/**
* @brief
* chk_array_doneness - check if all subjobs are expired and if so,
* purge the Array Job itself
*
* @param[in,out] parent - pointer to parent job.
*
* @return void
*/
void
chk_array_doneness(job *parent)
{
char acctbuf[40];
int e;
int i;
struct ajtrkhd *ptbl = parent->ji_ajtrk;
if (ptbl == NULL)
return;
if (ptbl->tkm_flags & TKMFLG_NO_DELETE)
return; /* delete of subjobs in progress, don't array */
if (ptbl->tkm_subjsct[JOB_STATE_QUEUED] + ptbl->tkm_subjsct[JOB_STATE_RUNNING] + ptbl->tkm_subjsct[JOB_STATE_EXITING] == 0) {
/* Array Job all done, do simple eoj processing */
for (e=i=0; i<ptbl->tkm_ct; ++i) {
if (ptbl->tkm_tbl[i].trk_error > 0)
e = 1;
else if (ptbl->tkm_tbl[i].trk_error < 0) {
e = 2;
break;
}
}
parent->ji_qs.ji_un_type = JOB_UNION_TYPE_EXEC;
parent->ji_qs.ji_un.ji_exect.ji_momaddr = 0;
parent->ji_qs.ji_un.ji_exect.ji_momport = 0;
parent->ji_qs.ji_un.ji_exect.ji_exitstat = e;
check_block(parent, "");
if (parent->ji_qs.ji_state == JOB_STATE_BEGUN) {
/* if BEGUN, issue 'E' account record */
sprintf(acctbuf, msg_job_end_stat, e);
account_job_update(parent, PBS_ACCT_LAST);
account_jobend(parent, acctbuf, PBS_ACCT_END);
svr_mailowner(parent, MAIL_END, MAIL_NORMAL, acctbuf);
}
if (parent->ji_wattr[(int)JOB_ATR_depend].at_flags & ATR_VFLAG_SET)
(void)depend_on_term(parent);
/*
* Check if the history of the finished job can be saved or it needs to be purged .
*/
svr_saveorpurge_finjobhist(parent);
} else {
(void)job_save(parent, SAVEJOB_FULL);
}
}
/**
* @brief
* force_reque - requeue (rerun) a job
*
* @param[in,out] pwt - job which needs to be rerun
*/
void
force_reque(job *pjob)
{
int newstate;
int newsubstate;
pjob->ji_modified = 1;
pjob->ji_momhandle = -1;
pjob->ji_mom_prot = PROT_INVALID;
/* simulate rerun: free nodes, clear checkpoint flag, and */
/* clear exec_vnode string */
rel_resc(pjob);
/* note in accounting file */
account_jobend(pjob, pjob->ji_acctrec, PBS_ACCT_RERUN);
/* if a subjob, we set substate to RERUN3 to cause trktbl entry */
/* to be reset to Qeued, and then blow away the job struct */
if (pjob->ji_qs.ji_svrflags & JOB_SVFLG_SubJob) {
pjob->ji_qs.ji_substate = JOB_SUBSTATE_RERUN3;
job_purge(pjob);
return;
}
/*
* Clear any JOB_SVFLG_Actsuspd flag too, as the job is no longer
* suspended (User busy). A suspended job is rerun in case of a
* MOM failure after the workstation becomes active(busy).
*/
pjob->ji_qs.ji_svrflags &= ~(JOB_SVFLG_Actsuspd | JOB_SVFLG_StagedIn | JOB_SVFLG_CHKPT);
job_attr_def[(int)JOB_ATR_exec_host].at_free(
&pjob->ji_wattr[(int)JOB_ATR_exec_host]);
job_attr_def[(int)JOB_ATR_exec_host2].at_free(
&pjob->ji_wattr[(int)JOB_ATR_exec_host2]);
job_attr_def[(int)JOB_ATR_exec_vnode].at_free(
&pjob->ji_wattr[(int)JOB_ATR_exec_vnode]);
job_attr_def[(int)JOB_ATR_pset].at_free(
&pjob->ji_wattr[(int)JOB_ATR_pset]);
/* job dir has no meaning for re-queued jobs, so unset it */
job_attr_def[(int)JOB_ATR_jobdir].at_free(&pjob->
ji_wattr[(int)JOB_ATR_jobdir]);
svr_evaljobstate(pjob, &newstate, &newsubstate, 1);
(void)svr_setjobstate(pjob, newstate, newsubstate);
}
