void stat_update(
struct batch_request *preq,
struct stat_cntl *cntl)
{
job *pjob;
struct batch_reply *preply;
struct brp_status *pstatus;
svrattrl *sattrl;
int oldsid;
int bad = 0;
time_t time_now = time(NULL);
char *msg_ptr = NULL;
char log_buf[LOCAL_LOG_BUF_SIZE];
preply = &preq->rq_reply;
if (preply->brp_un.brp_txt.brp_str != NULL)
{
msg_ptr = strstr(preply->brp_un.brp_txt.brp_str, PBS_MSG_EQUAL);
if (msg_ptr != NULL)
msg_ptr += strlen(PBS_MSG_EQUAL);
}
if (preply->brp_choice == BATCH_REPLY_CHOICE_Status)
{
pstatus = (struct brp_status *)GET_NEXT(preply->brp_un.brp_status);
while (pstatus != NULL)
{
if ((pjob = svr_find_job(pstatus->brp_objname, FALSE)) != NULL)
{
mutex_mgr job_mutex(pjob->ji_mutex, true);
sattrl = (svrattrl *)GET_NEXT(pstatus->brp_attr);
oldsid = pjob->ji_wattr[JOB_ATR_session_id].at_val.at_long;
modify_job_attr(
pjob,
sattrl,
ATR_DFLAG_MGWR | ATR_DFLAG_SvWR,
&bad);
if (oldsid != pjob->ji_wattr[JOB_ATR_session_id].at_val.at_long)
{
/* first save since running job (or the sid has changed), */
/* must save session id */
job_save(pjob, SAVEJOB_FULL, 0);
}
#ifdef USESAVEDRESOURCES
else
{
/* save so we can recover resources used */
job_save(pjob, SAVEJOB_FULL, 0);
}
#endif /* USESAVEDRESOURCES */
pjob->ji_momstat = time_now;
}
pstatus = (struct brp_status *)GET_NEXT(pstatus->brp_stlink);
} /* END while (pstatus != NULL) */
} /* END if (preply->brp_choice == BATCH_REPLY_CHOICE_Status) */
else if ((preply->brp_choice == BATCH_REPLY_CHOICE_Text) &&
(preply->brp_code == PBSE_UNKJOBID) &&
(msg_ptr != NULL) &&
(!strcmp(msg_ptr, preq->rq_ind.rq_status.rq_id)))
{
/* we sent a stat request, but mom says it doesn't know anything about
the job */
if ((pjob = svr_find_job(preq->rq_ind.rq_status.rq_id, FALSE)) != NULL)
{
/* job really isn't running any more - mom doesn't know anything about it
this can happen if a diskless node reboots and the mom_priv/jobs
directory is cleared, set its state to queued so job_abt doesn't
think it is still running */
mutex_mgr job_mutex(pjob->ji_mutex, true);
snprintf(log_buf, sizeof(log_buf),
"mother superior no longer recognizes %s as a valid job, aborting. Last reported time was %ld",
preq->rq_ind.rq_status.rq_id, pjob->ji_last_reported_time);
log_event(PBSEVENT_JOB, PBS_EVENTCLASS_JOB, __func__, log_buf);
svr_setjobstate(pjob, JOB_STATE_QUEUED, JOB_SUBSTATE_ABORT, FALSE);
rel_resc(pjob);
job_mutex.set_unlock_on_exit(false);
job_abt(&pjob, "Job does not exist on node");
/* TODO, if the job is rerunnable we should set its state back to queued */
}
}
else
{
snprintf(log_buf, sizeof(log_buf),
"Poll job request failed for job %s", preq->rq_ind.rq_status.rq_id);
log_err(preply->brp_code, __func__, log_buf);
}
cntl->sc_conn = -1;
if (cntl->sc_post)
cntl->sc_post(cntl); /* continue where we left off */
/* If sc_post has a value it is:
* req_stat_job_step2
* if so, it expects cntl to be free'd after the call
*/
free(cntl); /* a bit of a kludge but its saves an extra func */
return;
} /* END stat_update() */
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() */
void stat_update(
struct batch_request *preq,
struct stat_cntl *cntl)
{
job *pjob;
struct batch_reply *preply;
struct brp_status *pstatus;
svrattrl *sattrl;
int oldsid;
int bad = 0;
time_t time_now = time(NULL);
preply = &preq->rq_reply;
if (preply->brp_choice == BATCH_REPLY_CHOICE_Status)
{
pstatus = (struct brp_status *)GET_NEXT(preply->brp_un.brp_status);
while (pstatus != NULL)
{
if ((pjob = svr_find_job(pstatus->brp_objname, FALSE)) != NULL)
{
sattrl = (svrattrl *)GET_NEXT(pstatus->brp_attr);
oldsid = pjob->ji_wattr[JOB_ATR_session_id].at_val.at_long;
modify_job_attr(
pjob,
sattrl,
ATR_DFLAG_MGWR | ATR_DFLAG_SvWR,
&bad);
if (oldsid != pjob->ji_wattr[JOB_ATR_session_id].at_val.at_long)
{
/* first save since running job (or the sid has changed), */
/* must save session id */
job_save(pjob, SAVEJOB_FULL, 0);
}
#ifdef USESAVEDRESOURCES
else
{
/* save so we can recover resources used */
job_save(pjob, SAVEJOB_FULL, 0);
}
#endif /* USESAVEDRESOURCES */
pjob->ji_momstat = time_now;
unlock_ji_mutex(pjob, __func__, "1", LOGLEVEL);
}
pstatus = (struct brp_status *)GET_NEXT(pstatus->brp_stlink);
} /* END while (pstatus != NULL) */
} /* END if (preply->brp_choice == BATCH_REPLY_CHOICE_Status) */
else
{
if (preply->brp_code == PBSE_UNKJOBID)
{
/* we sent a stat request, but mom says it doesn't know anything about
the job */
if ((pjob = svr_find_job(preq->rq_ind.rq_status.rq_id, FALSE)) != NULL)
{
/* job really isn't running any more - mom doesn't know anything about it
this can happen if a diskless node reboots and the mom_priv/jobs
directory is cleared, set its state to queued so job_abt doesn't
think it is still running */
svr_setjobstate(pjob, JOB_STATE_QUEUED, JOB_SUBSTATE_ABORT, FALSE);
rel_resc(pjob);
job_abt(&pjob, "Job does not exist on node");
/* TODO, if the job is rerunnable we should set its state back to queued */
}
}
}
cntl->sc_conn = -1;
/* MUTSU - Unlock job here? */
if (cntl->sc_post)
cntl->sc_post(cntl); /* continue where we left off */
/* If sc_post has a value it is:
* req_stat_job_step2
* if so, it expects cntl to be free'd after the call
*/
free(cntl); /* a bit of a kludge but its saves an extra func */
return;
} /* END stat_update() */
void
req_modifyjob(struct batch_request *preq)
{
int add_to_am_list = 0; /* if altered during sched cycle */
int bad = 0;
int jt; /* job type */
int newstate;
int newsubstate;
resource_def *outsideselect = NULL;
job *pjob;
svrattrl *plist;
resource *presc;
resource_def *prsd;
int rc;
int running = 0;
int sendmom = 0;
char hook_msg[HOOK_MSG_SIZE];
int mod_project = 0;
pbs_sched *psched;
switch (process_hooks(preq, hook_msg, sizeof(hook_msg),
pbs_python_set_interrupt)) {
case 0: /* explicit reject */
reply_text(preq, PBSE_HOOKERROR, hook_msg);
return;
case 1: /* explicit accept */
if (recreate_request(preq) == -1) { /* error */
/* we have to reject the request, as 'preq' */
/* may have been partly modified */
strcpy(hook_msg,
"modifyjob event: rejected request");
log_event(PBSEVENT_ERROR, PBS_EVENTCLASS_HOOK,
LOG_ERR, "", hook_msg);
reply_text(preq, PBSE_HOOKERROR, hook_msg);
return;
}
break;
case 2: /* no hook script executed - go ahead and accept event*/
break;
default:
log_event(PBSEVENT_DEBUG2, PBS_EVENTCLASS_HOOK,
LOG_INFO, "", "modifyjob event: accept req by default");
}
if (pseldef == NULL) /* do one time to keep handy */
pseldef = find_resc_def(svr_resc_def, "select", svr_resc_size);
pjob = chk_job_request(preq->rq_ind.rq_modify.rq_objname, preq, &jt);
if (pjob == NULL)
return;
if ((jt == IS_ARRAY_Single) || (jt == IS_ARRAY_Range)) {
req_reject(PBSE_IVALREQ, 0, preq);
return;
}
psched = find_sched_from_sock(preq->rq_conn);
/* allow scheduler to modify job */
if (psched == NULL) {
/* provisioning job is not allowed to be modified */
if ((pjob->ji_qs.ji_state == JOB_STATE_RUNNING) &&
(pjob->ji_qs.ji_substate == JOB_SUBSTATE_PROVISION)) {
req_reject(PBSE_BADSTATE, 0, preq);
return;
}
}
/* cannot be in exiting or transit, exiting has already be checked */
if (pjob->ji_qs.ji_state == JOB_STATE_TRANSIT) {
req_reject(PBSE_BADSTATE, 0, preq);
return;
}
plist = (svrattrl *)GET_NEXT(preq->rq_ind.rq_modify.rq_attr);
if (plist == NULL) { /* nothing to do */
reply_ack(preq);
return;
}
/*
* Special checks must be made:
* if during a scheduling cycle and certain attributes are altered,
* make a note of the job to prevent it from being run now;
* if job is running, only certain attributes/resources can be
* altered.
*/
if (pjob->ji_qs.ji_state == JOB_STATE_RUNNING) {
running = 1;
}
while (plist) {
int i;
i = find_attr(job_attr_def, plist->al_name, JOB_ATR_LAST);
/*
* Is the attribute being altered one which could change
* scheduling (ATR_DFLAG_SCGALT set) and if a scheduling
* cycle is in progress, then set flag to add the job to list
* of jobs which cannot be run in this cycle.
* If the scheduler itself sends a modify job request,
* no need to delay the job until next cycle.
*/
if ((psched == NULL) && (scheduler_jobs_stat) && (job_attr_def[i].at_flags & ATR_DFLAG_SCGALT))
add_to_am_list = 1;
/* Is the attribute modifiable in RUN state ? */
if (i < 0) {
reply_badattr(PBSE_NOATTR, 1, plist, preq);
return;
}
if ((running == 1) &&
((job_attr_def[i].at_flags & ATR_DFLAG_ALTRUN) == 0)) {
reply_badattr(PBSE_MODATRRUN, 1, plist, preq);
return;
}
if (i == (int)JOB_ATR_resource) {
prsd = find_resc_def(svr_resc_def, plist->al_resc,
svr_resc_size);
if (prsd == 0) {
reply_badattr(PBSE_UNKRESC, 1, plist, preq);
return;
}
/* is the specified resource modifiable while */
/* the job is running */
if (running) {
if ((prsd->rs_flags & ATR_DFLAG_ALTRUN) == 0) {
reply_badattr(PBSE_MODATRRUN, 1, plist, preq);
return;
}
sendmom = 1;
}
/* should the resource be only in a select spec */
if (prsd->rs_flags & ATR_DFLAG_CVTSLT && !outsideselect &&
plist->al_atopl.value && plist->al_atopl.value[0]) {
/* if "-lresource" is set and has non-NULL value,
** remember as potential bad resource
** if this appears along "select".
*/
outsideselect = prsd;
}
}
if (strcmp(plist->al_name, ATTR_project) == 0) {
mod_project = 1;
} else if ((strcmp(plist->al_name, ATTR_runcount) == 0) &&
((plist->al_flags & ATR_VFLAG_HOOK) == 0) &&
(plist->al_value != NULL) &&
(plist->al_value[0] != '\0') &&
((preq->rq_perm & (ATR_DFLAG_MGWR | ATR_DFLAG_OPWR)) == 0) &&
(atol(plist->al_value) < \
pjob->ji_wattr[(int)JOB_ATR_runcount].at_val.at_long)) {
sprintf(log_buffer,
"regular user %s@%s cannot decrease '%s' attribute value from %ld to %ld",
preq->rq_user, preq->rq_host, ATTR_runcount,
pjob->ji_wattr[(int)JOB_ATR_runcount].at_val.at_long,
atol(plist->al_value));
log_event(PBSEVENT_JOB, PBS_EVENTCLASS_JOB, LOG_ERR,
pjob->ji_qs.ji_jobid, log_buffer);
req_reject(PBSE_PERM, 0, preq);
return;
}
plist = (svrattrl *)GET_NEXT(plist->al_link);
}
if (outsideselect) {
presc = find_resc_entry(&pjob->ji_wattr[(int)JOB_ATR_resource],
pseldef);
if (presc &&
((presc->rs_value.at_flags & ATR_VFLAG_DEFLT) == 0)) {
/* select is not a default, so reject qalter */
resc_in_err = strdup(outsideselect->rs_name);
req_reject(PBSE_INVALJOBRESC, 0, preq);
return;
}
}
/* modify the jobs 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) {
if (pjob->ji_clterrmsg)
reply_text(preq, rc, pjob->ji_clterrmsg);
else
reply_badattr(rc, bad, plist, preq);
return;
}
/* If certain attributes modified and if in scheduling cycle */
/* then add to list of jobs which cannot be run in this cycle */
if (add_to_am_list)
am_jobs_add(pjob); /* see req_runjob() */
/* check if project attribute was requested to be modified to */
/* be the default project value */
if (mod_project && (pjob->ji_wattr[(int)JOB_ATR_project].at_flags & \
ATR_VFLAG_SET)) {
if (strcmp(pjob->ji_wattr[(int)JOB_ATR_project].at_val.at_str,
PBS_DEFAULT_PROJECT) == 0) {
sprintf(log_buffer, msg_defproject,
ATTR_project, PBS_DEFAULT_PROJECT);
#ifdef NAS /* localmod 107 */
log_event(PBSEVENT_DEBUG4, PBS_EVENTCLASS_JOB, LOG_INFO,
pjob->ji_qs.ji_jobid, log_buffer);
#else
log_event(PBSEVENT_JOB, PBS_EVENTCLASS_JOB, LOG_INFO,
pjob->ji_qs.ji_jobid, log_buffer);
#endif /* localmod 107 */
}
}
if (pjob->ji_wattr[(int)JOB_ATR_resource].at_flags & ATR_VFLAG_MODIFY) {
presc = find_resc_entry(&pjob->ji_wattr[(int)JOB_ATR_resource],
pseldef);
if (presc && (presc->rs_value.at_flags & ATR_VFLAG_DEFLT)) {
/* changing Resource_List and select is a default */
/* clear "select" so it is rebuilt inset_resc_deflt */
pseldef->rs_free(&presc->rs_value);
}
}
/* Reset any defaults resource limit which might have been unset */
if ((rc = set_resc_deflt((void *)pjob, JOB_OBJECT, NULL)) != 0) {
req_reject(rc, 0, preq);
return;
}
/* 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);
(void)svr_setjobstate(pjob, newstate, newsubstate);
} else {
(void)job_save(pjob, SAVEJOB_FULL);
}
(void)sprintf(log_buffer, msg_manager, msg_jobmod,
preq->rq_user, preq->rq_host);
log_event(PBSEVENT_JOB, PBS_EVENTCLASS_JOB, LOG_INFO,
pjob->ji_qs.ji_jobid, log_buffer);
/* if a resource limit changed for a running job, send to MOM */
if (sendmom) {
rc = relay_to_mom(pjob, preq, post_modify_req);
if (rc)
req_reject(rc, 0, preq); /* unable to get to MOM */
return;
}
reply_ack(preq);
}
void update_job_data(
struct pbsnode *np, /* I */
const char *jobstring_in) /* I (changed attributes sent by mom) */
{
char *jobdata;
char *jobdata_ptr;
char *jobidstr;
char *attr_name;
char *attr_value;
char log_buf[LOCAL_LOG_BUF_SIZE];
job *pjob = NULL;
int on_node = FALSE;
if ((jobstring_in == NULL) || (!isdigit(*jobstring_in)))
{
/* NO-OP */
return;
}
/* FORMAT <JOBID>:<atrtributename=value>,<atrtributename=value>... */
jobdata = strdup(jobstring_in);
jobdata_ptr = jobdata;
jobidstr = threadsafe_tokenizer(&jobdata_ptr, ":");
if ((jobidstr != NULL) && isdigit(*jobidstr))
{
if (strstr(jobidstr, server_name) != NULL)
{
on_node = is_job_on_node(np, job_mapper.get_id(jobidstr));
pjob = svr_find_job(jobidstr, TRUE);
if (pjob != NULL)
{
int bad;
svrattrl tA;
mutex_mgr job_mutex(pjob->ji_mutex, true);
/* job exists, so get the attributes and update them */
attr_name = threadsafe_tokenizer(&jobdata_ptr, "=");
while (attr_name != NULL)
{
attr_value = threadsafe_tokenizer(&jobdata_ptr, ",");
if (LOGLEVEL >= 9)
{
sprintf(log_buf, "Mom sent changed attribute %s value %s for job %s",
attr_name,
attr_value,
pjob->ji_qs.ji_jobid);
log_event(PBSEVENT_JOB,PBS_EVENTCLASS_JOB,pjob->ji_qs.ji_jobid,log_buf);
}
memset(&tA, 0, sizeof(tA));
tA.al_name = attr_name;
tA.al_resc = (char *)"";
tA.al_value = attr_value;
tA.al_op = SET;
modify_job_attr(
pjob,
&tA, /* I: ATTR_sched_hint - svrattrl */
ATR_DFLAG_MGWR | ATR_DFLAG_SvWR,
&bad);
attr_name = threadsafe_tokenizer(&jobdata_ptr, "=");
}
}
if (on_node == FALSE)
{
/* job is reported by mom but server has no record of job */
sprintf(log_buf, "stray job %s reported on %s", jobidstr, np->nd_name);
log_err(-1, __func__, log_buf);
}
}
}
free(jobdata);
} /* END update_job_data() */
static void stat_update(
struct work_task *pwt)
{
struct stat_cntl *cntl;
job *pjob;
struct batch_request *preq;
struct batch_reply *preply;
struct brp_status *pstatus;
svrattrl *sattrl;
int oldsid;
preq = pwt->wt_parm1;
preply = &preq->rq_reply;
cntl = preq->rq_extra;
if (preply->brp_choice == BATCH_REPLY_CHOICE_Status)
{
pstatus = (struct brp_status *)GET_NEXT(preply->brp_un.brp_status);
while (pstatus != NULL)
{
if ((pjob = find_job(pstatus->brp_objname)))
{
sattrl = (svrattrl *)GET_NEXT(pstatus->brp_attr);
oldsid = pjob->ji_wattr[(int)JOB_ATR_session_id].at_val.at_long;
modify_job_attr(
pjob,
sattrl,
ATR_DFLAG_MGWR | ATR_DFLAG_SvWR,
&bad);
if (oldsid != pjob->ji_wattr[(int)JOB_ATR_session_id].at_val.at_long)
{
/* first save since running job (or the sid has changed), */
/* must save session id */
job_save(pjob, SAVEJOB_FULL);
svr_mailowner(pjob, MAIL_BEGIN, MAIL_NORMAL, NULL);
}
#ifdef USESAVEDRESOURCES
else
{
/* save so we can recover resources used */
job_save(pjob, SAVEJOB_FULL);
}
#endif /* USESAVEDRESOURCES */
pjob->ji_momstat = time_now;
}
pstatus = (struct brp_status *)GET_NEXT(pstatus->brp_stlink);
} /* END while (pstatus != NULL) */
} /* END if (preply->brp_choice == BATCH_REPLY_CHOICE_Status) */
else
{
if (preply->brp_code == PBSE_UNKJOBID)
{
/* we sent a stat request, but mom says it doesn't know anything about
the job */
if ((pjob = find_job(preq->rq_ind.rq_status.rq_id)))
{
/* job really isn't running any more - mom doesn't know anything about it
this can happen if a diskless node reboots and the mom_priv/jobs
directory is cleared, set its state to queued so job_abt doesn't
think it is still running */
svr_setjobstate(pjob, JOB_STATE_QUEUED, JOB_SUBSTATE_ABORT);
rel_resc(pjob);
job_abt(&pjob, "Job does not exist on node");
/* TODO, if the job is rerunnable we should set its state back to queued */
}
}
}
release_req(pwt);
cntl->sc_conn = -1;
if (cntl->sc_post)
cntl->sc_post(cntl); /* continue where we left off */
else
free(cntl); /* a bit of a kludge but its saves an extra func */
return;
} /* END stat_update() */
