void *req_messagejob(
batch_request *preq) /* I */
{
job *pjob;
int rc;
batch_request *dup_req = NULL;
if ((pjob = chk_job_request(preq->rq_ind.rq_message.rq_jid, preq)) == NULL)
return(NULL);
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(NULL);
}
if ((rc = copy_batchrequest(&dup_req, preq, 0, -1)) != 0)
{
req_reject(PBSE_MEM_MALLOC, 0, preq, NULL, NULL);
}
/* pass the request on to MOM */
/* The dup_req is freed in relay_to_mom (failure)
* or in issue_Drequest (success) */
else if ((rc = relay_to_mom(&pjob, dup_req, NULL)) != PBSE_NONE)
{
req_reject(rc, 0, preq, NULL, NULL); /* unable to get to MOM */
free_br(dup_req);
}
else
{
post_message_req(dup_req);
free_br(preq);
}
/* After MOM acts and replies to us, we pick up in post_message_req() */
if (pjob == NULL)
job_mutex.set_lock_on_exit(false);
return(NULL);
} /* END req_messagejob() */
void *req_messagejob(
void *vp)
{
struct batch_request *preq = (struct batch_request *)vp;
job *pjob;
int rc;
struct batch_request *dup_req = NULL;
if ((pjob = chk_job_request(preq->rq_ind.rq_message.rq_jid, preq)) == NULL)
return(NULL);
/* the job must be running */
if (pjob->ji_qs.ji_state != JOB_STATE_RUNNING)
{
req_reject(PBSE_BADSTATE, 0, preq, NULL, NULL);
unlock_ji_mutex(pjob, __func__, "1", LOGLEVEL);
return(NULL);
}
if ((rc = copy_batchrequest(&dup_req, preq, 0, -1)) != 0)
{
req_reject(PBSE_MEM_MALLOC, 0, preq, NULL, NULL);
}
/* pass the request on to MOM */
/* The dup_req is freed in relay_to_mom (failure)
* or in issue_Drequest (success) */
else if ((rc = relay_to_mom(&pjob, dup_req, post_message_req)) != 0)
req_reject(rc, 0, preq, NULL, NULL); /* unable to get to MOM */
else
free_br(preq);
/* After MOM acts and replies to us, we pick up in post_message_req() */
if (pjob != NULL)
unlock_ji_mutex(pjob, __func__, "2", LOGLEVEL);
return(NULL);
} /* END req_messagejob() */
/*
* modify_whole_array()
* modifies the entire job array
* @SEE req_modify_array PARENT
*/
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 = 0;
int mom_relay = 0;
char log_buf[LOCAL_LOG_BUF_SIZE];
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 */
rc = modify_job((void **)&pjob, plist, preq, checkpoint_req, NO_MOM_RELAY);
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 */
rc = copy_batchrequest(&array_req, preq, 0, i);
if (rc != 0)
{
unlock_ji_mutex(pjob, __func__, "1", LOGLEVEL);
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, post_modify_arrayreq)))
{
if (pjob != 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__, "2", LOGLEVEL);
}
return(rc); /* unable to get to MOM */
}
}
if (pjob != NULL)
unlock_ji_mutex(pjob, __func__, "3", LOGLEVEL);
}
} /* END foreach job in array */
if (mom_relay)
{
preq->rq_refcount--;
if (preq->rq_refcount == 0)
{
free_br(preq);
}
return(PBSE_RELAYED_TO_MOM);
}
return(rc);
} /* END modify_whole_array() */
int modify_job(
void **j, /* O */
svrattrl *plist, /* I */
struct batch_request *preq, /* I */
int checkpoint_req, /* I */
int flag) /* I */
{
int bad = 0;
int i;
int newstate;
int newsubstate;
resource_def *prsd;
int rc;
int sendmom = 0;
int copy_checkpoint_files = FALSE;
char log_buf[LOCAL_LOG_BUF_SIZE];
struct batch_request *dup_req = NULL;
job *pjob = (job *)*j;
if (pjob == NULL)
{
sprintf(log_buf, "job structure is NULL");
log_err(PBSE_IVALREQ, __func__, log_buf);
return(PBSE_IVALREQ);
}
/* cannot be in exiting or transit, exiting has already been checked */
if (pjob->ji_qs.ji_state == JOB_STATE_TRANSIT)
{
/* FAILURE */
snprintf(log_buf,sizeof(log_buf),
"Cannot modify job '%s' in transit\n",
pjob->ji_qs.ji_jobid);
log_err(PBSE_BADSTATE, __func__, log_buf);
return(PBSE_BADSTATE);
}
if (((checkpoint_req == CHK_HOLD) || (checkpoint_req == CHK_CONT)) &&
(pjob->ji_qs.ji_substate == JOB_SUBSTATE_RUNNING))
{
/* May need to request copy of the checkpoint file from mom */
copy_checkpoint_files = TRUE;
if (checkpoint_req == CHK_HOLD)
{
sprintf(log_buf,"setting jobsubstate for %s to RERUN\n", pjob->ji_qs.ji_jobid);
pjob->ji_qs.ji_substate = JOB_SUBSTATE_RERUN;
job_save(pjob, SAVEJOB_QUICK, 0);
log_event(PBSEVENT_JOB, PBS_EVENTCLASS_JOB, pjob->ji_qs.ji_jobid, log_buf);
/* remove checkpoint restart file if there is one */
if (pjob->ji_wattr[JOB_ATR_restart_name].at_flags & ATR_VFLAG_SET)
{
cleanup_restart_file(pjob);
}
}
}
/* if job is running, special checks must be made */
/* NOTE: must determine if job exists down at MOM - this will occur if
job is running, job is held, or job was held and just barely
released (ie qhold/qrls) */
/* COMMENTED OUT BY JOSH B IN 2.3 DUE TO MAJOR PROBLEMS w/ CUSTOMERS
* --FIX and uncomment once we know what is really going on.
*
* We now know that ji_destin gets set on a qmove and that the mom does not
* have the job at that point.
*
if ((pjob->ji_qs.ji_state == JOB_STATE_RUNNING) ||
((pjob->ji_qs.ji_state == JOB_STATE_HELD) && (pjob->ji_qs.ji_destin[0] != '\0')) ||
((pjob->ji_qs.ji_state == JOB_STATE_QUEUED) && (pjob->ji_qs.ji_destin[0] != '\0')))
*/
if (pjob->ji_qs.ji_state == JOB_STATE_RUNNING)
{
while (plist != NULL)
{
/* is the pbs_attribute modifiable in RUN state ? */
i = find_attr(job_attr_def, plist->al_name, JOB_ATR_LAST);
if ((i < 0) ||
((job_attr_def[i].at_flags & ATR_DFLAG_ALTRUN) == 0))
{
/* FAILURE */
snprintf(log_buf,sizeof(log_buf),
"Cannot modify attribute '%s' while running\n",
plist->al_name);
log_err(PBSE_MODATRRUN, __func__, log_buf);
return PBSE_MODATRRUN;
}
/* NOTE: only explicitly specified job attributes are routed down to MOM */
if (i == JOB_ATR_resource)
{
/* is the specified resource modifiable while */
/* the job is running */
prsd = find_resc_def(svr_resc_def, plist->al_resc, svr_resc_size);
if (prsd == NULL)
{
/* FAILURE */
snprintf(log_buf,sizeof(log_buf),
"Unknown attribute '%s'\n",
plist->al_name);
log_err(PBSE_UNKRESC, __func__, log_buf);
return(PBSE_UNKRESC);
}
if ((prsd->rs_flags & ATR_DFLAG_ALTRUN) == 0)
{
/* FAILURE */
snprintf(log_buf,sizeof(log_buf),
"Cannot modify attribute '%s' while running\n",
plist->al_name);
log_err(PBSE_MODATRRUN, __func__, log_buf);
return(PBSE_MODATRRUN);
}
sendmom = 1;
}
/*
else if ((i == JOB_ATR_checkpoint_name) || (i == JOB_ATR_variables))
{
sendmom = 1;
}
*/
plist = (svrattrl *)GET_NEXT(plist->al_link);
}
} /* END if (pjob->ji_qs.ji_state == JOB_STATE_RUNNING) */
/* modify the job's attributes */
bad = 0;
plist = (svrattrl *)GET_NEXT(preq->rq_ind.rq_modify.rq_attr);
rc = modify_job_attr(pjob, plist, preq->rq_perm, &bad);
if (rc)
{
/* FAILURE */
snprintf(log_buf,sizeof(log_buf),
"Cannot set attributes for job '%s'\n",
pjob->ji_qs.ji_jobid);
log_err(rc, __func__, log_buf);
if (rc == PBSE_JOBNOTFOUND)
*j = NULL;
return(rc);
}
/* Reset any defaults resource limit which might have been unset */
set_resc_deflt(pjob, NULL, FALSE);
/* if job is not running, may need to change its state */
if (pjob->ji_qs.ji_state != JOB_STATE_RUNNING)
{
svr_evaljobstate(pjob, &newstate, &newsubstate, 0);
svr_setjobstate(pjob, newstate, newsubstate, FALSE);
}
else
{
job_save(pjob, SAVEJOB_FULL, 0);
}
sprintf(log_buf, msg_manager, msg_jobmod, preq->rq_user, preq->rq_host);
log_event(PBSEVENT_JOB,PBS_EVENTCLASS_JOB,pjob->ji_qs.ji_jobid,log_buf);
/* if a resource limit changed for a running job, send to MOM */
if (sendmom)
{
/* if the NO_MOM_RELAY flag is set the calling function will call
relay_to_mom so we do not need to do it here */
if (flag != NO_MOM_RELAY)
{
/* The last number is unused unless this is an array */
if ((rc = copy_batchrequest(&dup_req, preq, 0, -1)) != 0)
{
}
/* The dup_req is freed in relay_to_mom (failure)
* or in issue_Drequest (success) */
else if ((rc = relay_to_mom(&pjob, dup_req, post_modify_req)))
{
if (pjob != 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);
}
return(rc); /* unable to get to MOM */
}
}
return(PBSE_RELAYED_TO_MOM);
}
if (copy_checkpoint_files)
{
struct batch_request *momreq = 0;
momreq = cpy_checkpoint(momreq, pjob, JOB_ATR_checkpoint_name, CKPT_DIR_OUT);
if (momreq != NULL)
{
/* have files to copy */
momreq->rq_extra = strdup(pjob->ji_qs.ji_jobid);
/* The momreq is freed in relay_to_mom (failure)
* or in issue_Drequest (success) */
if (checkpoint_req == CHK_HOLD)
{
rc = relay_to_mom(&pjob, momreq, chkpt_xfr_hold);
}
else
{
rc = relay_to_mom(&pjob, momreq, chkpt_xfr_done);
}
if (rc != 0)
{
if (pjob != 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);
}
return(PBSE_NONE); /* come back when mom replies */
}
}
else
{
log_err(-1, __func__, "Failed to get batch request");
}
}
return(PBSE_NONE);
} /* END modify_job() */
int req_signaljob(
void *vp) /* I */
{
struct batch_request *preq = (struct batch_request *)vp;
job *pjob;
int rc;
char log_buf[LOCAL_LOG_BUF_SIZE];
struct 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);
}
/* the job must be running */
if (pjob->ji_qs.ji_state != JOB_STATE_RUNNING)
{
req_reject(PBSE_BADSTATE, 0, preq, NULL, NULL);
unlock_ji_mutex(pjob, __func__, (char *)"1", LOGLEVEL);
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);
unlock_ji_mutex(pjob, __func__, (char *)"2", LOGLEVEL);
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);
unlock_ji_mutex(pjob, __func__, (char *)"3", LOGLEVEL);
return(PBSE_NONE);
}
#endif
if (LOGLEVEL >= 6)
{
sprintf(log_buf, "relaying signal request to mom %lu", pjob->ji_qs.ji_un.ji_exect.ji_momaddr);
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(preq);
preq->rq_noreply = TRUE;
}
/* pass the request on to MOM */
if ((rc = copy_batchrequest(&dup_req, preq, 0, -1)) != 0)
{
req_reject(rc, 0, preq, NULL, "can not allocate memory");
unlock_ji_mutex(pjob, __func__, (char *)"4", LOGLEVEL);
}
/* 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)
unlock_ji_mutex(pjob, __func__, (char *)"4", LOGLEVEL);
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 modify_array_range(
job_array *pa, /* I/O */
char *range, /* I */
svrattrl *plist, /* I */
struct batch_request *preq, /* I */
int checkpoint_req) /* I */
{
char id[] = "modify_array_range";
tlist_head tl;
int i, rc;
int mom_relay = 0;
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->jobs[i] == NULL))
continue;
rc = modify_job(pa->jobs[i],plist,preq,checkpoint_req, NO_MOM_RELAY);
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 */
rc = copy_batchrequest(&array_req, preq, 0, i);
if (rc != 0)
{
return(rc);
}
preq->rq_refcount++;
if (mom_relay == 0)
{
preq->rq_refcount++;
}
mom_relay++;
if ((rc = relay_to_mom(
pa->jobs[i],
array_req,
post_modify_arrayreq)))
{
snprintf(log_buffer,sizeof(log_buffer),
"Unable to relay information to mom for job '%s'\n",
pa->jobs[i]->ji_qs.ji_jobid);
log_err(rc,id,log_buffer);
return(rc); /* unable to get to MOM */
}
}
}
/* 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() */
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() */
void *req_checkpointjob(
void *vp)
{
struct batch_request *preq = (struct batch_request *)vp;
job *pjob;
int rc;
pbs_attribute *pattr;
char log_buf[LOCAL_LOG_BUF_SIZE];
struct batch_request *dup_req = NULL;
if ((pjob = chk_job_request(preq->rq_ind.rq_manager.rq_objname, preq)) == NULL)
{
return(NULL);
}
pattr = &pjob->ji_wattr[JOB_ATR_checkpoint];
if ((pjob->ji_qs.ji_state == JOB_STATE_RUNNING) &&
((pattr->at_flags & ATR_VFLAG_SET) &&
((csv_find_string(pattr->at_val.at_str, "s") != NULL) ||
(csv_find_string(pattr->at_val.at_str, "c") != NULL) ||
(csv_find_string(pattr->at_val.at_str, "enabled") != NULL))))
{
/* have MOM attempt checkpointing */
if ((rc = copy_batchrequest(&dup_req, preq, 0, -1)) != 0)
{
req_reject(rc, 0, preq, NULL, "failure to allocate memory");
}
/* The dup_req is freed in relay_to_mom (failure)
* or in issue_Drequest (success) */
else if ((rc = relay_to_mom(&pjob, dup_req, NULL)) != PBSE_NONE)
{
req_reject(rc, 0, preq, NULL, NULL);
free_br(dup_req);
}
else
{
if (pjob != NULL)
{
pjob->ji_qs.ji_svrflags |= JOB_SVFLG_CHECKPOINT_FILE;
job_save(pjob, SAVEJOB_QUICK, 0);
log_event(PBSEVENT_JOB, PBS_EVENTCLASS_JOB, pjob->ji_qs.ji_jobid, log_buf);
unlock_ji_mutex(pjob, __func__, (char *)"1", LOGLEVEL);
pjob = NULL;
}
process_checkpoint_reply(dup_req);
}
}
else
{
/* Job does not have checkpointing enabled, so reject the request */
log_event(PBSEVENT_JOB,PBS_EVENTCLASS_JOB,pjob->ji_qs.ji_jobid,log_buf);
req_reject(PBSE_IVALREQ, 0, preq, NULL, "job is not checkpointable");
}
if (pjob != NULL)
unlock_ji_mutex(pjob, __func__, (char *)"1", LOGLEVEL);
return(NULL);
} /* END req_checkpointjob() */
int req_holdjob(
void *vp) /* I */
{
long *hold_val;
int newstate;
int newsub;
long old_hold;
job *pjob;
char *pset;
int rc;
pbs_attribute temphold;
pbs_attribute *pattr;
struct batch_request *preq = (struct batch_request *)vp;
char log_buf[LOCAL_LOG_BUF_SIZE];
struct batch_request *dup_req = NULL;
pjob = chk_job_request(preq->rq_ind.rq_hold.rq_orig.rq_objname, preq);
if (pjob == NULL)
{
return(PBSE_NONE);
}
/* cannot do anything until we decode the holds to be set */
if ((rc = get_hold(&preq->rq_ind.rq_hold.rq_orig.rq_attr, &pset,
&temphold)) != 0)
{
req_reject(rc, 0, preq, NULL, NULL);
unlock_ji_mutex(pjob, __func__, (char *)"1", LOGLEVEL);
return(PBSE_NONE);
}
/* if other than HOLD_u is being set, must have privil */
if ((rc = chk_hold_priv(temphold.at_val.at_long, preq->rq_perm)) != 0)
{
req_reject(rc, 0, preq, NULL, NULL);
unlock_ji_mutex(pjob, __func__, (char *)"2", LOGLEVEL);
return(PBSE_NONE);
}
hold_val = &pjob->ji_wattr[JOB_ATR_hold].at_val.at_long;
old_hold = *hold_val;
*hold_val |= temphold.at_val.at_long;
pjob->ji_wattr[JOB_ATR_hold].at_flags |= ATR_VFLAG_SET;
sprintf(log_buf, msg_jobholdset, pset, preq->rq_user, preq->rq_host);
pattr = &pjob->ji_wattr[JOB_ATR_checkpoint];
if ((pjob->ji_qs.ji_state == JOB_STATE_RUNNING) &&
((pattr->at_flags & ATR_VFLAG_SET) &&
((csv_find_string(pattr->at_val.at_str, "s") != NULL) ||
(csv_find_string(pattr->at_val.at_str, "c") != NULL) ||
(csv_find_string(pattr->at_val.at_str, "enabled") != NULL))))
{
/* have MOM attempt checkpointing */
if ((rc = copy_batchrequest(&dup_req, preq, 0, -1)) != 0)
{
req_reject(rc, 0, preq, NULL, "memory allocation failure");
}
/* The dup_req is freed in relay_to_mom (failure)
* or in issue_Drequest (success) */
else if ((rc = relay_to_mom(&pjob, dup_req, NULL)) != PBSE_NONE)
{
free_br(dup_req);
*hold_val = old_hold; /* reset to the old value */
req_reject(rc, 0, preq, NULL, NULL);
}
else
{
if (pjob != NULL)
{
pjob->ji_qs.ji_svrflags |= JOB_SVFLG_HASRUN | JOB_SVFLG_CHECKPOINT_FILE;
job_save(pjob, SAVEJOB_QUICK, 0);
/* fill in log_buf again, since relay_to_mom changed it */
sprintf(log_buf, msg_jobholdset, pset, preq->rq_user, preq->rq_host);
log_event(PBSEVENT_JOB, PBS_EVENTCLASS_JOB, pjob->ji_qs.ji_jobid, log_buf);
unlock_ji_mutex(pjob, __func__, (char *)"3", LOGLEVEL);
pjob = NULL;
req_reject(rc, 0, preq, NULL, "relay to mom failed");
}
process_hold_reply(dup_req);
}
}
#ifdef ENABLE_BLCR
else if (pjob->ji_qs.ji_state == JOB_STATE_RUNNING)
{
/*
* This system is configured with BLCR checkpointing to be used,
* but this Running job does not have checkpointing enabled,
* so we reject the request
*/
log_event(PBSEVENT_JOB,PBS_EVENTCLASS_JOB,pjob->ji_qs.ji_jobid,log_buf);
req_reject(PBSE_IVALREQ, 0, preq, NULL,
"job not held since checkpointing is expected but not enabled for job");
}
#endif
else
{
/* everything went well, may need to update the job state */
log_event(PBSEVENT_JOB,PBS_EVENTCLASS_JOB,pjob->ji_qs.ji_jobid,log_buf);
if (old_hold != *hold_val)
{
/* indicate attributes changed */
pjob->ji_modified = 1;
svr_evaljobstate(pjob, &newstate, &newsub, 0);
svr_setjobstate(pjob, newstate, newsub, FALSE);
}
reply_ack(preq);
}
if (pjob != NULL)
unlock_ji_mutex(pjob, __func__, (char *)"3", LOGLEVEL);
return(PBSE_NONE);
} /* END req_holdjob() */
/*
* modify_whole_array()
* modifies the entire job array
* @SEE req_modify_array PARENT
*/
int modify_whole_array(
job_array *pa, /* I/O */
svrattrl *plist, /* I */
struct batch_request *preq, /* I */
int checkpoint_req) /* I */
{
char id[] = "modify_whole_array";
int i;
int rc = 0;
int mom_relay = 0;
for (i = 0; i < pa->ai_qs.array_size; i++)
{
if (pa->jobs[i] == NULL)
continue;
/* NO_MOM_RELAY will prevent modify_job from calling relay_to_mom */
rc = modify_job(pa->jobs[i],plist,preq,checkpoint_req, NO_MOM_RELAY);
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 */
rc = copy_batchrequest(&array_req, preq, 0, i);
if(rc != 0)
{
return(rc);
}
preq->rq_refcount++;
if(mom_relay == 0)
{
preq->rq_refcount++;
}
mom_relay++;
if ((rc = relay_to_mom(
pa->jobs[i]->ji_qs.ji_un.ji_exect.ji_momaddr,
array_req,
post_modify_arrayreq)))
{
snprintf(log_buffer,sizeof(log_buffer),
"Unable to relay information to mom for job '%s'\n",
pa->jobs[i]->ji_qs.ji_jobid);
log_err(rc,id,log_buffer);
return(rc); /* unable to get to MOM */
}
}
}
if(mom_relay)
{
preq->rq_refcount--;
if(preq->rq_refcount == 0)
{
free_br(preq);
}
return(PBSE_RELAYED_TO_MOM);
}
return(rc);
} /* END modify_whole_array() */
