void
req_relnodesjob(struct batch_request *preq)
{
int jt; /* job type */
job *pjob;
int rc;
char *jid;
int i, offset;
char *nodeslist = NULL;
char msg[LOG_BUF_SIZE];
if (preq == NULL)
return;
jid = preq->rq_ind.rq_relnodes.rq_jid;
if (jid == NULL)
return;
/*
** Returns job pointer for singleton job or "parent" of
** an array job.
*/
pjob = chk_job_request(jid, preq, &jt);
if (pjob == NULL) {
return;
}
if (jt == IS_ARRAY_NO) { /* a regular job is okay */
/* the job must be running */
if ((pjob->ji_qs.ji_state != JOB_STATE_RUNNING) ||
(pjob->ji_qs.ji_substate !=
JOB_SUBSTATE_RUNNING)) {
req_reject(PBSE_BADSTATE, 0, preq);
return;
}
}
else if (jt == IS_ARRAY_Single) { /* a single subjob is okay */
offset = subjob_index_to_offset(pjob,
get_index_from_jid(jid));
if (offset == -1) {
req_reject(PBSE_UNKJOBID, 0, preq);
return;
}
i = get_subjob_state(pjob, offset);
if (i == -1) {
req_reject(PBSE_IVALREQ, 0, preq);
return;
}
if (i != JOB_STATE_RUNNING) {
req_reject(PBSE_BADSTATE, 0, preq);
return;
}
if ((pjob = pjob->ji_ajtrk->tkm_tbl[offset].trk_psubjob) == NULL) {
req_reject(PBSE_UNKJOBID, 0, preq);
return;
}
if (pjob->ji_qs.ji_substate != JOB_SUBSTATE_RUNNING) {
req_reject(PBSE_BADSTATE, 0, preq);
return;
}
} else {
reply_text(preq, PBSE_NOSUP,
"not supported for Array Jobs or multiple sub-jobs");
return;
}
nodeslist = preq->rq_ind.rq_relnodes.rq_node_list;
if ((nodeslist != NULL) && (nodeslist[0] == '\0')) {
nodeslist = NULL;
}
rc = free_sister_vnodes(pjob, nodeslist, msg, LOG_BUF_SIZE, preq);
if (rc != 0) {
reply_text(preq, PBSE_SYSTEM, msg);
}
}
void
req_py_spawn(struct batch_request *preq)
{
int jt; /* job type */
job *pjob;
int rc;
char *jid = preq->rq_ind.rq_py_spawn.rq_jid;
int i, offset;
/*
** Returns job pointer for singleton job or "parent" of
** an array job.
*/
pjob = chk_job_request(jid, preq, &jt);
if (pjob == NULL)
return;
/* see if requestor is the job owner */
if (svr_chk_owner(preq, pjob) != 0) {
req_reject(PBSE_PERM, 0, preq);
return;
}
if (jt == IS_ARRAY_NO) { /* a regular job is okay */
/* the job must be running */
if ((pjob->ji_qs.ji_state != JOB_STATE_RUNNING) ||
(pjob->ji_qs.ji_substate !=
JOB_SUBSTATE_RUNNING)) {
req_reject(PBSE_BADSTATE, 0, preq);
return;
}
}
else if (jt == IS_ARRAY_Single) { /* a single subjob is okay */
offset = subjob_index_to_offset(pjob,
get_index_from_jid(jid));
if (offset == -1) {
req_reject(PBSE_UNKJOBID, 0, preq);
return;
}
i = get_subjob_state(pjob, offset);
if (i == -1) {
req_reject(PBSE_IVALREQ, 0, preq);
return;
}
if (i != JOB_STATE_RUNNING) {
req_reject(PBSE_BADSTATE, 0, preq);
return;
}
if ((pjob = pjob->ji_ajtrk->tkm_tbl[offset].trk_psubjob) == NULL) {
req_reject(PBSE_UNKJOBID, 0, preq);
return;
}
if (pjob->ji_qs.ji_substate != JOB_SUBSTATE_RUNNING) {
req_reject(PBSE_BADSTATE, 0, preq);
return;
}
} else {
reply_text(preq, PBSE_NOSUP,
"not supported for Array Jobs or multiple sub-jobs");
return;
}
/*
** Pass the request on to MOM. If this works, the function
** post_py_spawn_req will be called to handle the reply.
** If it fails, send the reply now.
*/
rc = relay_to_mom(pjob, preq, post_py_spawn_req);
if (rc)
req_reject(rc, 0, preq); /* unable to get to MOM */
}
/**
* @brief
* Support function for req_stat_job().
* Builds status reply for a single job id, which may be: a normal job,
* an Array job, a single subjob or a range of subjobs.
* Finds the job structure for the job id and calls either do_stat_of_a_job()
* or status_subjob() to build that actual status reply.
*
* @param[in,out] preq - pointer to the stat job batch request, reply updated
* @param[in] name - job id to be statused
* @param[in] dohistjobs - flag to include job if it is a history job
* @param[in] dosubjobs - flag to expand a Array job to include all subjobs
*
* @return int
* @retval PBSE_NONE (0) : no error
* @retval non-zero : PBS error code to return to client
*/
static int
stat_a_jobidname(struct batch_request *preq, char *name, int dohistjobs, int dosubjobs)
{
int i, indx, x, y, z;
char *pc;
char *range;
int rc;
job *pjob;
struct batch_reply *preply = &preq->rq_reply;
svrattrl *pal;
if ((i = is_job_array(name)) == IS_ARRAY_Single) {
pjob = find_arrayparent(name);
if (pjob == NULL) {
return (PBSE_UNKJOBID);
} else if ((!dohistjobs) && (rc = svr_chk_histjob(pjob))) {
return (rc);
}
indx = subjob_index_to_offset(pjob, get_index_from_jid(name));
if (indx != -1) {
pal = (svrattrl *)GET_NEXT(preq->rq_ind.rq_status.rq_attr);
rc = status_subjob(pjob, preq, pal, indx, &preply->brp_un.brp_status, &bad);
} else {
rc = PBSE_UNKJOBID;
}
return (rc); /* no job still needs to be stat-ed */
} else if ((i == IS_ARRAY_NO) || (i == IS_ARRAY_ArrayJob)) {
pjob = find_job(name);
if (pjob == NULL) {
return (PBSE_UNKJOBID);
} else if ((!dohistjobs) && (rc = svr_chk_histjob(pjob))) {
return (rc);
}
return (do_stat_of_a_job(preq, pjob, dohistjobs, dosubjobs));
} else {
/* range of sub jobs */
range = get_index_from_jid(name);
if (range == NULL) {
return (PBSE_IVALREQ);
}
pjob = find_arrayparent(name);
if (pjob == NULL) {
return (PBSE_UNKJOBID);
} else if ((!dohistjobs) && (rc = svr_chk_histjob(pjob))) {
return (rc);
}
pal = (svrattrl *)GET_NEXT(preq->rq_ind.rq_status.rq_attr);
while (1) {
if ((i=parse_subjob_index(range,&pc,&x,&y,&z,&i)) == -1) {
return (PBSE_IVALREQ);
} else if (i == 1)
break;
while (x <= y) {
indx = numindex_to_offset(pjob, x);
if (indx < 0) {
x += z;
continue;
}
rc = status_subjob(pjob, preq, pal, indx, &preply->brp_un.brp_status, &bad);
if (rc && (rc != PBSE_PERM)) {
return (rc);
}
x += z;
}
range = pc;
}
/* stat-ed the range, no more to stat for this id */
return (PBSE_NONE);
}
}
/**
* @brief
* create_subjob - create a Subjob from the parent Array Job
* Certain attributes are changed or left out
* @param[in] parent - pointer to parent Job
* @param[in] newjid - new job id
* @param[in] rc - return code
* @return pointer to new job
* @retval NULL - error
*/
job *
create_subjob(job *parent, char *newjid, int *rc)
{
pbs_list_head attrl;
int i;
int j;
int indx;
char *index;
attribute_def *pdef;
attribute *ppar;
attribute *psub;
svrattrl *psatl;
job *subj;
long eligibletime;
long time_msec;
#ifdef WIN32
struct _timeb tval;
#else
struct timeval tval;
#endif
if ((parent->ji_qs.ji_svrflags & JOB_SVFLG_ArrayJob) == 0) {
*rc = PBSE_IVALREQ;
return NULL; /* parent not an array job */
}
/* find and copy the index */
if ((index = get_index_from_jid(newjid)) == NULL) {
*rc = PBSE_IVALREQ;
return NULL;
}
if ((indx = subjob_index_to_offset(parent, index)) == -1) {
*rc = PBSE_UNKJOBID;
return NULL;
}
if (parent->ji_ajtrk->tkm_tbl[indx].trk_status != JOB_STATE_QUEUED) {
*rc = PBSE_BADSTATE;
return NULL;
}
/*
* allocate and clear basic structure
* cannot copy job attributes because cannot share strings and other
* malloc-ed data, so copy ji_qs as a whole and then copy the
* non-saved items before ji_qs.
*/
subj = job_alloc();
subj->ji_qs = parent->ji_qs; /* copy the fixed save area */
#ifdef PBS_CRED_GRIDPROXY
subj->ji_gsscontext = parent->ji_gsscontext;
#endif
subj->ji_qhdr = parent->ji_qhdr;
subj->ji_resvp = parent->ji_resvp;
subj->ji_myResv = parent->ji_myResv;
subj->ji_parentaj = parent;
strcpy(subj->ji_qs.ji_jobid, newjid); /* replace job id */
*subj->ji_qs.ji_fileprefix = '\0';
subj->ji_subjindx = indx;
/*
* now that is all done, copy the required attributes by
* encoding and then decoding into the new array. Then add the
* subjob specific attributes.
*/
resc_access_perm = ATR_DFLAG_ACCESS;
CLEAR_HEAD(attrl);
for (i = 0; attrs_to_copy[i] != JOB_ATR_LAST; i++) {
j = (int)attrs_to_copy[i];
ppar = &parent->ji_wattr[j];
psub = &subj->ji_wattr[j];
pdef = &job_attr_def[j];
if (pdef->at_encode(ppar, &attrl, pdef->at_name, NULL,
ATR_ENCODE_MOM, &psatl) > 0) {
for (psatl = (svrattrl *)GET_NEXT(attrl); psatl;
psatl = ((svrattrl *)GET_NEXT(psatl->al_link))) {
pdef->at_decode(psub, psatl->al_name, psatl->al_resc,
psatl->al_value);
}
/* carry forward the default bit if set */
psub->at_flags |= (ppar->at_flags & ATR_VFLAG_DEFLT);
free_attrlist(&attrl);
}
}
psub = &subj->ji_wattr[(int)JOB_ATR_array_id];
job_attr_def[(int)JOB_ATR_array_id].at_decode(psub, NULL, NULL,
parent->ji_qs.ji_jobid);
psub = &subj->ji_wattr[(int)JOB_ATR_array_index];
job_attr_def[(int)JOB_ATR_array_index].at_decode(psub, NULL, NULL, index);
/* Lastly, set or clear a few flags and link in the structure */
subj->ji_qs.ji_svrflags &= ~JOB_SVFLG_ArrayJob;
subj->ji_qs.ji_svrflags |= JOB_SVFLG_SubJob;
subj->ji_modified = 1; /* ** will likely take this out ** */
subj->ji_qs.ji_substate = JOB_SUBSTATE_TRANSICM;
(void)svr_setjobstate(subj, JOB_STATE_QUEUED, JOB_SUBSTATE_QUEUED);
subj->ji_wattr[(int)JOB_ATR_state].at_flags |= ATR_VFLAG_SET;
subj->ji_wattr[(int)JOB_ATR_substate].at_flags |= ATR_VFLAG_SET;
/* subjob needs to borrow eligible time from parent job array.
* expecting only to accrue eligible_time and nothing else.
*/
if (server.sv_attr[(int)SRV_ATR_EligibleTimeEnable].at_val.at_long == 1) {
eligibletime = parent->ji_wattr[(int)JOB_ATR_eligible_time].at_val.at_long;
if (parent->ji_wattr[(int)JOB_ATR_accrue_type].at_val.at_long == JOB_ELIGIBLE)
eligibletime += subj->ji_wattr[(int)JOB_ATR_sample_starttime].at_val.at_long - parent->ji_wattr[(int)JOB_ATR_sample_starttime].at_val.at_long;
subj->ji_wattr[(int)JOB_ATR_eligible_time].at_val.at_long = eligibletime;
subj->ji_wattr[(int)JOB_ATR_eligible_time].at_flags |= ATR_VFLAG_MODIFY | ATR_VFLAG_MODCACHE;
}
#ifdef WIN32
_ftime_s(&tval);
time_msec = (tval.time * 1000L) + tval.millitm;
#else
gettimeofday(&tval, NULL);
time_msec = (tval.tv_sec * 1000L) + (tval.tv_usec/1000L);
#endif
/* set the queue rank attribute */
subj->ji_wattr[(int)JOB_ATR_qrank].at_val.at_long = time_msec;
subj->ji_wattr[(int)JOB_ATR_qrank].at_flags |= ATR_VFLAG_SET|ATR_VFLAG_MODCACHE;
if (svr_enquejob(subj) != 0) {
job_purge(subj);
*rc = PBSE_IVALREQ;
return NULL;
}
*rc = PBSE_NONE;
return subj;
}
void
req_rerunjob(struct batch_request *preq)
{
int anygood = 0;
int i;
int j;
char *jid;
int jt; /* job type */
int offset;
char *pc;
job *pjob;
job *parent;
char *range;
char *vrange;
int x, y, z;
jid = preq->rq_ind.rq_signal.rq_jid;
parent = chk_job_request(jid, preq, &jt);
if (parent == (job *)0)
return; /* note, req_reject already called */
if ((preq->rq_perm & (ATR_DFLAG_MGWR | ATR_DFLAG_OPWR)) == 0) {
req_reject(PBSE_PERM, 0, preq);
return;
}
if (jt == IS_ARRAY_NO) {
/* just a regular job, pass it on down the line and be done */
req_rerunjob2(preq, parent);
return;
} else if (jt == IS_ARRAY_Single) {
/* single subjob, if running can signal */
offset = subjob_index_to_offset(parent, get_index_from_jid(jid));
if (offset == -1) {
req_reject(PBSE_UNKJOBID, 0, preq);
return;
}
i = get_subjob_state(parent, offset);
if (i == -1) {
req_reject(PBSE_IVALREQ, 0, preq);
return;
} else if (i == JOB_STATE_RUNNING) {
pjob = find_job(jid); /* get ptr to the subjob */
if (pjob) {
req_rerunjob2(preq, pjob);
} else {
req_reject(PBSE_BADSTATE, 0, preq);
return;
}
} else {
req_reject(PBSE_BADSTATE, 0, preq);
return;
}
return;
} else if (jt == IS_ARRAY_ArrayJob) {
/* The Array Job itself ... */
if (parent->ji_qs.ji_state != JOB_STATE_BEGUN) {
req_reject(PBSE_BADSTATE, 0, preq);
return;
}
/* for each subjob that is running, call req_rerunjob2 */
++preq->rq_refct; /* protect the request/reply struct */
/* Setting deleted subjobs count to 0,
* since all the deleted subjobs will be moved to Q state
*/
parent->ji_ajtrk->tkm_dsubjsct = 0;
for (i=0; i<parent->ji_ajtrk->tkm_ct; i++) {
if (get_subjob_state(parent, i) == JOB_STATE_RUNNING) {
pjob = find_job(mk_subjob_id(parent, i));
if (pjob) {
dup_br_for_subjob(preq, pjob, req_rerunjob2);
}
} else {
set_subjob_tblstate(parent, i, JOB_STATE_QUEUED);
}
}
/* if not waiting on any running subjobs, can reply; else */
/* it is taken care of when last running subjob responds */
if (--preq->rq_refct == 0)
reply_send(preq);
return;
}
/* what's left to handle is a range of subjobs, foreach subjob */
/* if running, all req_rerunjob2 */
range = get_index_from_jid(jid);
if (range == NULL) {
req_reject(PBSE_IVALREQ, 0, preq);
return;
}
/* first check that all in the subrange are in fact running */
vrange = range;
while (1) {
if ((i = parse_subjob_index(vrange, &pc, &x, &y, &z, &j)) == -1) {
req_reject(PBSE_IVALREQ, 0, preq);
return;
} else if (i == 1)
break;
while (x <= y) {
i = numindex_to_offset(parent, x);
if (i >= 0) {
if (get_subjob_state(parent, i) == JOB_STATE_RUNNING)
anygood++;
}
x += z;
}
vrange = pc;
}
if (anygood == 0) {
req_reject(PBSE_BADSTATE, 0, preq);
return;
}
/* now do the deed */
++preq->rq_refct; /* protect the request/reply struct */
while (1) {
if ((i = parse_subjob_index(range, &pc, &x, &y, &z, &j)) == -1) {
req_reject(PBSE_IVALREQ, 0, preq);
break;
} else if (i == 1)
break;
while (x <= y) {
i = numindex_to_offset(parent, x);
if (i < 0) {
x += z;
continue;
}
if (get_subjob_state(parent, i) == JOB_STATE_RUNNING) {
pjob = find_job(mk_subjob_id(parent, i));
if (pjob) {
dup_br_for_subjob(preq, pjob, req_rerunjob2);
}
}
x += z;
}
range = pc;
}
/* if not waiting on any running subjobs, can reply; else */
/* it is taken care of when last running subjob responds */
if (--preq->rq_refct == 0)
reply_send(preq);
return;
}