static int
status_que(pbs_queue *pque, struct batch_request *preq, pbs_list_head *pstathd)
{
struct brp_status *pstat;
svrattrl *pal;
if ((preq->rq_perm & ATR_DFLAG_RDACC) == 0)
return (PBSE_PERM);
/* ok going to do status, update count and state counts from qu_qs */
if (!svr_chk_history_conf()) {
pque->qu_attr[(int)QA_ATR_TotalJobs].at_val.at_long = pque->qu_numjobs;
} else {
pque->qu_attr[(int)QA_ATR_TotalJobs].at_val.at_long = pque->qu_numjobs -
(pque->qu_njstate[JOB_STATE_MOVED] + pque->qu_njstate[JOB_STATE_FINISHED]);
}
pque->qu_attr[(int)QA_ATR_TotalJobs].at_flags |= ATR_VFLAG_SET|ATR_VFLAG_MODCACHE;
update_state_ct(&pque->qu_attr[(int)QA_ATR_JobsByState],
pque->qu_njstate,
pque->qu_jobstbuf);
/* allocate status sub-structure and fill in header portion */
pstat = (struct brp_status *)malloc(sizeof(struct brp_status));
if (pstat == NULL)
return (PBSE_SYSTEM);
pstat->brp_objtype = MGR_OBJ_QUEUE;
(void)strcpy(pstat->brp_objname, pque->qu_qs.qu_name);
CLEAR_LINK(pstat->brp_stlink);
CLEAR_HEAD(pstat->brp_attr);
append_link(pstathd, &pstat->brp_stlink, pstat);
/* add attributes to the status reply */
bad = 0;
pal = (svrattrl *)GET_NEXT(preq->rq_ind.rq_status.rq_attr);
if (status_attrib(pal, que_attr_def, pque->qu_attr, QA_ATR_LAST,
preq->rq_perm, &pstat->brp_attr, &bad))
return (PBSE_NOATTR);
return (0);
}
/**
* @brief
* update_subjob_state - update the subjob state in the table entry for
* the subjob and the total counts for each state.
* If job going into EXPIRED state, the job exitstatus is saved in the tbl
*
* @param[in] pjob - pointer to the actual subjob job entry
* @param[in] newstate - newstate of the sub job.
*
* @return void
*/
void
update_subjob_state(job *pjob, int newstate)
{
int len;
job *parent;
char *pc;
struct ajtrkhd *ptbl;
parent = pjob->ji_parentaj;
if (parent == NULL)
return;
ptbl = parent->ji_ajtrk;
if (ptbl == NULL)
return;
/* verify that parent job is in fact the parent Array Job */
pc = strchr(pjob->ji_qs.ji_jobid, (int)'[');
len = pc - pjob->ji_qs.ji_jobid - 1;
if ((strncmp(pjob->ji_qs.ji_jobid, parent->ji_qs.ji_jobid, len) != 0) ||
(ptbl == NULL))
return; /* nope, not the parent */
set_subjob_tblstate(parent, pjob->ji_subjindx, newstate);
if (newstate == JOB_STATE_EXPIRED) {
ptbl->tkm_tbl[pjob->ji_subjindx].trk_error =
pjob->ji_qs.ji_un.ji_exect.ji_exitstat;
if (svr_chk_history_conf()) {
if ((pjob->ji_wattr[(int)JOB_ATR_stageout_status].at_flags) & ATR_VFLAG_SET) {
ptbl->tkm_tbl[pjob->ji_subjindx].trk_stgout =
pjob->ji_wattr[(int)JOB_ATR_stageout_status].at_val.at_long;
}
if ((pjob->ji_wattr[(int)JOB_ATR_exit_status].at_flags) & ATR_VFLAG_SET) {
ptbl->tkm_tbl[pjob->ji_subjindx].trk_exitstat = 1;
}
}
ptbl->tkm_tbl[pjob->ji_subjindx].trk_substate = pjob->ji_qs.ji_substate;
parent->ji_modified = 1;
}
chk_array_doneness(parent);
}
/**
* @brief
* post_movejob - clean up action for child started in net_move/send_job
* to "move" a job to another server
* @par
* If move was successfull, delete server's copy of thejob structure,
* and reply to request.
* @par
* If route didn't work, reject the request.
*
* @param[in] pwt - work task structure
*
* @return none.
*/
static void
post_movejob(struct work_task *pwt)
{
char *id = "post_movejob";
struct batch_request *req;
int newstate;
int newsub;
int stat;
int r;
job *jobp;
req = (struct batch_request *)pwt->wt_parm1;
stat = pwt->wt_aux;
pbs_errno = PBSE_NONE;
if (req->rq_type != PBS_BATCH_MoveJob) {
sprintf(log_buffer, "bad request type %d", req->rq_type);
log_err(-1, __func__, log_buffer);
return;
}
jobp = find_job(req->rq_ind.rq_move.rq_jid);
if ((jobp == NULL) || (jobp != (job *)pwt->wt_parm2)) {
sprintf(log_buffer,
"job %s not found",
req->rq_ind.rq_move.rq_jid);
log_err(-1, __func__, log_buffer);
}
if (WIFEXITED(stat)) {
r = WEXITSTATUS(stat);
if (r == SEND_JOB_OK) { /* purge server's job structure */
if (jobp->ji_qs.ji_svrflags & JOB_SVFLG_StagedIn)
remove_stagein(jobp);
(void)strcpy(log_buffer, msg_movejob);
(void)sprintf(log_buffer+strlen(log_buffer),
msg_manager,
req->rq_ind.rq_move.rq_destin,
req->rq_user, req->rq_host);
/*
* If server is configured to keep job history info and
* the job is created here, then keep the job struture
* for history purpose without purging. No need to check
* for sub-jobs as sub jobs can't be moved.
*/
if (svr_chk_history_conf())
svr_setjob_histinfo(jobp, T_MOV_JOB);
else
job_purge(jobp);
} else
r = PBSE_ROUTEREJ;
} else {
r = PBSE_SYSTEM;
(void)sprintf(log_buffer, msg_badexit, stat);
(void)strcat(log_buffer, __func__);
log_event(PBSEVENT_SYSTEM, PBS_EVENTCLASS_JOB, LOG_NOTICE,
jobp->ji_qs.ji_jobid, log_buffer);
}
if (r) {
if (jobp) {
/* force re-eval of job state out of Transit */
svr_evaljobstate(jobp, &newstate, &newsub, 1);
svr_setjobstate(jobp, newstate, newsub);
}
req_reject(r, 0, req);
} else
reply_ack(req);
return;
}
/**
* @brief
* post_routejob - clean up action for child started in net_move/send_job
* to "route" a job to another server
* @par
* If route was successfull, delete job.
* @par
* If route didn't work, mark destination not to be tried again for this
* job and call route again.
*
* @param[in] pwt - work task structure
*
* @return none.
*/
static void
post_routejob(struct work_task *pwt)
{
int newstate;
int newsub;
int r;
int stat = pwt->wt_aux;
job *jobp = (job *)pwt->wt_parm2;
if (jobp == NULL) {
log_event(PBSEVENT_ERROR, PBS_EVENTCLASS_JOB, LOG_INFO, "", "post_routejob failed, jobp NULL");
return;
}
if (WIFEXITED(stat)) {
r = WEXITSTATUS(stat);
} else {
r = SEND_JOB_FATAL;
(void)sprintf(log_buffer, msg_badexit, stat);
(void)strcat(log_buffer, __func__);
log_event(PBSEVENT_SYSTEM, PBS_EVENTCLASS_JOB, LOG_NOTICE,
jobp->ji_qs.ji_jobid, log_buffer);
}
switch (r) {
case SEND_JOB_OK: /* normal return, job was routed */
if (jobp->ji_qs.ji_svrflags & JOB_SVFLG_StagedIn)
remove_stagein(jobp);
/*
* If the server is configured to keep job history and the job
* is created here, do not purge the job structure but save
* it for history purpose. No need to check for sub-jobs as
* sub jobs can not be routed.
*/
if (svr_chk_history_conf())
svr_setjob_histinfo(jobp, T_MOV_JOB);
else
job_purge(jobp); /* need to remove server job struct */
return;
case SEND_JOB_FATAL: /* permanent rejection (or signal) */
if (jobp->ji_qs.ji_substate == JOB_SUBSTATE_ABORT) {
/* Job Delete in progress, just set to queued status */
(void)svr_setjobstate(jobp, JOB_STATE_QUEUED,
JOB_SUBSTATE_ABORT);
return;
}
add_dest(jobp); /* else mark destination as bad */
/* fall through */
default : /* try routing again */
/* force re-eval of job state out of Transit */
svr_evaljobstate(jobp, &newstate, &newsub, 1);
(void)svr_setjobstate(jobp, newstate, newsub);
jobp->ji_retryok = 1;
if ((r = job_route(jobp)) == PBSE_ROUTEREJ)
(void)job_abt(jobp, msg_routebad);
else if (r != 0)
(void)job_abt(jobp, msg_routexceed);
break;
}
return;
}
