncpus;
-	job->nodes = xstrdup(temp->nodes);
-	destroy_job(temp);
+	
 }
 
 void process_step(char *f[], int lc)
@@ -253,7 +253,7 @@ void process_step(char *f[], int lc)
 		return;
 	}
 	if (!job) {	/* fake it for now */
-		job = _init_job_rec(temp->header, lc);
+		job = _init_job_rec(temp->header);
 		if ((params.opt_verbose > 1) 
 		    && (params.opt_jobstep_list==NULL)) 
 			fprintf(stderr, 
@@ -291,6 +291,7 @@ void process_step(char *f[], int lc)
 		step->psize = temp->psize;
 		xfree(step->stepname);
 		step->stepname = xstrdup(temp->stepname);
+		destroy_step(temp);
 		goto got_step;
 	}
 	step = temp;
@@ -305,7 +306,7 @@ void process_step(char *f[], int lc)
 	}
 	
 got_step:
-	destroy_step(temp);
+	
 		
 	if (job->job_terminated_seen == 0) {	/* If the job is still running,
 						   this is the most recent
@@ -359,7 +360,7 @@ void process_suspend(char *f[], int lc)
 	_parse_line(f, (void **)&temp);
 	job = _find_job_record(temp->header);
 	if (!job)    
-		job = _init_job_rec(temp->header, lc);
+		job = _init_job_rec(temp->header);
 	
 	if (job->status == JOB_SUSPENDED) 
 		job->elapsed -= temp->elapsed;
@@ -377,7 +378,7 @@ void process_terminated(char *f[], int lc)
 	_parse_line(f, (void **)&temp);
 	job = _find_job_record(temp->header);
 	if (!job) {	/* fake it for now */
-		job = _init_job_rec(temp->header, lc);
+		job = _init_job_rec(temp->header);
 		if (params.opt_verbose > 1) 
 			fprintf(stderr, "Note: JOB_TERMINATED record for job "
 				"%u preceded "
@@ -418,13 +419,21 @@ finished:
 	destroy_job(temp);
 }
 
+void destroy_acct_header(void *object)
+{
+	acct_header_t *header = (acct_header_t *)object;
+	if(header) {
+		xfree(header->partition);
+		xfree(header->blockid);
+	}
+}
 void destroy_job(void *object)
 {
 	job_rec_t *job = (job_rec_t *)object;
 	if (job) {
 		if(job->steps)
 			list_destroy(job->steps);
-		xfree(job->header.partition);
+		destroy_acct_header(&job->header);
 		xfree(job->jobname);
 		xfree(job->nodes);
 		xfree(job);
@@ -435,7 +444,7 @@ void destroy_step(void *object)
 {
 	step_rec_t *step = (step_rec_t *)object;
 	if (step) {
-		xfree(step->header.partition);
+		destroy_acct_header(&step->header);
 		xfree(step->stepname);
 		xfree(step->nodes);
 		xfree(step);

src/sacct/sacct.c

@@ -165,6 +165,7 @@ fields_t fields[] = {{"cpu", print_cpu},
 		     {"nvcsw", print_nvcsw}, 
 		     {"outblocks", print_outblocks}, 
 		     {"partition", print_partition}, 
+		     {"blockid", print_blockid}, 
 		     {"psize", print_psize}, 
 		     {"rss", print_rss}, 
 		     {"status", print_status}, 

src/sacct/sacct.h

@@ -83,7 +83,7 @@ enum {	F_JOB =	0,
 	F_TIMESTAMP,	
 	F_UID,	
 	F_GID,	
-	F_RESERVED1,	
+	F_BLOCKID,	
 	F_RESERVED2,	
 	F_RECTYPE,	
 	HEADER_LENGTH
@@ -159,6 +159,7 @@ enum {	CANCELLED,
 typedef struct header {
 	uint32_t jobnum;
 	char	*partition;
+	char	*blockid;
 	time_t 	job_start;
 	time_t	timestamp;
 	uint32_t uid;
@@ -181,24 +182,28 @@ typedef struct job_rec {
 	int32_t	status;
 	int32_t	exitcode;
 	uint32_t elapsed;
-	uint32_t tot_cpu_sec, tot_cpu_usec;
-	uint32_t vsize, psize;
+	uint32_t tot_cpu_sec;
+	uint32_t tot_cpu_usec;
+	uint32_t vsize;
+	uint32_t psize;
 	struct rusage rusage;
 	List    steps;
 } job_rec_t;
 
 typedef struct step_rec {
-	acct_header_t header;
+	acct_header_t   header;
 	uint32_t	stepnum;	/* job's step number */
-	uint32_t	next;		/* linked list of job steps */
 	char	        *nodes;
 	char	        *stepname;
 	int32_t	        status;
 	int32_t	        exitcode;
-	uint32_t	ntasks, ncpus;
+	uint32_t	ntasks; 
+	uint32_t        ncpus;
 	uint32_t	elapsed;
-	uint32_t	tot_cpu_sec, tot_cpu_usec;
-	uint32_t	vsize, psize;
+	uint32_t	tot_cpu_sec;
+	uint32_t        tot_cpu_usec;
+	uint32_t	vsize;
+	uint32_t        psize;
 	struct rusage rusage;
 } step_rec_t;
 
@@ -251,6 +256,7 @@ void process_start(char *f[], int lc);
 void process_step(char *f[], int lc);
 void process_suspend(char *f[], int lc);
 void process_terminated(char *f[], int lc);
+void destroy_acct_header(void *object);
 void destroy_job(void *object);
 void destroy_step(void *object);
 
@@ -281,6 +287,7 @@ void print_ntasks(type_t type, void *object);
 void print_nvcsw(type_t type, void *object);
 void print_outblocks(type_t type, void *object);
 void print_partition(type_t type, void *object);
+void print_blockid(type_t type, void *object);
 void print_psize(type_t type, void *object);
 void print_rss(type_t type, void *object);
 void print_status(type_t type, void *object);

src/slurmctld/jobacct.c

@@ -79,24 +79,33 @@ const char *_jobstep_format =
 static int _print_record(struct job_record *job_ptr, 
 			 time_t time, char *data)
 { 
-	struct tm   *ts; /* timestamp decoder */
 	static int   rc=SLURM_SUCCESS;
+	char *block_id = NULL;
 
-	ts = xmalloc(sizeof(struct tm));
-	gmtime_r(&time, ts);
 	debug3("_print_record, job=%u, \"%s\"",
 	      job_ptr->job_id, data);
+#ifdef HAVE_BG
+	select_g_get_jobinfo(job_ptr->select_jobinfo, 
+			     SELECT_DATA_BLOCK_ID, 
+			     &block_id);
+		
+#endif
+	if(!block_id)
+		block_id = xstrdup("-");
+
 	slurm_mutex_lock( &logfile_lock );
+
 	if (fprintf(LOGFILE,
-		    "%u %s %u %u %d %d - - %s\n",
+		    "%u %s %u %u %d %d %s - %s\n",
 		    job_ptr->job_id, job_ptr->partition,
 		    (int)job_ptr->start_time, (int)time, 
-		    job_ptr->user_id, job_ptr->group_id, data)
+		    job_ptr->user_id, job_ptr->group_id, block_id, data)
 	    < 0)
 		rc=SLURM_ERROR;
 	fdatasync(LOGFILE_FD);
 	slurm_mutex_unlock( &logfile_lock );
-	xfree(ts);
+	xfree(block_id);
+
 	return rc;
 }
 
@@ -146,6 +155,7 @@ int jobacct_job_start(struct job_record *job_ptr)
 		debug("jobacct init was not called or it failed");
 		return SLURM_ERROR;
 	}
+
 	debug2("jobacct_job_start() called");
 	for (i=0; i < job_ptr->num_cpu_groups; i++)
 		ncpus += (job_ptr->cpus_per_node[i])
@@ -185,17 +195,39 @@ int jobacct_step_start(struct step_record *step)
 {
 	char buf[BUFFER_SIZE];
 	int cpus = 0;
+	char node_list[BUFFER_SIZE];
+#ifdef HAVE_BG
+	uint16_t quarter = (uint16_t)NO_VAL;
+	uint16_t nodecard = (uint16_t)NO_VAL;
+#endif
 
 	if(!init) {
 		debug("jobacct init was not called or it failed");
 		return SLURM_ERROR;
 	}
+
 #ifdef HAVE_BG
-	cpus = step->job_ptr->num_procs;
+	select_g_get_jobinfo(step->job_ptr->select_jobinfo, 
+			     SELECT_DATA_QUARTER, 
+			     &quarter);
+	select_g_get_jobinfo(step->job_ptr->select_jobinfo, 
+			     SELECT_DATA_NODECARD, 
+			     &nodecard);
+	if(quarter != (uint16_t)NO_VAL 
+	   && nodecard != (uint16_t)NO_VAL)
+		snprintf(node_list, BUFFER_SIZE, 
+			 "%s.%d.%d", step->step_node_list, quarter, nodecard);
+	else if(quarter != (uint16_t)NO_VAL)
+		snprintf(node_list, BUFFER_SIZE, 
+			 "%s.%d", step->step_node_list, quarter);
+	else
+		snprintf(node_list, BUFFER_SIZE, "%s", step->step_node_list);
+	
 #else
 	cpus = step->num_cpus;
+	snprintf(node_list, BUFFER_SIZE, "%s", step->step_node_list);
+	block_id = xstrdup("-");
 #endif
-	
 	snprintf(buf, BUFFER_SIZE, _jobstep_format,
 		 JOB_STEP,
 		 step->step_id,	/* stepid */
@@ -226,8 +258,9 @@ int jobacct_step_start(struct step_record *step)
 		 0,	/* total nivcsw */
 		 0,		/* max vsize */
 		 0,		/* max psize */
-		 step->name,      	/* step exe name */
-		 step->step_node_list); /* name of nodes step running on */
+		 step->name,    /* step exe name */
+		 node_list);     /* name of nodes step running on */
+	
 	return _print_record(step->job_ptr, step->start_time, buf);	
 }
 
@@ -238,6 +271,11 @@ int jobacct_step_complete(struct step_record *step)
 	int elapsed;
 	int comp_status;
 	int cpus = 0;
+	char node_list[BUFFER_SIZE];
+#ifdef HAVE_BG
+	uint16_t quarter = (uint16_t)NO_VAL;
+	uint16_t nodecard = (uint16_t)NO_VAL;
+#endif
 
 	if(!init) {
 		debug("jobacct init was not called or it failed");
@@ -254,9 +292,26 @@ int jobacct_step_complete(struct step_record *step)
 		comp_status = JOB_COMPLETE;
 
 #ifdef HAVE_BG
-	cpus = step->job_ptr->num_procs;
+	select_g_get_jobinfo(step->job_ptr->select_jobinfo, 
+			     SELECT_DATA_QUARTER, 
+			     &quarter);
+	select_g_get_jobinfo(step->job_ptr->select_jobinfo, 
+			     SELECT_DATA_NODECARD, 
+			     &nodecard);
+	if(quarter != (uint16_t)NO_VAL 
+	   && nodecard != (uint16_t)NO_VAL)
+		snprintf(node_list, BUFFER_SIZE, 
+			 "%s.%d.%d", step->step_node_list, quarter, nodecard);
+	else if(quarter != (uint16_t)NO_VAL)
+		snprintf(node_list, BUFFER_SIZE, 
+			 "%s.%d", step->step_node_list, quarter);
+	else
+		snprintf(node_list, BUFFER_SIZE, "%s", step->step_node_list);
+	
 #else
 	cpus = step->num_cpus;
+	snprintf(node_list, BUFFER_SIZE, "%s", step->step_node_list);
+	block_id = xstrdup("-");
 #endif
 	
 	snprintf(buf, BUFFER_SIZE, _jobstep_format,
@@ -294,8 +349,8 @@ int jobacct_step_complete(struct step_record *step)
 		 step->max_vsize,		/* max vsize */
 		 step->max_psize,		/* max psize */
 		 step->name,      	/* step exe name */
-		 step->step_node_list); /* name of nodes step running on */
-       
+		 node_list); /* name of nodes step running on */
+
 	return _print_record(step->job_ptr, now, buf);	
 }
 

src/slurmctld/jobacct.h

@@ -35,6 +35,7 @@
 #include <slurm/slurm_errno.h>
 #include <sys/stat.h>
 #include "src/common/xstring.h"
+#include "src/common/node_select.h"
 #include "slurmctld.h"
 
 int jobacct_init(char *job_acct_log);