/*
* get_job_script_path()
*
* @pre-cond: parameters must be valid
* @post-cond: script_path will be populated with the path to the job's script on success
* @return: PBSE_NONE on success, PBSE_JOB_RECYCLED if job disappears or -1 if array can't be
* found for an array job.
*/
int get_job_script_path(
job *pjob,
std::string &script_path)
{
// get the adjusted path to the script
script_path = get_path_jobdata(pjob->ji_qs.ji_jobid, path_jobs);
if (pjob->ji_arraystructid[0] != '\0')
{
job_array *pa = get_jobs_array(&pjob);
if (pa != NULL)
{
script_path += pa->ai_qs.fileprefix;
unlock_ai_mutex(pa, __func__, NULL, LOGLEVEL);
}
else if (pjob == NULL)
return(PBSE_JOB_RECYCLED);
else
return(-1);
}
else
{
script_path += pjob->ji_qs.ji_fileprefix;
}
script_path += JOB_SCRIPT_SUFFIX;
return(PBSE_NONE);
} /* END get_job_script_path() */
END_TEST
START_TEST(get_jobs_array_test)
{
struct job *test_job = NULL;
struct job_array *result = get_jobs_array(NULL);
fail_unless(result == NULL, "NULL input pointer to pointer to job fail");
result = get_jobs_array(&test_job);
fail_unless(result == NULL, "NULL input pointer to job fail");
test_job = job_alloc();
result = get_jobs_array(&test_job);
fail_unless(result == NULL, "get job array fail");
}
int modify_whole_array(
job_array *pa, /* I/O */
svrattrl *plist, /* I */
struct batch_request *preq, /* I */
int checkpoint_req) /* I */
{
int i;
int rc = PBSE_NONE;
int modify_job_rc = PBSE_NONE;
job *pjob;
for (i = 0; i < pa->ai_qs.array_size; i++)
{
if (pa->job_ids[i] == NULL)
continue;
if ((pjob = svr_find_job(pa->job_ids[i], FALSE)) == NULL)
{
free(pa->job_ids[i]);
pa->job_ids[i] = NULL;
}
else
{
/* NO_MOM_RELAY will prevent modify_job from calling relay_to_mom */
batch_request *array_req = duplicate_request(preq, i);
mutex_mgr job_mutex(pjob->ji_mutex, true);
pthread_mutex_unlock(pa->ai_mutex);
array_req->rq_noreply = TRUE;
rc = modify_job((void **)&pjob, plist, array_req, checkpoint_req, NO_MOM_RELAY);
if (rc != PBSE_NONE)
{
modify_job_rc = rc;
}
pa = get_jobs_array(&pjob);
if (pa == NULL)
{
if (pjob == NULL)
job_mutex.set_lock_on_exit(false);
return(PBSE_JOB_RECYCLED);
}
if (pjob == NULL)
{
pa->job_ids[i] = NULL;
job_mutex.set_lock_on_exit(false);
continue;
}
}
} /* END foreach job in array */
return(modify_job_rc);
} /* END modify_whole_array() */
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 modify_array_range(
job_array *pa, /* I/O */
char *range, /* I */
svrattrl *plist, /* I */
struct batch_request *preq, /* I */
int checkpoint_req) /* I */
{
char log_buf[LOCAL_LOG_BUF_SIZE];
tlist_head tl;
int i;
int rc;
int mom_relay = 0;
job *pjob;
array_request_node *rn;
array_request_node *to_free;
CLEAR_HEAD(tl);
if (parse_array_request(range,&tl) > 0)
{
/* don't hold the jobs if range error */
return(FAILURE);
}
else
{
/* hold just that range from the array */
rn = (array_request_node*)GET_NEXT(tl);
while (rn != NULL)
{
for (i = rn->start; i <= rn->end; i++)
{
if ((i >= pa->ai_qs.array_size) ||
(pa->job_ids[i] == NULL))
continue;
if ((pjob = svr_find_job(pa->job_ids[i], FALSE)) == NULL)
{
free(pa->job_ids[i]);
pa->job_ids[i] = NULL;
}
else
{
pthread_mutex_unlock(pa->ai_mutex);
rc = modify_job((void **)&pjob, plist, preq, checkpoint_req, NO_MOM_RELAY);
pa = get_jobs_array(&pjob);
if (pjob != NULL)
{
if (rc == PBSE_RELAYED_TO_MOM)
{
struct batch_request *array_req = NULL;
/* We told modify_job not to call relay_to_mom so we need to contact the mom */
if ((rc = copy_batchrequest(&array_req, preq, 0, i)) != PBSE_NONE)
{
return(rc);
}
preq->rq_refcount++;
if (mom_relay == 0)
{
preq->rq_refcount++;
}
mom_relay++;
/* The array_req is freed in relay_to_mom (failure)
* or in issue_Drequest (success) */
if ((rc = relay_to_mom(&pjob, array_req, NULL)))
{
snprintf(log_buf,sizeof(log_buf),
"Unable to relay information to mom for job '%s'\n",
pjob->ji_qs.ji_jobid);
log_err(rc, __func__, log_buf);
unlock_ji_mutex(pjob, __func__, "1", LOGLEVEL);
return(rc); /* unable to get to MOM */
}
else
{
unlock_ji_mutex(pjob, __func__, "2", LOGLEVEL);
post_modify_arrayreq(array_req);
}
}
else
unlock_ji_mutex(pjob, __func__, "2", LOGLEVEL);
}
else
pa->job_ids[i] = NULL;
}
}
/* release mem */
to_free = rn;
rn = (array_request_node*)GET_NEXT(rn->request_tokens_link);
free(to_free);
}
}
if (mom_relay)
{
preq->rq_refcount--;
if (preq->rq_refcount == 0)
{
free_br(preq);
}
return(PBSE_RELAYED_TO_MOM);
}
return(PBSE_NONE);
} /* END modify_array_range() */
