void scan_for_terminated(void)
{
static char id[] = "scan_for_terminated";
int exiteval = 0;
pid_t pid;
job *pjob;
task *ptask = NULL;
int statloc;
unsigned int momport = 0;
int tcount;
if (LOGLEVEL >= 7)
{
log_record(
PBSEVENT_JOB,
PBS_EVENTCLASS_JOB,
id,
"entered");
}
/* update the latest intelligence about the running jobs; */
/* must be done before we reap the zombies, else we lose the info */
termin_child = 0;
if (mom_get_sample() == PBSE_NONE)
{
pjob = (job *)GET_PRIOR(svr_alljobs);
while (pjob != NULL)
{
mom_set_use(pjob);
pjob = (job *)GET_PRIOR(pjob->ji_alljobs);
}
}
/* Now figure out which task(s) have terminated (are zombies) */
/* NOTE: does a job's tasks include its epilog? */
while ((pid = waitpid(-1, &statloc, WNOHANG)) > 0)
{
pjob = (job *)GET_PRIOR(svr_alljobs);
while (pjob != NULL)
{
/*
* see if process was a child doing a special
* function for MOM
*/
if (LOGLEVEL >= 7)
{
snprintf(log_buffer, 1024, "checking job w/subtask pid=%d (child pid=%d)",
pjob->ji_momsubt,
pid);
LOG_EVENT(
PBSEVENT_DEBUG,
PBS_EVENTCLASS_JOB,
pjob->ji_qs.ji_jobid,
log_buffer);
}
if (pid == pjob->ji_momsubt)
{
if (LOGLEVEL >= 7)
{
snprintf(log_buffer, 1024, "found match with job subtask for pid=%d",
pid);
LOG_EVENT(
PBSEVENT_DEBUG,
PBS_EVENTCLASS_JOB,
pjob->ji_qs.ji_jobid,
log_buffer);
}
break;
}
/* look for task */
ptask = (task *)GET_NEXT(pjob->ji_tasks);
/* locate task with associated process id */
tcount = 0;
while (ptask != NULL)
{
if (ptask->ti_qs.ti_sid == pid)
{
if (LOGLEVEL >= 7)
{
snprintf(log_buffer, 1024, "found match with job task %d for pid=%d",
tcount,
pid);
LOG_EVENT(
PBSEVENT_DEBUG,
PBS_EVENTCLASS_JOB,
pjob->ji_qs.ji_jobid,
log_buffer);
}
break;
}
ptask = (task *)GET_NEXT(ptask->ti_jobtask);
tcount++;
} /* END while (ptask) */
if (ptask != NULL)
{
/* pid match located - break out of job loop */
break;
}
pjob = (job *)GET_PRIOR(pjob->ji_alljobs);
} /* END while (pjob != NULL) */
if (pjob == NULL)
{
if (LOGLEVEL >= 1)
{
sprintf(log_buffer, "pid %d not tracked, exitcode=%d",
pid,
statloc);
log_record(
PBSEVENT_JOB,
PBS_EVENTCLASS_JOB,
id,
log_buffer);
}
continue;
} /* END if (pjob == NULL) */
if (WIFEXITED(statloc))
exiteval = WEXITSTATUS(statloc);
else if (WIFSIGNALED(statloc))
exiteval = WTERMSIG(statloc) + 0x100;
else
exiteval = 1;
if (pid == pjob->ji_momsubt)
{
/* PID matches job mom subtask */
/* NOTE: both ji_momsubt and ji_mompost normally set in routine
preobit_reply() after epilog child is successfully forked */
if (pjob->ji_mompost != NULL)
{
if (pjob->ji_mompost(pjob, exiteval) == 0)
{
/* success */
pjob->ji_mompost = NULL;
}
} /* END if (pjob->ji_mompost != NULL) */
else
{
log_record(
PBSEVENT_JOB,
PBS_EVENTCLASS_JOB,
pjob->ji_qs.ji_jobid,
"job has no postprocessing routine registered");
}
/* clear mom sub-task */
pjob->ji_momsubt = 0;
if(multi_mom)
{
momport = pbs_rm_port;
}
job_save(pjob, SAVEJOB_QUICK, momport);
continue;
} /* END if (pid == pjob->ji_momsubt) */
/* what happens if mom PID is reaped before subtask? */
if (LOGLEVEL >= 2)
{
sprintf(log_buffer, "pid %d harvested for job %s, task %d, exitcode=%d",
pid,
pjob->ji_qs.ji_jobid,
ptask->ti_qs.ti_task,
exiteval);
log_record(
PBSEVENT_JOB,
PBS_EVENTCLASS_JOB,
id,
log_buffer);
}
/* where is job purged? How do we keep job from progressing in state until the obit is sent? */
kill_task(ptask, SIGKILL, 0);
ptask->ti_qs.ti_exitstat = exiteval;
ptask->ti_qs.ti_status = TI_STATE_EXITED;
task_save(ptask);
sprintf(log_buffer, "%s: job %s task %d terminated, sid=%d",
id,
pjob->ji_qs.ji_jobid,
ptask->ti_qs.ti_task,
ptask->ti_qs.ti_sid);
LOG_EVENT(
PBSEVENT_DEBUG,
PBS_EVENTCLASS_JOB,
pjob->ji_qs.ji_jobid,
log_buffer);
exiting_tasks = 1;
} /* END while ((pid = waitpid(-1,&statloc,WNOHANG)) > 0) */
return;
} /* END scan_for_terminated() */
static int parse_array_request(char *request, tlist_head *tl)
{
char *temp_str;
int num_tokens;
char **tokens;
int i;
int j;
int num_elements;
int start;
int end;
int num_bad_tokens;
int searching;
array_request_node *rn;
array_request_node *rn2;
temp_str = strdup(request);
num_tokens = array_request_token_count(request);
num_bad_tokens = 0;
tokens = (char**)malloc(sizeof(char*) * num_tokens);
j = num_tokens - 1;
/* start from back and scan backwards setting pointers to tokens and changing ',' to '\0' */
for (i = strlen(temp_str) - 1; i >= 0; i--)
{
if (temp_str[i] == ',')
{
tokens[j--] = &temp_str[i+1];
temp_str[i] = '\0';
}
else if (i == 0)
{
tokens[0] = temp_str;
}
}
for (i = 0; i < num_tokens; i++)
{
num_elements = array_request_parse_token(tokens[i], &start, &end);
if (num_elements == 0)
{
num_bad_tokens++;
}
else
{
rn = (array_request_node*)malloc(sizeof(array_request_node));
rn->start = start;
rn->end = end;
CLEAR_LINK(rn->request_tokens_link);
rn2 = GET_NEXT(*tl);
searching = TRUE;
while (searching)
{
if (rn2 == NULL)
{
append_link(tl, &rn->request_tokens_link, (void*)rn);
searching = FALSE;
}
else if (rn->start < rn2->start)
{
insert_link(&rn2->request_tokens_link, &rn->request_tokens_link, (void*)rn,
LINK_INSET_BEFORE);
searching = FALSE;
}
else
{
rn2 = GET_NEXT(rn2->request_tokens_link);
}
}
rn2 = GET_PRIOR(rn->request_tokens_link);
if (rn2 != NULL && rn2->end >= rn->start)
{
num_bad_tokens++;
}
rn2 = GET_NEXT(rn->request_tokens_link);
if (rn2 != NULL && rn2->start <= rn->end)
{
num_bad_tokens++;
}
}
}
free(tokens);
free(temp_str);
return num_bad_tokens;
}
