void *req_modifyjob(
batch_request *preq) /* I */
{
job *pjob;
svrattrl *plist;
char log_buf[LOCAL_LOG_BUF_SIZE];
pjob = chk_job_request(preq->rq_ind.rq_modify.rq_objname, preq);
if (pjob == NULL)
{
return(NULL);
}
mutex_mgr job_mutex(pjob->ji_mutex, true);
plist = (svrattrl *)GET_NEXT(preq->rq_ind.rq_modify.rq_attr);
if (plist == NULL)
{
/* nothing to do */
reply_ack(preq);
/* SUCCESS */
return(NULL);
}
job_mutex.unlock();
/* If async modify, reply now; otherwise reply is handled later */
if (preq->rq_type == PBS_BATCH_AsyModifyJob)
{
/* reply_ack will free preq. We need to copy it before we call reply_ack */
batch_request *new_preq;
new_preq = duplicate_request(preq, -1);
if (new_preq == NULL)
{
sprintf(log_buf, "failed to duplicate batch request");
log_event(PBSEVENT_JOB, PBS_EVENTCLASS_JOB, __func__, log_buf);
return(NULL);
}
get_batch_request_id(new_preq);
reply_ack(preq);
new_preq->rq_noreply = TRUE; /* set for no more replies */
enqueue_threadpool_request((void *(*)(void *))modify_job_work, new_preq);
}
else
modify_job_work(preq);
return(NULL);
} /* END req_modifyjob() */
int que_to_local_svr(struct batch_request *preq) /* I */
{
preq->rq_fromsvr = 1;
preq->rq_perm = ATR_DFLAG_MGRD | ATR_DFLAG_MGWR | ATR_DFLAG_SvWR;
if (preq->rq_id == NULL)
get_batch_request_id(preq);
set_task(WORK_Immed, 0, reissue_to_svr, preq->rq_id, TRUE);
return(PBSE_NONE);
} /* END que_to_local_svr() */
void queue_a_retry_task(
batch_request *preq, /* I */
void (*replyfunc)(struct work_task *)) /* I */
{
/* create a new batch_request because preq is going to be freed when issue_to_svr returns success */
batch_request *new_preq = duplicate_request(preq, -1);
struct work_task *pwt;
get_batch_request_id(new_preq);
pwt = set_task(WORK_Timed, (time(NULL) + PBS_NET_RETRY_TIME), reissue_to_svr, new_preq->rq_id, TRUE);
pwt->wt_parmfunc = replyfunc;
pthread_mutex_unlock(pwt->wt_mutex);
} /* END queue_a_retry_task() */
int execute_job_delete(
job *pjob, /* M */
char *Msg, /* I */
struct batch_request *preq) /* I */
{
struct work_task *pwtnew;
int rc;
char *sigt = "SIGTERM";
int has_mutex = TRUE;
char log_buf[LOCAL_LOG_BUF_SIZE];
time_t time_now = time(NULL);
long force_cancel = FALSE;
long array_compatible = FALSE;
chk_job_req_permissions(&pjob,preq);
if (pjob == NULL)
{
/* preq is rejected in chk_job_req_permissions here */
return(-1);
}
if (pjob->ji_qs.ji_state == JOB_STATE_TRANSIT)
{
/* see note in req_delete - not sure this is possible still,
* but the deleted code is irrelevant now. I will leave this
* part --dbeer */
unlock_ji_mutex(pjob, __func__, "1", LOGLEVEL);
return(-1);
}
if (pjob->ji_qs.ji_substate == JOB_SUBSTATE_PRERUN ||
pjob->ji_qs.ji_substate == JOB_SUBSTATE_RERUN ||
pjob->ji_qs.ji_substate == JOB_SUBSTATE_RERUN1 ||
pjob->ji_qs.ji_substate == JOB_SUBSTATE_RERUN2 ||
pjob->ji_qs.ji_substate == JOB_SUBSTATE_RERUN3 )
{
/* If JOB_SUBSTATE_PRERUN being sent to MOM, wait till she gets it going */
/* retry in one second */
/* If JOB_SUBSTATE_RERUN, RERUN1, RERUN2 or RERUN3 the
job is being requeued. Wait until finished */
static time_t cycle_check_when = 0;
static char cycle_check_jid[PBS_MAXSVRJOBID + 1];
if (cycle_check_when != 0)
{
if (!strcmp(pjob->ji_qs.ji_jobid, cycle_check_jid) &&
(time_now - cycle_check_when > 10))
{
/* state not updated after 10 seconds */
/* did the mom ever get it? delete it anyways... */
cycle_check_jid[0] = '\0';
cycle_check_when = 0;
goto jump;
}
if (time_now - cycle_check_when > 20)
{
/* give up after 20 seconds */
cycle_check_jid[0] = '\0';
cycle_check_when = 0;
}
} /* END if (cycle_check_when != 0) */
if (cycle_check_when == 0)
{
/* new PRERUN job located */
cycle_check_when = time_now;
strcpy(cycle_check_jid, pjob->ji_qs.ji_jobid);
}
sprintf(log_buf, "job cannot be deleted, state=PRERUN, requeuing delete request");
log_event(PBSEVENT_JOB,PBS_EVENTCLASS_JOB,pjob->ji_qs.ji_jobid,log_buf);
pwtnew = set_task(WORK_Timed,time_now + 1,post_delete_route,preq,FALSE);
unlock_ji_mutex(pjob, __func__, "2", LOGLEVEL);
if (pwtnew == NULL)
{
req_reject(PBSE_SYSTEM, 0, preq, NULL, NULL);
return(-1);
}
else
{
return(ROUTE_DELETE);
}
} /* END if (pjob->ji_qs.ji_substate == JOB_SUBSTATE_PRERUN) */
jump:
/*
* Log delete and if requesting client is not job owner, send mail.
*/
sprintf(log_buf, "requestor=%s@%s", preq->rq_user, preq->rq_host);
/* NOTE: should annotate accounting record with extend message (NYI) */
account_record(PBS_ACCT_DEL, pjob, log_buf);
sprintf(log_buf, msg_manager, msg_deletejob, preq->rq_user, preq->rq_host);
log_event(PBSEVENT_JOB,PBS_EVENTCLASS_JOB,pjob->ji_qs.ji_jobid,log_buf);
/* NOTE: should incorporate job delete message */
if (Msg != NULL)
{
/* have text message in request extension, add it */
strcat(log_buf, "\n");
strcat(log_buf, Msg);
}
if ((svr_chk_owner(preq, pjob) != 0) &&
(pjob->ji_has_delete_nanny == FALSE))
{
/* only send email if owner did not delete job and job deleted
has not been previously attempted */
svr_mailowner(pjob, MAIL_DEL, MAIL_FORCE, log_buf);
/*
* If we sent mail and already sent the extra message
* then reset message so we don't trigger a redundant email
* in job_abt()
*/
if (Msg != NULL)
{
Msg = NULL;
}
}
if ((pjob->ji_qs.ji_svrflags & JOB_SVFLG_CHECKPOINT_FILE) != 0)
{
/* job has restart file at mom, change restart comment if failed */
change_restart_comment_if_needed(pjob);
}
if (pjob->ji_qs.ji_state == JOB_STATE_RUNNING)
{
/*
* setup a nanny task to make sure the job is actually deleted (see the
* comments at job_delete_nanny()).
*/
if (pjob->ji_has_delete_nanny == TRUE)
{
unlock_ji_mutex(pjob, __func__, "3", LOGLEVEL);
req_reject(PBSE_IVALREQ, 0, preq, NULL, "job cancel in progress");
return(-1);
}
apply_job_delete_nanny(pjob, time_now + 60);
/*
* Send signal request to MOM. The server will automagically
* pick up and "finish" off the client request when MOM replies.
*/
get_batch_request_id(preq);
if ((rc = issue_signal(&pjob, sigt, post_delete_mom1, strdup(preq->rq_id))))
{
/* cant send to MOM */
req_reject(rc, 0, preq, NULL, NULL);
}
/* normally will ack reply when mom responds */
if (pjob != NULL)
{
sprintf(log_buf, msg_delrunjobsig, sigt);
log_event(PBSEVENT_JOB,PBS_EVENTCLASS_JOB,pjob->ji_qs.ji_jobid,log_buf);
unlock_ji_mutex(pjob, __func__, "4", LOGLEVEL);
}
return(-1);
} /* END if (pjob->ji_qs.ji_state == JOB_STATE_RUNNING) */
/* make a cleanup task if set */
get_svr_attr_l(SRV_ATR_JobForceCancelTime, &force_cancel);
if (force_cancel > 0)
{
char *dup_jobid = strdup(pjob->ji_qs.ji_jobid);
set_task(WORK_Timed, time_now + force_cancel, ensure_deleted, dup_jobid, FALSE);
}
/* if configured, and this job didn't have a slot limit hold, free a job
* held with the slot limit hold */
get_svr_attr_l(SRV_ATR_MoabArrayCompatible, &array_compatible);
if ((array_compatible != FALSE) &&
((pjob->ji_wattr[JOB_ATR_hold].at_val.at_long & HOLD_l) == FALSE))
{
if ((pjob->ji_arraystruct != NULL) &&
(pjob->ji_is_array_template == FALSE))
{
int i;
int newstate;
int newsub;
job *tmp;
job_array *pa = get_jobs_array(&pjob);
if (pjob == NULL)
return(-1);
for (i = 0; i < pa->ai_qs.array_size; i++)
{
if (pa->job_ids[i] == NULL)
continue;
if (!strcmp(pa->job_ids[i], pjob->ji_qs.ji_jobid))
continue;
if ((tmp = svr_find_job(pa->job_ids[i], FALSE)) == NULL)
{
free(pa->job_ids[i]);
pa->job_ids[i] = NULL;
}
else
{
if (tmp->ji_wattr[JOB_ATR_hold].at_val.at_long & HOLD_l)
{
tmp->ji_wattr[JOB_ATR_hold].at_val.at_long &= ~HOLD_l;
if (tmp->ji_wattr[JOB_ATR_hold].at_val.at_long == 0)
{
tmp->ji_wattr[JOB_ATR_hold].at_flags &= ~ATR_VFLAG_SET;
}
svr_evaljobstate(tmp, &newstate, &newsub, 1);
svr_setjobstate(tmp, newstate, newsub, FALSE);
job_save(tmp, SAVEJOB_FULL, 0);
unlock_ji_mutex(tmp, __func__, "5", LOGLEVEL);
break;
}
unlock_ji_mutex(tmp, __func__, "6", LOGLEVEL);
}
}
if (LOGLEVEL >= 7)
{
sprintf(log_buf, "%s: unlocking ai_mutex", __func__);
log_event(PBSEVENT_JOB, PBS_EVENTCLASS_JOB, pjob->ji_qs.ji_jobid, log_buf);
}
pthread_mutex_unlock(pa->ai_mutex);
}
} /* END MoabArrayCompatible check */
if ((pjob->ji_qs.ji_svrflags & JOB_SVFLG_CHECKPOINT_FILE) != 0)
{
/* job has restart file at mom, do end job processing */
svr_setjobstate(pjob, JOB_STATE_EXITING, JOB_SUBSTATE_EXITING, FALSE);
/* force new connection */
pjob->ji_momhandle = -1;
if (LOGLEVEL >= 7)
{
sprintf(log_buf, "calling on_job_exit from %s", __func__);
log_event(PBSEVENT_JOB, PBS_EVENTCLASS_JOB, pjob->ji_qs.ji_jobid, log_buf);
}
set_task(WORK_Immed, 0, on_job_exit, strdup(pjob->ji_qs.ji_jobid), FALSE);
}
else if ((pjob->ji_qs.ji_svrflags & JOB_SVFLG_StagedIn) != 0)
{
/* job has staged-in file, should remove them */
remove_stagein(&pjob);
if (pjob != NULL)
job_abt(&pjob, Msg);
has_mutex = FALSE;
}
else
{
/*
* the job is not transitting (though it may have been) and
* is not running, so put in into a complete state.
*/
struct pbs_queue *pque;
int KeepSeconds = 0;
svr_setjobstate(pjob, JOB_STATE_COMPLETE, JOB_SUBSTATE_COMPLETE, FALSE);
if ((pque = get_jobs_queue(&pjob)) != NULL)
{
pque->qu_numcompleted++;
unlock_queue(pque, __func__, NULL, LOGLEVEL);
if (LOGLEVEL >= 7)
{
sprintf(log_buf, "calling on_job_exit from %s", __func__);
log_event(PBSEVENT_JOB, PBS_EVENTCLASS_JOB, pjob->ji_qs.ji_jobid, log_buf);
}
pthread_mutex_lock(server.sv_attr_mutex);
KeepSeconds = attr_ifelse_long(
&pque->qu_attr[QE_ATR_KeepCompleted],
&server.sv_attr[SRV_ATR_KeepCompleted],
0);
pthread_mutex_unlock(server.sv_attr_mutex);
}
else
KeepSeconds = 0;
if (pjob != NULL)
{
set_task(WORK_Timed, time_now + KeepSeconds, on_job_exit, strdup(pjob->ji_qs.ji_jobid), FALSE);
}
else
has_mutex = FALSE;
} /* END else if ((pjob->ji_qs.ji_svrflags & JOB_SVFLG_CHECKPOINT_FILE) != 0) */
if (has_mutex == TRUE)
unlock_ji_mutex(pjob, __func__, "7", LOGLEVEL);
return(PBSE_NONE);
} /* END execute_job_delete() */
int req_rerunjob(
batch_request *preq)
{
int rc = PBSE_NONE;
job *pjob;
int MgrRequired = TRUE;
char log_buf[LOCAL_LOG_BUF_SIZE];
/* check if requestor is admin, job owner, etc */
if (!strcasecmp(preq->rq_ind.rq_rerun, "all"))
{
return(handle_requeue_all(preq));
}
if ((pjob = chk_job_request(preq->rq_ind.rq_rerun, preq)) == 0)
{
/* FAILURE */
/* chk_job_request calls req_reject() */
rc = PBSE_SYSTEM;
return rc; /* This needs to fixed to return an accurate error */
}
mutex_mgr pjob_mutex = mutex_mgr(pjob->ji_mutex, true);
/* the job must be running or completed */
if (pjob->ji_qs.ji_state >= JOB_STATE_EXITING)
{
if (pjob->ji_wattr[JOB_ATR_checkpoint_name].at_flags & ATR_VFLAG_SET)
{
/* allow end-users to rerun checkpointed jobs */
MgrRequired = FALSE;
}
}
else if (pjob->ji_qs.ji_state == JOB_STATE_RUNNING)
{
/* job is running */
/* NO-OP */
}
else if (pjob->ji_qs.ji_state == JOB_STATE_QUEUED)
{
// If we are already queued, then there is nothing to do.
rc = PBSE_NONE;
reply_ack(preq);
return(rc);
}
else
{
/* FAILURE - job is in bad state */
rc = PBSE_BADSTATE;
snprintf(log_buf, LOCAL_LOG_BUF_SIZE, "job %s is in a bad state",
preq->rq_ind.rq_rerun);
req_reject(rc, 0, preq, NULL, log_buf);
return rc;
}
if ((MgrRequired == TRUE) &&
((preq->rq_perm & (ATR_DFLAG_MGWR | ATR_DFLAG_OPWR)) == 0))
{
/* FAILURE */
rc = PBSE_PERM;
snprintf(log_buf, LOCAL_LOG_BUF_SIZE,
"additional permissions required (ATR_DFLAG_MGWR | ATR_DFLAG_OPWR)");
req_reject(rc, 0, preq, NULL, log_buf);
return rc;
}
/* the job must be rerunnable */
if (pjob->ji_wattr[JOB_ATR_rerunable].at_val.at_long == 0)
{
/* NOTE: should force override this constraint? maybe (???) */
/* no, the user is saying that the job will break, and
IEEE Std 1003.1 specifically says rerun is to be rejected
if rerunable==FALSE -garrick */
rc = PBSE_NORERUN;
snprintf(log_buf, LOCAL_LOG_BUF_SIZE, "job %s not rerunnable",
preq->rq_ind.rq_rerun);
req_reject(rc, 0, preq, NULL, log_buf);
return rc;
}
if (pjob->ji_qs.ji_state == JOB_STATE_RUNNING)
{
/* ask MOM to kill off the job if it is running */
int delay = 0;
pbs_queue *pque;
// Apply the user delay first so it takes precedence.
if (pjob->ji_wattr[JOB_ATR_user_kill_delay].at_flags & ATR_VFLAG_SET)
delay = pjob->ji_wattr[JOB_ATR_user_kill_delay].at_val.at_long;
if ((pque = get_jobs_queue(&pjob)) != NULL)
{
mutex_mgr pque_mutex = mutex_mgr(pque->qu_mutex, true);
mutex_mgr server_mutex = mutex_mgr(server.sv_attr_mutex, false);
if (delay == 0)
{
delay = attr_ifelse_long(&pque->qu_attr[QE_ATR_KillDelay],
&server.sv_attr[SRV_ATR_KillDelay],
0);
}
}
else
{
/* why is the pque null. Something went wrong */
snprintf(log_buf, LOCAL_LOG_BUF_SIZE, "jobid %s returned a null queue", pjob->ji_qs.ji_jobid);
req_reject(PBSE_UNKQUE, 0, preq, NULL, log_buf);
return(PBSE_UNKQUE);
}
pjob->ji_qs.ji_substate = JOB_SUBSTATE_RERUN;
if (delay != 0)
{
static const char *rerun = "rerun";
char *extra = strdup(rerun);
get_batch_request_id(preq);
/* If a qrerun -f is given requeue the job regardless of the outcome of issue_signal*/
if ((preq->rq_extend) &&
(!strncasecmp(preq->rq_extend, RERUNFORCE, strlen(RERUNFORCE))))
{
std::string extend = RERUNFORCE;
batch_request *dup = new batch_request(*preq);
get_batch_request_id(dup);
rc = issue_signal(&pjob, "SIGTERM", delay_and_send_sig_kill, extra, strdup(dup->rq_id.c_str()));
if (rc == PBSE_NORELYMOM)
{
dup->rq_reply.brp_code = PBSE_NORELYMOM;
pjob_mutex.unlock();
post_rerun(dup);
pjob = svr_find_job(preq->rq_ind.rq_signal.rq_jid, FALSE);
if (pjob == NULL)
{
delete dup;
return(PBSE_NONE);
}
pjob_mutex.set_lock_state(true);
rc = PBSE_NONE;
}
delete dup;
}
else
{
rc = issue_signal(&pjob, "SIGTERM", delay_and_send_sig_kill, extra, strdup(preq->rq_id.c_str()));
if (rc != PBSE_NONE)
{
/* cant send to MOM */
req_reject(rc, 0, preq, NULL, NULL);
}
return(rc);
}
}
else
{
static const char *rerun = "rerun";
char *extra = strdup(rerun);
/* If a qrerun -f is given requeue the job regardless of the outcome of issue_signal*/
if (preq->rq_extend && !strncasecmp(preq->rq_extend, RERUNFORCE, strlen(RERUNFORCE)))
{
std::string extend = RERUNFORCE;
rc = issue_signal(&pjob, "SIGKILL", post_rerun, extra, strdup(extend.c_str()));
if (rc == PBSE_NORELYMOM)
rc = PBSE_NONE;
}
else
rc = issue_signal(&pjob, "SIGKILL", post_rerun, extra, NULL);
}
}
else
{
if (pjob->ji_wattr[JOB_ATR_hold].at_val.at_long == HOLD_n)
{
svr_setjobstate(pjob, JOB_STATE_QUEUED, JOB_SUBSTATE_QUEUED, FALSE);
}
else
{
svr_setjobstate(pjob, JOB_STATE_HELD, JOB_SUBSTATE_HELD, FALSE);
}
/* reset some job attributes */
pjob->ji_wattr[JOB_ATR_comp_time].at_flags &= ~ATR_VFLAG_SET;
pjob->ji_wattr[JOB_ATR_reported].at_flags &= ~ATR_VFLAG_SET;
set_statechar(pjob);
rc = -1;
}
/* finalize_rerunjob will return with pjob->ji_mutex unlocked */
pjob_mutex.set_unlock_on_exit(false);
return finalize_rerunjob(preq,pjob,rc);
}
int issue_to_svr(
char *servern, /* I */
struct batch_request *preq, /* I */
void (*replyfunc) (struct work_task *)) /* I */
{
int rc = PBSE_NONE;
int do_retry = 0;
int handle;
int my_err = 0;
pbs_net_t svraddr;
char *svrname;
unsigned int port = pbs_server_port_dis;
struct work_task *pwt;
time_t time_now = time(NULL);
snprintf(preq->rq_host, sizeof(preq->rq_host), "%s", servern);
preq->rq_fromsvr = 1;
preq->rq_perm = ATR_DFLAG_MGRD | ATR_DFLAG_MGWR | ATR_DFLAG_SvWR;
svrname = parse_servername(servern, &port);
svraddr = get_hostaddr(&my_err,svrname);
free(svrname);
if (svraddr == (pbs_net_t)0)
{
if (my_err == PBS_NET_RC_RETRY)
{
/* Non fatal error - retry */
do_retry = 1;
}
}
else
{
handle = svr_connect(svraddr, port, &my_err, NULL, NULL, ToServerDIS);
if (handle >= 0)
{
if (((rc = issue_Drequest(handle, preq)) == PBSE_NONE) &&
(handle != PBS_LOCAL_CONNECTION))
{
/* preq is already freed if handle == PBS_LOCAL_CONNECTION - a reply
* has always been sent */
rc = preq->rq_reply.brp_code;
}
return(rc);
}
else if (handle == PBS_NET_RC_RETRY)
{
do_retry = 1;
}
}
/* if reached here, it didn`t go, do we retry? */
if (do_retry)
{
if (preq->rq_id == NULL)
get_batch_request_id(preq);
pwt = set_task(WORK_Timed, (long)(time_now + PBS_NET_RETRY_TIME), reissue_to_svr, preq->rq_id, TRUE);
pwt->wt_parmfunc = replyfunc;
pthread_mutex_unlock(pwt->wt_mutex);
return(PBSE_NONE);
}
/* FAILURE */
return(PBSE_INTERNAL);
} /* END issue_to_svr() */
int req_signaljob(
batch_request *preq) /* I */
{
job *pjob;
int rc;
char log_buf[LOCAL_LOG_BUF_SIZE];
batch_request *dup_req = NULL;
/* preq free'd in error cases */
if ((pjob = chk_job_request(preq->rq_ind.rq_signal.rq_jid, preq)) == 0)
{
return(PBSE_NONE);
}
mutex_mgr job_mutex(pjob->ji_mutex, true);
/* the job must be running */
if (pjob->ji_qs.ji_state != JOB_STATE_RUNNING)
{
req_reject(PBSE_BADSTATE, 0, preq, NULL, NULL);
return(PBSE_NONE);
}
/* Special pseudo signals for suspend and resume require op/mgr */
if (!strcmp(preq->rq_ind.rq_signal.rq_signame, SIG_RESUME) ||
!strcmp(preq->rq_ind.rq_signal.rq_signame, SIG_SUSPEND))
{
if ((preq->rq_perm & (ATR_DFLAG_OPRD | ATR_DFLAG_OPWR | ATR_DFLAG_MGRD | ATR_DFLAG_MGWR)) == 0)
{
/* for suspend/resume, must be mgr/op */
req_reject(PBSE_PERM, 0, preq, NULL, NULL);
return(PBSE_NONE);
}
}
/* save job ptr for post_signal_req() */
preq->rq_extra = strdup(pjob->ji_qs.ji_jobid);
/* FIXME: need a race-free check for available free subnodes before
* resuming a suspended job */
#ifdef DONOTSUSPINTJOB
/* interactive jobs don't resume correctly so don't allow a suspend */
if (!strcmp(preq->rq_ind.rq_signal.rq_signame, SIG_SUSPEND) &&
(pjob->ji_wattr[JOB_ATR_interactive].at_flags & ATR_VFLAG_SET) &&
(pjob->ji_wattr[JOB_ATR_interactive].at_val.at_long > 0))
{
req_reject(PBSE_JOBTYPE, 0, preq, NULL, NULL);
return(PBSE_NONE);
}
#endif
if (LOGLEVEL >= 6)
{
char ipstr[128];
sprintf(log_buf, "relaying signal request to mom %s", netaddr_long(pjob->ji_qs.ji_un.ji_exect.ji_momaddr,ipstr));
log_record(PBSEVENT_SCHED,PBS_EVENTCLASS_REQUEST,"req_signaljob",log_buf);
}
/* send reply for asynchronous suspend */
if (preq->rq_type == PBS_BATCH_AsySignalJob)
{
/* reply_ack will free preq. We need to copy it before we call reply_ack */
batch_request *new_preq;
new_preq = duplicate_request(preq, -1);
if (new_preq == NULL)
{
sprintf(log_buf, "failed to duplicate batch request");
log_event(PBSEVENT_JOB, PBS_EVENTCLASS_JOB, __func__, log_buf);
return(PBSE_MEM_MALLOC);
}
get_batch_request_id(new_preq);
reply_ack(new_preq);
preq->rq_noreply = TRUE;
}
/* pass the request on to MOM */
if ((dup_req = duplicate_request(preq)) == NULL)
{
req_reject(PBSE_SYSTEM, 0, preq, NULL, "can not allocate memory");
}
/* The dup_req is freed in relay_to_mom (failure)
* or in issue_Drequest (success) */
else
{
rc = relay_to_mom(&pjob, dup_req, NULL);
if (pjob != NULL)
job_mutex.unlock();
else
job_mutex.set_unlock_on_exit(false);
if (rc != PBSE_NONE)
{
free_br(dup_req);
req_reject(rc, 0, preq, NULL, NULL); /* unable to get to MOM */
}
else
{
post_signal_req(dup_req);
free_br(preq);
}
}
/* If successful we ack after mom replies to us, we pick up in post_signal_req() */
return(PBSE_NONE);
} /* END req_signaljob() */
int req_rerunjob(
struct batch_request *preq)
{
int rc = PBSE_NONE;
job *pjob;
int MgrRequired = TRUE;
char log_buf[LOCAL_LOG_BUF_SIZE];
/* check if requestor is admin, job owner, etc */
if ((pjob = chk_job_request(preq->rq_ind.rq_rerun, preq)) == 0)
{
/* FAILURE */
/* chk_job_request calls req_reject() */
rc = PBSE_SYSTEM;
return rc; /* This needs to fixed to return an accurate error */
}
/* the job must be running or completed */
if (pjob->ji_qs.ji_state >= JOB_STATE_EXITING)
{
if (pjob->ji_wattr[JOB_ATR_checkpoint_name].at_flags & ATR_VFLAG_SET)
{
/* allow end-users to rerun checkpointed jobs */
MgrRequired = FALSE;
}
}
else if (pjob->ji_qs.ji_state == JOB_STATE_RUNNING)
{
/* job is running */
/* NO-OP */
}
else
{
/* FAILURE - job is in bad state */
rc = PBSE_BADSTATE;
snprintf(log_buf, LOCAL_LOG_BUF_SIZE, "job %s is in a bad state",
preq->rq_ind.rq_rerun);
req_reject(rc, 0, preq, NULL, log_buf);
unlock_ji_mutex(pjob, __func__, "2", LOGLEVEL);
return rc;
}
if ((MgrRequired == TRUE) &&
((preq->rq_perm & (ATR_DFLAG_MGWR | ATR_DFLAG_OPWR)) == 0))
{
/* FAILURE */
rc = PBSE_PERM;
snprintf(log_buf, LOCAL_LOG_BUF_SIZE,
"additional permissions required (ATR_DFLAG_MGWR | ATR_DFLAG_OPWR)");
req_reject(rc, 0, preq, NULL, log_buf);
unlock_ji_mutex(pjob, __func__, "3", LOGLEVEL);
return rc;
}
/* the job must be rerunnable */
if (pjob->ji_wattr[JOB_ATR_rerunable].at_val.at_long == 0)
{
/* NOTE: should force override this constraint? maybe (???) */
/* no, the user is saying that the job will break, and
IEEE Std 1003.1 specifically says rerun is to be rejected
if rerunable==FALSE -garrick */
rc = PBSE_NORERUN;
snprintf(log_buf, LOCAL_LOG_BUF_SIZE, "job %s not rerunnable",
preq->rq_ind.rq_rerun);
req_reject(rc, 0, preq, NULL, log_buf);
unlock_ji_mutex(pjob, __func__, "4", LOGLEVEL);
return rc;
}
if (pjob->ji_qs.ji_state == JOB_STATE_RUNNING)
{
/* ask MOM to kill off the job if it is running */
int delay = 0;
pbs_queue *pque;
if ((pque = get_jobs_queue(&pjob)) != NULL)
{
mutex_mgr pque_mutex = mutex_mgr(pque->qu_mutex, true);
pthread_mutex_lock(server.sv_attr_mutex);
delay = attr_ifelse_long(&pque->qu_attr[QE_ATR_KillDelay],
&server.sv_attr[SRV_ATR_KillDelay],
0);
pthread_mutex_unlock(server.sv_attr_mutex);
}
else if (pjob == NULL)
{
rc = PBSE_NORERUN;
unlock_ji_mutex(pjob, __func__, "4", LOGLEVEL);
return rc;
}
if(delay != 0)
{
static const char *rerun = "rerun";
char *extra = strdup(rerun);
get_batch_request_id(preq);
if ((rc = issue_signal(&pjob, "SIGTERM", delay_and_send_sig_kill, extra, strdup(preq->rq_id))))
{
/* cant send to MOM */
req_reject(rc, 0, preq, NULL, NULL);
}
unlock_ji_mutex(pjob, __func__, "4", LOGLEVEL);
return rc;
}
else
{
static const char *rerun = "rerun";
char *extra = strdup(rerun);
rc = issue_signal(&pjob, "SIGKILL", post_rerun, extra, NULL);
}
}
else
{
if (pjob->ji_wattr[JOB_ATR_hold].at_val.at_long == HOLD_n)
{
svr_setjobstate(pjob, JOB_STATE_QUEUED, JOB_SUBSTATE_QUEUED, FALSE);
}
else
{
svr_setjobstate(pjob, JOB_STATE_HELD, JOB_SUBSTATE_HELD, FALSE);
}
/* reset some job attributes */
pjob->ji_wattr[JOB_ATR_comp_time].at_flags &= ~ATR_VFLAG_SET;
pjob->ji_wattr[JOB_ATR_reported].at_flags &= ~ATR_VFLAG_SET;
set_statechar(pjob);
rc = -1;
}
return finalize_rerunjob(preq,pjob,rc);
}
