static void _process_start(List job_list, char *f[], int lc,
int show_full, int len)
{
filetxt_job_rec_t *job = NULL;
filetxt_job_rec_t *temp = NULL;
_parse_line(f, (void **)&temp, len);
job = _find_job_record(job_list, temp->header, JOB_START);
if (job) {
/* in slurm we can get 2 start records one for submit
* and one for start, so look at the last one */
xfree(job->jobname);
job->jobname = xstrdup(temp->jobname);
job->track_steps = temp->track_steps;
job->priority = temp->priority;
job->ncpus = temp->ncpus;
xfree(job->nodes);
job->nodes = xstrdup(temp->nodes);
xfree(job->account);
job->account = xstrdup(temp->account);
_destroy_filetxt_job_rec(temp);
return;
}
job = temp;
job->show_full = show_full;
list_append(job_list, job);
job->job_start_seen = 1;
}
static void _process_terminated(List job_list, char *f[], int lc,
int show_full, int len)
{
filetxt_job_rec_t *job = NULL;
filetxt_job_rec_t *temp = NULL;
_parse_line(f, (void **)&temp, len);
if (temp == NULL) {
error("Unknown proccess terminated");
return;
}
job = _find_job_record(job_list, temp->header, JOB_TERMINATED);
if (!job) { /* fake it for now */
job = _create_filetxt_job_rec(temp->header);
job->jobname = xstrdup("(unknown)");
debug("Note: JOB_TERMINATED record for job "
"%u preceded "
"other job records at line %d\n",
temp->header.jobnum, lc);
} else if (job->job_terminated_seen) {
if (temp->status == JOB_NODE_FAIL) {
/* multiple node failures - extra TERMINATED records */
debug("Note: Duplicate JOB_TERMINATED "
"record (nf) for job %u at "
"line %d\n",
temp->header.jobnum, lc);
/* JOB_TERMINATED/NF records may be preceded
* by a JOB_TERMINATED/CA record; NF is much
* more interesting.
*/
job->status = temp->status;
goto finished;
}
fprintf(stderr,
"Conflicting JOB_TERMINATED record (%s) for "
"job %u at line %d -- ignoring it\n",
job_state_string(temp->status),
job->header.jobnum, lc);
goto finished;
}
job->job_terminated_seen = 1;
job->elapsed = temp->elapsed;
job->end = temp->header.timestamp;
job->status = temp->status;
job->requid = temp->requid;
job->exitcode = temp->exitcode;
if (list_count(job->steps) > 1)
job->track_steps = 1;
job->show_full = show_full;
finished:
_destroy_filetxt_job_rec(temp);
}
static void _process_suspend(List job_list, char *f[], int lc,
int show_full, int len)
{
filetxt_job_rec_t *job = NULL;
filetxt_job_rec_t *temp = NULL;
_parse_line(f, (void **)&temp, len);
job = _find_job_record(job_list, temp->header, JOB_SUSPEND);
if (!job) { /* fake it for now */
job = _create_filetxt_job_rec(temp->header);
job->jobname = xstrdup("(unknown)");
}
job->show_full = show_full;
if (job->status == JOB_SUSPENDED)
job->elapsed -= temp->elapsed;
//job->header.timestamp = temp->header.timestamp;
job->status = temp->status;
_destroy_filetxt_job_rec(temp);
}
static void _process_step(List job_list, char *f[], int lc,
int show_full, int len)
{
filetxt_job_rec_t *job = NULL;
filetxt_step_rec_t *step = NULL;
filetxt_step_rec_t *temp = NULL;
_parse_line(f, (void **)&temp, len);
job = _find_job_record(job_list, temp->header, JOB_STEP);
if (temp->stepnum == -2) {
_destroy_filetxt_step_rec(temp);
return;
}
if (!job) { /* fake it for now */
job = _create_filetxt_job_rec(temp->header);
job->jobname = xstrdup("(unknown)");
debug2("Note: JOB_STEP record %u.%u preceded "
"JOB_START record at line %d\n",
temp->header.jobnum, temp->stepnum, lc);
}
job->show_full = show_full;
if ((step = _find_step_record(job, temp->stepnum))) {
if (temp->status == JOB_RUNNING) {
_destroy_filetxt_step_rec(temp);
return;/* if "R" record preceded by F or CD; unusual */
}
if (step->status != JOB_RUNNING) { /* if not JOB_RUNNING */
fprintf(stderr,
"Conflicting JOB_STEP record for "
"jobstep %u.%u at line %d "
"-- ignoring it\n",
step->header.jobnum,
step->stepnum, lc);
_destroy_filetxt_step_rec(temp);
return;
}
step->status = temp->status;
step->exitcode = temp->exitcode;
step->ntasks = temp->ntasks;
step->ncpus = temp->ncpus;
step->elapsed = temp->elapsed;
step->tot_cpu_sec = temp->tot_cpu_sec;
step->tot_cpu_usec = temp->tot_cpu_usec;
job->requid = temp->requid;
step->requid = temp->requid;
memcpy(&step->rusage, &temp->rusage, sizeof(struct rusage));
memcpy(&step->stats, &temp->stats, sizeof(slurmdb_stats_t));
xfree(step->stepname);
step->stepname = xstrdup(temp->stepname);
step->end = temp->header.timestamp;
_destroy_filetxt_step_rec(temp);
goto got_step;
}
step = temp;
temp = NULL;
list_append(job->steps, step);
if (!job->track_steps) {
/* If we don't have track_steps we want to see
if we have multiple steps. If we only have
1 step check the job name against the step
name in most all cases it will be
different. If it is different print out
the step separate.
*/
if (list_count(job->steps) > 1)
job->track_steps = 1;
else if (step && step->stepname && job->jobname) {
if (strcmp(step->stepname, job->jobname))
job->track_steps = 1;
}
}
if (job->header.timestamp == 0)
job->header.timestamp = step->header.timestamp;
job->job_step_seen = 1;
job->ntasks += step->ntasks;
if (!job->nodes || !strcmp(job->nodes, "(unknown)")) {
xfree(job->nodes);
job->nodes = xstrdup(step->nodes);
}
got_step:
if (job->job_terminated_seen == 0) { /* If the job is still running,
this is the most recent
status */
if ( job->exitcode == 0 )
job->exitcode = step->exitcode;
job->status = JOB_RUNNING;
job->elapsed = step->header.timestamp - job->header.timestamp;
}
}
