void req_commit(
struct batch_request *preq) /* I */
{
job *pj;
pj = locate_new_job(preq->rq_conn, preq->rq_ind.rq_commit);
if (LOGLEVEL >= 6)
{
log_record(
PBSEVENT_JOB,
PBS_EVENTCLASS_JOB,
(pj != NULL) ? pj->ji_qs.ji_jobid : "NULL",
"committing job");
}
if (pj == NULL)
{
req_reject(PBSE_UNKJOBID, 0, preq, NULL, NULL);
return;
}
if (pj->ji_qs.ji_substate != JOB_SUBSTATE_TRANSICM)
{
log_err(errno, "req_commit", "cannot commit job in unexpected state");
req_reject(PBSE_IVALREQ, 0, preq, NULL, NULL);
return;
}
/* move job from new job list to "all" job list, set to running state */
delete_link(&pj->ji_alljobs);
append_link(&svr_alljobs, &pj->ji_alljobs, pj);
/*
** Set JOB_SVFLG_HERE to indicate that this is Mother Superior.
*/
pj->ji_qs.ji_svrflags |= JOB_SVFLG_HERE;
pj->ji_qs.ji_state = JOB_STATE_RUNNING;
pj->ji_qs.ji_substate = JOB_SUBSTATE_PRERUN;
pj->ji_qs.ji_un_type = JOB_UNION_TYPE_MOM;
pj->ji_qs.ji_un.ji_momt.ji_svraddr = get_connectaddr(preq->rq_conn);
pj->ji_qs.ji_un.ji_momt.ji_exitstat = 0;
#ifdef HAVE_GLITE_LB
svr_logjobstate(pj, JOB_STATE_RUNNING, JOB_SUBSTATE_PRERUN, preq);
#endif
/* For MOM - start up the job (blocks) */
if (LOGLEVEL >= 6)
{
log_record(
PBSEVENT_JOB,
PBS_EVENTCLASS_JOB,
(pj != NULL) ? pj->ji_qs.ji_jobid : "NULL",
"starting job execution");
}
// force poll of other jobs to prevent memory crashes
last_poll_time = 0;
start_exec(pj);
if (LOGLEVEL >= 6)
{
log_record(
PBSEVENT_JOB,
PBS_EVENTCLASS_JOB,
(pj != NULL) ? pj->ji_qs.ji_jobid : "NULL",
"job execution started");
}
/* if start request fails, reply with failure string */
if (pj->ji_qs.ji_substate == JOB_SUBSTATE_EXITING)
{
char tmpLine[1024];
if ((pj->ji_hosts != NULL) &&
(pj->ji_nodekill >= 0) &&
(pj->ji_hosts[pj->ji_nodekill].hn_host != NULL))
{
sprintf(tmpLine, "start failed on node %s",
pj->ji_hosts[pj->ji_nodekill].hn_host);
}
else
{
sprintf(tmpLine, "start failed on unknown node");
}
if (LOGLEVEL >= 6)
{
log_record(
PBSEVENT_JOB,
PBS_EVENTCLASS_JOB,
(pj != NULL) ? pj->ji_qs.ji_jobid : "NULL",
tmpLine);
}
reply_text(preq, 0, tmpLine);
}
else
{
reply_jobid(preq, pj->ji_qs.ji_jobid, BATCH_REPLY_CHOICE_Commit);
}
if (is_cloud_job(pj) && pj->ji_numnodes == 1)
{
if (cloud_exec(pj, 1) == 0)
cloud_set_running(pj);
if (mom_do_poll(pj) != 0)
append_link(&mom_polljobs, &pj->ji_jobque, pj);
}
job_save(pj, SAVEJOB_FULL);
/* NOTE: we used to flag JOB_ATR_errpath, JOB_ATR_outpath,
* JOB_ATR_session_id, and JOB_ATR_altid as modified at this point to make sure
* pbs_server got these attr values. This worked fine before TORQUE modified
* job launched into an async process. At 2.0.0p6, a new attribute "SEND" flag
* was added to handle this process. */
return;
} /* END req_commit() */
/**
* @brief
* que_purge - purge queue from system
* The queue is dequeued, the queue file is unlinked.
* If the queue contains any jobs, the purge is not allowed.
* Eventually the queue is deleted from the database
*
* @param[in] pque - The pointer to the queue to purge
*
* @return error code
* @retval 0 - queue purged or queue not valid
* @retval PBSE_OBJBUSY - queue deletion not allowed
*/
int
que_purge(pbs_queue *pque)
{
pbs_db_obj_info_t obj;
pbs_db_que_info_t dbque;
pbs_db_conn_t *conn = (pbs_db_conn_t *) svr_db_conn;
/*
* If the queue (pque) is not valid, then nothing to
* do, just return 0.
*/
if (pque == NULL)
return (0);
/* are there any jobs still in the queue */
if (pque->qu_numjobs != 0) {
/*
* If the queue still has job(s), check if the SERVER
* is configured for history info and all the jobs in
* queue are history jobs. If yes, then allow queue
* deletion otherwise return PBSE_OBJBUSY.
*/
if (svr_history_enable) { /* SVR histconf chk */
job *pjob = (job *)0;
job *nxpjob = (job *)0;
pjob = (job *)GET_NEXT(pque->qu_jobs);
while (pjob) {
/*
* If it is not a history job (MOVED/FINISHED), then
* return with PBSE_OBJBUSY error.
*/
if ((pjob->ji_qs.ji_state != JOB_STATE_MOVED) &&
(pjob->ji_qs.ji_state != JOB_STATE_FINISHED))
return (PBSE_OBJBUSY);
pjob = (job *)GET_NEXT(pjob->ji_jobque);
}
/*
* All are history jobs, unlink all of them from queue.
* Update the number of jobs in the queue and their state
* count as the queue is going to be purged. No job(s)
* should point to the queue to be purged, make the queue
* header pointer of job(pjob->ji_qhdr) to NULL.
*/
pjob = (job *)GET_NEXT(pque->qu_jobs);
while (pjob) {
nxpjob = (job *)GET_NEXT(pjob->ji_jobque);
delete_link(&pjob->ji_jobque);
--pque->qu_numjobs;
--pque->qu_njstate[pjob->ji_qs.ji_state];
pjob->ji_qhdr = (pbs_queue *)0;
pjob = nxpjob;
}
} else {
return (PBSE_OBJBUSY);
}
}
/* delete queue from database */
strcpy(dbque.qu_name, pque->qu_qs.qu_name);
obj.pbs_db_obj_type = PBS_DB_QUEUE;
obj.pbs_db_un.pbs_db_que = &dbque;
if (pbs_db_delete_obj(conn, &obj) != 0) {
(void)sprintf(log_buffer,
"delete of que %s from datastore failed",
pque->qu_qs.qu_name);
log_err(errno, "queue_purge", log_buffer);
}
que_free(pque);
return (0);
}
void req_quejob(
struct batch_request *preq) /* ptr to the decoded request */
{
char *id = "req_quejob";
char basename[PBS_JOBBASE + 1];
int created_here = 0;
int index;
char *jid;
attribute_def *pdef;
job *pj;
svrattrl *psatl;
int rc;
int sock = preq->rq_conn;
int IsCheckpoint = 0;
/* set basic (user) level access permission */
resc_access_perm = ATR_DFLAG_USWR | ATR_DFLAG_Creat;
if (PBSNodeCheckProlog)
{
check_state(1);
mom_server_all_update_stat();
if (internal_state & INUSE_DOWN)
{
req_reject(PBSE_MOMREJECT,0,preq,NULL,NULL);
return;
}
}
if (preq->rq_fromsvr)
{
/* from another server - accept the extra attributes */
resc_access_perm |= ATR_DFLAG_MGWR | ATR_DFLAG_SvWR | ATR_DFLAG_MOM;
jid = preq->rq_ind.rq_queuejob.rq_jid;
}
else
{
/* request must be from server */
log_err(errno, id, "request not from server");
req_reject(PBSE_IVALREQ, 0, preq, NULL, "request not received from server");
return;
}
/* does job already exist, check both old and new jobs */
if ((pj = find_job(jid)) == NULL)
{
pj = (job *)GET_NEXT(svr_newjobs);
while (pj != NULL)
{
if (!strcmp(pj->ji_qs.ji_jobid, jid))
break;
pj = (job *)GET_NEXT(pj->ji_alljobs);
}
}
/*
* New job ...
*
* for MOM - rather than make up a hashname, we use the name sent
* to us by the server as an attribute.
*/
psatl = (svrattrl *)GET_NEXT(preq->rq_ind.rq_queuejob.rq_attr);
while (psatl != NULL)
{
if (!strcmp(psatl->al_name,ATTR_hashname))
{
strcpy(basename,psatl->al_value);
break;
}
psatl = (svrattrl *)GET_NEXT(psatl->al_link);
}
if (pj != NULL)
{
/* newly queued job already exists */
if (pj->ji_qs.ji_substate == JOB_SUBSTATE_RUNNING)
{
/* FAILURE - job exists and is running */
log_err(errno,id,"cannot queue new job, job exists and is running");
req_reject(PBSE_JOBEXIST,0,preq,NULL,"job is running");
return;
}
/* if checkpointed, then keep old and skip rest of process */
if (pj->ji_qs.ji_svrflags & JOB_SVFLG_CHECKPOINT_FILE)
{
IsCheckpoint = 1;
} /* END if (pj->ji_qs.ji_svrflags & JOB_SVFLG_CHECKPOINT_FILE) */
else
{
/* unlink job from svr_alljobs since it will be placed on newjobs */
delete_link(&pj->ji_alljobs);
}
} /* END if (pj != NULL) */
else
{
/* if not already here, allocate job struct */
if ((pj = job_alloc()) == NULL)
{
/* FAILURE */
req_reject(PBSE_SYSTEM, 0, preq, NULL, "cannot allocate new job structure");
return;
}
} /* END else (pj != NULL) */
if (IsCheckpoint == 0)
{
strcpy(pj->ji_qs.ji_jobid,jid);
strcpy(pj->ji_qs.ji_fileprefix,basename);
pj->ji_modified = 1;
pj->ji_qs.ji_svrflags = created_here;
pj->ji_qs.ji_un_type = JOB_UNION_TYPE_NEW;
}
/* decode attributes from request into job structure */
psatl = (svrattrl *)GET_NEXT(preq->rq_ind.rq_queuejob.rq_attr);
while (psatl != NULL)
{
if (IsCheckpoint == 1)
{
if (strcmp(psatl->al_name,ATTR_checkpoint_name) &&
strcmp(psatl->al_name,ATTR_v))
{
psatl = (svrattrl *)GET_NEXT(psatl->al_link);
continue;
}
}
/* identify the attribute by name */
index = find_attr(job_attr_def,psatl->al_name,JOB_ATR_LAST);
if (index < 0)
{
/* FAILURE */
#if 0
/* old implementation, refuse jobs with unknown attributes */
/* didn`t recognize the name */
job_purge(pj); /* CRI - 12/20/2004 */
reply_badattr(PBSE_NOATTR,1,psatl,preq);
return;
#endif
/* new implementation, ignore unknown attributes */
sprintf(log_buffer, "Unknown attribute \"%s\" received in req_quejob request", psatl->al_name);
LOG_EVENT(PBSEVENT_JOB, PBS_EVENTCLASS_JOB, pj->ji_qs.ji_jobid, log_buffer);
psatl = (svrattrl *)GET_NEXT(psatl->al_link);
continue;
}
pdef = &job_attr_def[index];
/* Is attribute not writeable by manager or by a server? */
if ((pdef->at_flags & resc_access_perm) == 0)
{
/* FAILURE */
job_purge(pj);
reply_badattr(PBSE_ATTRRO,1,psatl,preq);
return;
}
/* decode attribute */
if (!strcmp(psatl->al_name,ATTR_v))
{
rc = decode_arst_merge(
&pj->ji_wattr[index],
psatl->al_name,
psatl->al_resc,
psatl->al_value);
}
else
{
rc = pdef->at_decode(
&pj->ji_wattr[index],
psatl->al_name,
psatl->al_resc,
psatl->al_value);
}
if (rc != 0)
{
/* FAILURE */
/* all errors are fatal for MOM */
if (rc != PBSE_UNKRESC)
{
job_purge(pj);
reply_badattr(rc,1,psatl,preq);
return;
}
psatl = (svrattrl *)GET_NEXT(psatl->al_link);
continue;
}
if (psatl->al_op == DFLT)
{
if (psatl->al_resc)
{
#if 0 /* don't mark resources as default on nodes, we need all resources stored in the jobfile */
resource *presc;
resource_def *prdef;
prdef = find_resc_def(svr_resc_def,psatl->al_resc,svr_resc_size);
if (prdef == NULL)
{
job_purge(pj);
reply_badattr(rc,1,psatl, preq);
return;
}
presc = find_resc_entry(&pj->ji_wattr[index],prdef);
if (presc != NULL)
presc->rs_value.at_flags |= ATR_VFLAG_DEFLT;
#endif
}
else
{
pj->ji_wattr[index].at_flags |= ATR_VFLAG_DEFLT;
}
} /* END if (psatl->al_op == DFLT) */
psatl = (svrattrl *)GET_NEXT(psatl->al_link);
} /* END while (psatl != NULL) */
if (IsCheckpoint == 1)
{
pj->ji_qs.ji_substate = JOB_SUBSTATE_TRANSIN;
if (reply_jobid(preq,pj->ji_qs.ji_jobid,BATCH_REPLY_CHOICE_Queue) == 0)
{
delete_link(&pj->ji_alljobs);
append_link(&svr_newjobs,&pj->ji_alljobs,pj);
pj->ji_qs.ji_un_type = JOB_UNION_TYPE_NEW;
pj->ji_qs.ji_un.ji_newt.ji_fromsock = sock;
pj->ji_qs.ji_un.ji_newt.ji_fromaddr = get_connectaddr(sock);
pj->ji_qs.ji_un.ji_newt.ji_scriptsz = 0;
/* Per Eric R., req_mvjobfile was giving error in open_std_file,
showed up as fishy error message */
if (pj->ji_grpcache != NULL)
{
free(pj->ji_grpcache);
pj->ji_grpcache = NULL;
}
}
else
{
close_conn(sock);
}
/* SUCCESS */
return;
}
/* set remaining job structure elements */
pj->ji_qs.ji_state = JOB_STATE_TRANSIT;
pj->ji_qs.ji_substate = JOB_SUBSTATE_TRANSIN;
pj->ji_wattr[(int)JOB_ATR_mtime].at_val.at_long = (long)time_now;
pj->ji_wattr[(int)JOB_ATR_mtime].at_flags |= ATR_VFLAG_SET;
pj->ji_qs.ji_un_type = JOB_UNION_TYPE_NEW;
pj->ji_qs.ji_un.ji_newt.ji_fromsock = sock;
pj->ji_qs.ji_un.ji_newt.ji_fromaddr = get_connectaddr(sock);
pj->ji_qs.ji_un.ji_newt.ji_scriptsz = 0;
/* acknowledge the request with the job id */
if (reply_jobid(preq, pj->ji_qs.ji_jobid, BATCH_REPLY_CHOICE_Queue) != 0)
{
/* reply failed, purge the job and close the connection */
close_conn(sock);
job_purge(pj);
return;
}
/* link job into server's new jobs list request */
append_link(&svr_newjobs, &pj->ji_alljobs, pj);
return;
} /* END req_quejob() */
int acct_job(
job *pjob, /* I */
dynamic_string *ds) /* O */
{
int rc;
long cray_enabled = FALSE;
int resc_access_perm = READ_ONLY;
char local_buf[MAXLINE*4];
pbs_queue *pque;
tlist_head attrlist;
svrattrl *pal;
if (pjob == NULL)
{
return(PBSE_NONE);
}
CLEAR_HEAD(attrlist);
/* user */
/* acct_job is only called from account_jobstr and account_jobend. BufSize should be
PBS_ACCT_MAX_RCD + 1 in size. */
sprintf(local_buf, "user=%s ",
pjob->ji_wattr[JOB_ATR_euser].at_val.at_str);
if ((rc = append_dynamic_string(ds, local_buf)) != PBSE_NONE)
return(rc);
/* group */
sprintf(local_buf, "group=%s ",
pjob->ji_wattr[JOB_ATR_egroup].at_val.at_str);
if ((rc = append_dynamic_string(ds, local_buf)) != PBSE_NONE)
return(rc);
/* account */
if (pjob->ji_wattr[JOB_ATR_account].at_flags & ATR_VFLAG_SET)
{
sprintf(local_buf, "account=%s ",
pjob->ji_wattr[JOB_ATR_account].at_val.at_str);
if ((rc = append_dynamic_string(ds, local_buf)) != PBSE_NONE)
return(rc);
}
/* job name */
sprintf(local_buf, "jobname=%s ",
pjob->ji_wattr[JOB_ATR_jobname].at_val.at_str);
if ((rc = append_dynamic_string(ds, local_buf)) != PBSE_NONE)
return(rc);
if ((pque = get_jobs_queue(&pjob)) != NULL)
{
/* queue name */
sprintf(local_buf, "queue=%s ", pque->qu_qs.qu_name);
unlock_queue(pque, __func__, NULL, LOGLEVEL);
if ((rc = append_dynamic_string(ds, local_buf)) != PBSE_NONE)
return(rc);
}
else if (pjob == NULL)
{
log_err(PBSE_JOBNOTFOUND, __func__, "Job lost while acquiring queue 1");
return(PBSE_JOBNOTFOUND);
}
/* create time */
sprintf(local_buf, "ctime=%ld ",
pjob->ji_wattr[JOB_ATR_ctime].at_val.at_long);
if ((rc = append_dynamic_string(ds, local_buf)) != PBSE_NONE)
return(rc);
/* queued time */
sprintf(local_buf, "qtime=%ld ",
pjob->ji_wattr[JOB_ATR_qtime].at_val.at_long);
if ((rc = append_dynamic_string(ds, local_buf)) != PBSE_NONE)
return(rc);
/* eligible time, how long ready to run */
sprintf(local_buf, "etime=%ld ",
pjob->ji_wattr[JOB_ATR_etime].at_val.at_long);
if ((rc = append_dynamic_string(ds, local_buf)) != PBSE_NONE)
return(rc);
/* execution start time */
sprintf(local_buf, "start=%ld ",
(long)pjob->ji_qs.ji_stime);
if ((rc = append_dynamic_string(ds, local_buf)) != PBSE_NONE)
return(rc);
/* user */
sprintf(local_buf, "owner=%s ",
pjob->ji_wattr[JOB_ATR_job_owner].at_val.at_str);
if ((rc = append_dynamic_string(ds, local_buf)) != PBSE_NONE)
return(rc);
/* For large clusters strings can get pretty long. We need to see if there
is a need to allocate a bigger buffer */
/* execution host name */
append_dynamic_string(ds, "exec_host=");
append_dynamic_string(ds, pjob->ji_wattr[JOB_ATR_exec_host].at_val.at_str);
if ((rc = append_dynamic_string(ds, " ")) != PBSE_NONE)
return(rc);
get_svr_attr_l(SRV_ATR_CrayEnabled, &cray_enabled);
if ((cray_enabled == TRUE) &&
(pjob->ji_wattr[JOB_ATR_login_node_id].at_flags & ATR_VFLAG_SET))
{
append_dynamic_string(ds, "login_node=");
append_dynamic_string(ds, pjob->ji_wattr[JOB_ATR_login_node_id].at_val.at_str);
if ((rc = append_dynamic_string(ds, " ")) != PBSE_NONE)
return(rc);
}
/* now encode the job's resource_list pbs_attribute */
job_attr_def[JOB_ATR_resource].at_encode(
&pjob->ji_wattr[JOB_ATR_resource],
&attrlist,
job_attr_def[JOB_ATR_resource].at_name,
NULL,
ATR_ENCODE_CLIENT,
resc_access_perm);
while ((pal = GET_NEXT(attrlist)) != NULL)
{
/* exec_host can use a lot of buffer space. Use a dynamic string */
append_dynamic_string(ds, pal->al_name);
if (pal->al_resc != NULL)
{
append_dynamic_string(ds, ".");
append_dynamic_string(ds, pal->al_resc);
}
append_dynamic_string(ds, "=");
append_dynamic_string(ds, pal->al_value);
if ((rc = append_dynamic_string(ds, " ")) != PBSE_NONE)
return(rc);
delete_link(&pal->al_link);
free(pal);
} /* END while (pal != NULL) */
#ifdef ATTR_X_ACCT
/* x attributes */
if (pjob->ji_wattr[JOB_SITE_ATR_x].at_flags & ATR_VFLAG_SET)
{
sprintf(local_buf, "x=%s ",
pjob->ji_wattr[JOB_SITE_ATR_x].at_val.at_str);
if ((rc = append_dynamic_string(ds, local_buf)) != PBSE_NONE)
return(rc);
}
#endif
/* SUCCESS */
return(PBSE_NONE);
} /* END acct_job() */
void req_commit(
struct batch_request *preq) /* I */
{
unsigned int momport = 0;
int rc;
job *pj = locate_new_job(preq->rq_conn, preq->rq_ind.rq_commit);
if (LOGLEVEL >= 6)
{
log_record(
PBSEVENT_JOB,
PBS_EVENTCLASS_JOB,
(pj != NULL) ? pj->ji_qs.ji_jobid : "NULL",
"committing job");
}
if (pj == NULL)
{
req_reject(PBSE_UNKJOBID, 0, preq, NULL, NULL);
return;
}
if (pj->ji_qs.ji_substate != JOB_SUBSTATE_TRANSICM)
{
log_err(errno, "req_commit", (char *)"cannot commit job in unexpected state");
req_reject(PBSE_IVALREQ, 0, preq, NULL, NULL);
return;
}
/* move job from new job list to "all" job list, set to running state */
delete_link(&pj->ji_alljobs);
append_link(&svr_alljobs, &pj->ji_alljobs, pj);
/*
** Set JOB_SVFLG_HERE to indicate that this is Mother Superior.
*/
pj->ji_qs.ji_svrflags |= JOB_SVFLG_HERE;
pj->ji_qs.ji_state = JOB_STATE_RUNNING;
pj->ji_qs.ji_substate = JOB_SUBSTATE_PRERUN;
pj->ji_qs.ji_un_type = JOB_UNION_TYPE_MOM;
pj->ji_qs.ji_un.ji_momt.ji_svraddr = get_connectaddr(preq->rq_conn,FALSE);
pj->ji_qs.ji_un.ji_momt.ji_exitstat = 0;
/* For MOM - start up the job (blocks) */
if (LOGLEVEL >= 6)
{
log_record(
PBSEVENT_JOB,
PBS_EVENTCLASS_JOB,
(pj != NULL) ? pj->ji_qs.ji_jobid : "NULL",
"req_commit:starting job execution");
}
rc = start_exec(pj);
if (LOGLEVEL >= 6)
{
log_record(
PBSEVENT_JOB,
PBS_EVENTCLASS_JOB,
pj->ji_qs.ji_jobid,
"req_commit:job execution started");
}
/* if start request fails, reply with failure string */
if (pj->ji_qs.ji_substate == JOB_SUBSTATE_EXITING)
{
char tmpLine[1024];
if ((pj->ji_hosts != NULL) &&
(pj->ji_nodekill >= 0) &&
(pj->ji_hosts[pj->ji_nodekill].hn_host != NULL))
{
sprintf(tmpLine, "start failed on node %s",
pj->ji_hosts[pj->ji_nodekill].hn_host);
}
else
{
sprintf(tmpLine, "start failed on unknown node");
}
if (LOGLEVEL >= 6)
{
log_record(
PBSEVENT_JOB,
PBS_EVENTCLASS_JOB,
(pj != NULL) ? pj->ji_qs.ji_jobid : "NULL",
tmpLine);
}
reply_text(preq, rc, tmpLine);
}
else
{
reply_sid(preq, pj->ji_wattr[JOB_ATR_session_id].at_val.at_long,BATCH_REPLY_CHOICE_Text);
}
if (multi_mom)
{
momport = pbs_rm_port;
}
job_save(pj, SAVEJOB_FULL, momport);
#ifdef NVIDIA_GPUS
/*
* Does this job have a gpuid assigned?
* if so, then update gpu status
*/
if ((use_nvidia_gpu) &&
((pj->ji_wattr[JOB_ATR_exec_gpus].at_flags & ATR_VFLAG_SET) != 0) &&
(pj->ji_wattr[JOB_ATR_exec_gpus].at_val.at_str != NULL))
{
send_update_soon();
}
#endif /* NVIDIA_GPUS */
/* NOTE: we used to flag JOB_ATR_errpath, JOB_ATR_outpath,
* JOB_ATR_session_id, and JOB_ATR_altid as modified at this point to make sure
* pbs_server got these attr values. This worked fine before TORQUE modified
* job launched into an async process. At 2.0.0p6, a new pbs_attribute "SEND" flag
* was added to handle this process. */
return;
} /* END req_commit() */
void free_br(
struct batch_request *preq)
{
delete_link(&preq->rq_link);
reply_free(&preq->rq_reply);
if (preq->rq_extend)
free(preq->rq_extend);
switch (preq->rq_type)
{
case PBS_BATCH_QueueJob:
free_attrlist(&preq->rq_ind.rq_queuejob.rq_attr);
break;
case PBS_BATCH_JobCred:
if (preq->rq_ind.rq_jobcred.rq_data)
free(preq->rq_ind.rq_jobcred.rq_data);
break;
case PBS_BATCH_MvJobFile:
case PBS_BATCH_jobscript:
if (preq->rq_ind.rq_jobfile.rq_data)
free(preq->rq_ind.rq_jobfile.rq_data);
break;
case PBS_BATCH_HoldJob:
freebr_manage(&preq->rq_ind.rq_hold.rq_orig);
break;
case PBS_BATCH_CheckpointJob:
freebr_manage(&preq->rq_ind.rq_manager);
break;
case PBS_BATCH_MessJob:
if (preq->rq_ind.rq_message.rq_text)
free(preq->rq_ind.rq_message.rq_text);
break;
case PBS_BATCH_ModifyJob:
case PBS_BATCH_AsyModifyJob:
freebr_manage(&preq->rq_ind.rq_modify);
break;
case PBS_BATCH_StatusJob:
case PBS_BATCH_StatusQue:
case PBS_BATCH_StatusNode:
case PBS_BATCH_StatusSvr:
/* DIAGTODO: handle PBS_BATCH_StatusDiag */
free_attrlist(&preq->rq_ind.rq_status.rq_attr);
break;
case PBS_BATCH_JobObit:
free_attrlist(&preq->rq_ind.rq_jobobit.rq_attr);
break;
case PBS_BATCH_CopyFiles:
case PBS_BATCH_DelFiles:
freebr_cpyfile(&preq->rq_ind.rq_cpyfile);
break;
default:
/* NO-OP */
break;
} /* END switch (preq->rq_type) */
free(preq);
return;
} /* END free_br() */
int save_attr(
struct attribute_def *padef, /* pbs_attribute definition array */
pbs_attribute *pattr, /* ptr to pbs_attribute value array */
int numattr, /* number of attributes in array */
int fds,
char *buf_ptr, /* M */
size_t *space_remaining, /* O */
size_t buf_size) /* I */
{
svrattrl dummy;
int errct = 0;
tlist_head lhead;
int i;
int resc_access_perm = ATR_DFLAG_ACCESS;
svrattrl *pal;
int rc;
/* encode each pbs_attribute which has a value (not non-set) */
CLEAR_HEAD(lhead);
for (i = 0;i < numattr;i++)
{
if ((padef + i)->at_type != ATR_TYPE_ACL)
{
/* NOTE: access lists are not saved this way */
rc = (padef + i)->at_encode(
pattr + i,
&lhead,
(padef + i)->at_name,
NULL,
ATR_ENCODE_SAVE,
resc_access_perm);
if (rc < 0)
errct++;
(pattr + i)->at_flags &= ~ATR_VFLAG_MODIFY;
/* now that it has been encoded, block and save it */
while ((pal = (svrattrl *)GET_NEXT(lhead)) != NULL)
{
if (save_struct((char *)pal, pal->al_tsize, fds, buf_ptr, space_remaining, buf_size) < 0)
errct++;
delete_link(&pal->al_link);
free(pal);
}
}
} /* END for (i) */
/* indicate last of attributes by writing dummy entry */
memset(&dummy, 0, sizeof(dummy));
dummy.al_tsize = ENDATTRIBUTES;
if (save_struct((char *)&dummy, sizeof(dummy), fds, buf_ptr, space_remaining, buf_size) < 0)
errct++;
if (errct != 0)
{
return(-1);
}
/* SUCCESS */
return(0);
} /* END save_attr() */
/* delete a job array struct from memory and disk. This is used when the number
* of jobs that belong to the array becomes zero.
* returns zero if there are no errors, non-zero otherwise
*/
int array_delete(
job_array *pa)
{
int i;
char path[MAXPATHLEN + 1];
char log_buf[LOCAL_LOG_BUF_SIZE];
array_request_node *rn;
struct array_depend *pdep;
struct array_depend_job *pdj;
/* first thing to do is take this out of the servers list of all arrays */
remove_array(pa);
/* unlock the mutex and free it */
unlock_ai_mutex(pa, __func__, "1", LOGLEVEL);
free(pa->ai_mutex);
/* delete the on disk copy of the struct */
snprintf(path, sizeof(path), "%s%s%s",
path_arrays, pa->ai_qs.fileprefix, ARRAY_FILE_SUFFIX);
if (unlink(path))
{
sprintf(log_buf, "unable to delete %s", path);
log_err(errno, "array_delete", log_buf);
}
/* clear array request linked list */
for (rn = (array_request_node *)GET_NEXT(pa->request_tokens);
rn != NULL;
rn = (array_request_node *)GET_NEXT(pa->request_tokens))
{
delete_link(&rn->request_tokens_link);
free(rn);
}
/* free the memory for the job pointers */
for (i = 0; i < pa->ai_qs.array_size; i++)
{
if (pa->job_ids[i] != NULL)
free(pa->job_ids[i]);
}
free(pa->job_ids);
/* free the dependencies, if any */
for (pdep = (struct array_depend *)GET_NEXT(pa->ai_qs.deps);
pdep != NULL;
pdep = (struct array_depend *)GET_NEXT(pa->ai_qs.deps))
{
delete_link(&pdep->dp_link);
for (pdj = (struct array_depend_job *)GET_NEXT(pdep->dp_jobs);
pdj != NULL;
pdj = (struct array_depend_job *)GET_NEXT(pdep->dp_jobs))
{
delete_link(&pdj->dc_link);
free(pdj);
}
free(pdep);
}
/* purge the "template" job,
this also deletes the shared script file for the array*/
if (pa->ai_qs.parent_id[0] != '\0')
{
job *pjob;
if ((pjob = svr_find_job(pa->ai_qs.parent_id, FALSE)) != NULL)
svr_job_purge(pjob);
}
/* free the memory allocated for the struct */
free(pa);
return(PBSE_NONE);
} /* END array_delete() */
/* array_recov reads in an array struct saved to disk and inserts it into
the servers list of arrays */
int array_recov(
char *path,
job_array **new_pa)
{
job_array *pa;
array_request_node *rn;
char log_buf[LOCAL_LOG_BUF_SIZE];
int fd;
int old_version;
int num_tokens;
int i;
int len;
int rc;
*new_pa = NULL;
old_version = ARRAY_QS_STRUCT_VERSION;
/* allocate the storage for the struct */
pa = (job_array*)calloc(1,sizeof(job_array));
if (pa == NULL)
{
return(PBSE_SYSTEM);
}
/* initialize the linked list nodes */
CLEAR_HEAD(pa->request_tokens);
fd = open(path, O_RDONLY, 0);
if(fd < 0)
{
free(pa);
return(PBSE_SYSTEM);
}
if (array_259_upgrade)
{
rc = read_and_convert_259_array(fd, pa, path);
if (rc != PBSE_NONE)
{
free(pa);
close(fd);
return(rc);
}
}
else
{
/* read the file into the struct previously allocated.
*/
len = read_ac_socket(fd, &(pa->ai_qs), sizeof(pa->ai_qs));
if ((len < 0) || ((len < (int)sizeof(pa->ai_qs)) && (pa->ai_qs.struct_version == ARRAY_QS_STRUCT_VERSION)))
{
sprintf(log_buf, "error reading %s", path);
log_err(errno, __func__, log_buf);
free(pa);
close(fd);
return(PBSE_SYSTEM);
}
if (pa->ai_qs.struct_version != ARRAY_QS_STRUCT_VERSION)
{
rc = array_upgrade(pa, fd, pa->ai_qs.struct_version, &old_version);
if (rc)
{
sprintf(log_buf, "Cannot upgrade array version %d to %d", pa->ai_qs.struct_version, ARRAY_QS_STRUCT_VERSION);
log_err(errno, __func__, log_buf);
free(pa);
close(fd);
return(rc);
}
}
}
pa->job_ids = (char **)calloc(pa->ai_qs.array_size, sizeof(char *));
/* check to see if there is any additional info saved in the array file */
/* check if there are any array request tokens that haven't been fully
processed */
if (old_version > 1)
{
if (read_ac_socket(fd, &num_tokens, sizeof(int)) != sizeof(int))
{
sprintf(log_buf, "error reading token count from %s", path);
log_err(errno, __func__, log_buf);
free(pa);
close(fd);
return(PBSE_SYSTEM);
}
for (i = 0; i < num_tokens; i++)
{
rn = (array_request_node *)calloc(1, sizeof(array_request_node));
if (read_ac_socket(fd, rn, sizeof(array_request_node)) != sizeof(array_request_node))
{
sprintf(log_buf, "error reading array_request_node from %s", path);
log_err(errno, __func__, log_buf);
free(rn);
for (rn = (array_request_node*)GET_NEXT(pa->request_tokens);
rn != NULL;
rn = (array_request_node*)GET_NEXT(pa->request_tokens))
{
delete_link(&rn->request_tokens_link);
free(rn);
}
free(pa);
close(fd);
return(PBSE_SYSTEM);
}
CLEAR_LINK(rn->request_tokens_link);
append_link(&pa->request_tokens, &rn->request_tokens_link, (void*)rn);
}
}
close(fd);
CLEAR_HEAD(pa->ai_qs.deps);
if (old_version != ARRAY_QS_STRUCT_VERSION)
{
/* resave the array struct if the version on disk is older than the current */
array_save(pa);
}
pa->ai_mutex = (pthread_mutex_t *)calloc(1, sizeof(pthread_mutex_t));
pthread_mutex_init(pa->ai_mutex,NULL);
lock_ai_mutex(pa, __func__, NULL, LOGLEVEL);
/* link the struct into the servers list of job arrays */
insert_array(pa);
*new_pa = pa;
return(PBSE_NONE);
} /* END array_recov() */
void mom_job_purge(
job *pjob) /* I (modified) */
{
job_file_delete_info *jfdi;
jfdi = (job_file_delete_info *)calloc(1, sizeof(job_file_delete_info));
if (jfdi == NULL)
{
log_err(ENOMEM,__func__, (char *)"No space to allocate info for job file deletion");
return;
}
#ifdef NVIDIA_GPUS
/*
* Did this job have a gpuid assigned?
* if so, then update gpu status
*/
if (((pjob->ji_wattr[JOB_ATR_exec_gpus].at_flags & ATR_VFLAG_SET) != 0) &&
(pjob->ji_wattr[JOB_ATR_exec_gpus].at_val.at_str != NULL))
{
send_update_soon();
}
#endif /* NVIDIA_GPUS */
/* initialize struct information */
if (pjob->ji_flags & MOM_HAS_TMPDIR)
{
jfdi->has_temp_dir = TRUE;
pjob->ji_flags &= ~MOM_HAS_TMPDIR;
}
else
jfdi->has_temp_dir = FALSE;
strcpy(jfdi->jobid,pjob->ji_qs.ji_jobid);
strcpy(jfdi->prefix,pjob->ji_qs.ji_fileprefix);
if ((pjob->ji_wattr[JOB_ATR_checkpoint_dir].at_flags & ATR_VFLAG_SET) &&
(pjob->ji_wattr[JOB_ATR_checkpoint_name].at_flags & ATR_VFLAG_SET))
jfdi->checkpoint_dir = strdup(pjob->ji_wattr[JOB_ATR_checkpoint_dir].at_val.at_str);
jfdi->gid = pjob->ji_qs.ji_un.ji_momt.ji_exgid;
jfdi->uid = pjob->ji_qs.ji_un.ji_momt.ji_exuid;
if (thread_unlink_calls == TRUE)
enqueue_threadpool_request(delete_job_files,jfdi);
else
delete_job_files(jfdi);
/* remove this job from the global queue */
delete_link(&pjob->ji_jobque);
delete_link(&pjob->ji_alljobs);
if (LOGLEVEL >= 6)
{
sprintf(log_buffer,"removing job");
log_record(PBSEVENT_DEBUG, PBS_EVENTCLASS_JOB, pjob->ji_qs.ji_jobid, log_buffer);
}
#if IBM_SP2==2 /* IBM SP PSSP 3.1 */
unload_sp_switch(pjob);
#endif /* IBM SP */
mom_job_free(pjob);
/* if no jobs are left, check if MOM should be restarted */
if (((job *)GET_NEXT(svr_alljobs)) == NULL)
MOMCheckRestart();
return;
} /* END mom_job_purge() */
int add_encoded_attributes(
xmlNodePtr *attr_node, /* M attribute node */
pbs_attribute *pattr) /* M ptr to pbs_attribute value array */
{
tlist_head lhead;
int i;
int resc_access_perm = ATR_DFLAG_ACCESS;
svrattrl *pal;
int rc = PBSE_NONE;
xmlNodePtr attributeNode = *attr_node;
char buf[BUFSIZE];
xmlNodePtr pal_xmlNode;
CLEAR_HEAD(lhead);
xmlNodePtr resource_list_head_node = NULL;
xmlNodePtr resource_used_head_node = NULL;
xmlNodePtr complete_req_head_node = NULL;
for (i = 0; ((i < JOB_ATR_LAST) && (rc >= 0)); i++)
{
if ((job_attr_def[i].at_type != ATR_TYPE_ACL) &&
((pattr + i)->at_flags & ATR_VFLAG_SET))
{
if ((i != JOB_ATR_resource) &&
(i != JOB_ATR_resc_used) &&
(i != JOB_ATR_req_information))
{
std::string value;
#ifndef PBS_MOM
if (i == JOB_ATR_depend)
translate_dependency_to_string(pattr + i, value);
else
#endif
attr_to_str(value, job_attr_def + i, pattr[i], true);
if (value.size() == 0)
continue;
pal_xmlNode = xmlNewChild(attributeNode,
NULL,
(xmlChar *)job_attr_def[i].at_name,
(const xmlChar *)value.c_str());
if (pal_xmlNode)
{
snprintf(buf, sizeof(buf), "%u", (unsigned int)pattr[i].at_flags);
xmlSetProp(pal_xmlNode, (const xmlChar *)AL_FLAGS_ATTR, (const xmlChar *)buf);
(pattr + i)->at_flags &= ~ATR_VFLAG_MODIFY;
}
}
else
{
rc = job_attr_def[i].at_encode(pattr + i,
&lhead,
job_attr_def[i].at_name,
NULL,
ATR_ENCODE_SAVE,
resc_access_perm);
if (rc < 0)
return -1;
(pattr + i)->at_flags &= ~ATR_VFLAG_MODIFY;
while ((pal = (svrattrl *)GET_NEXT(lhead)) != NULL)
{
if (i == JOB_ATR_resource)
{
pal_xmlNode = add_resource_list_attribute(ATTR_l,
attr_node,
&resource_list_head_node,
pal);
}
else if (i == JOB_ATR_req_information)
{
pal_xmlNode = add_resource_list_attribute(ATTR_req_information,
attr_node,
&complete_req_head_node,
pal);
}
else
{
pal_xmlNode = add_resource_list_attribute(ATTR_used,
attr_node,
&resource_used_head_node,
pal);
}
if (pal_xmlNode)
{
snprintf(buf, sizeof(buf), "%u", (unsigned int)pal->al_flags);
xmlSetProp(pal_xmlNode, (const xmlChar *)AL_FLAGS_ATTR, (const xmlChar *)buf);
}
delete_link(&pal->al_link);
free(pal);
if (!pal_xmlNode)
rc = -1;
}
}
}
}
return (0);
} /* END add_encoded_attributes */
void
free_br(struct batch_request *preq)
{
delete_link(&preq->rq_link);
reply_free(&preq->rq_reply);
if (preq->rq_parentbr) {
/*
* have a parent who has the original info, so we cannot
* free any data malloc-ed outside of the basic structure;
* decrement the reference count in the parent and when it
* goes to zero, reply_send() it
*/
if (preq->rq_parentbr->rq_refct > 0) {
if (--preq->rq_parentbr->rq_refct == 0)
reply_send(preq->rq_parentbr);
}
if (preq->rppcmd_msgid)
free(preq->rppcmd_msgid);
(void)free(preq);
return;
}
/*
* IMPORTANT - free any data that is malloc-ed outside of the
* basic batch_request structure below here so it is not freed
* when a copy of the structure (for a Array subjob) is freed
*/
if (preq->rq_extend)
(void)free(preq->rq_extend);
switch (preq->rq_type) {
case PBS_BATCH_QueueJob:
free_attrlist(&preq->rq_ind.rq_queuejob.rq_attr);
break;
case PBS_BATCH_JobCred:
if (preq->rq_ind.rq_jobcred.rq_data)
(void)free(preq->rq_ind.rq_jobcred.rq_data);
break;
case PBS_BATCH_UserCred:
if (preq->rq_ind.rq_usercred.rq_data)
(void)free(preq->rq_ind.rq_usercred.rq_data);
break;
case PBS_BATCH_jobscript:
if (preq->rq_ind.rq_jobfile.rq_data)
(void)free(preq->rq_ind.rq_jobfile.rq_data);
break;
case PBS_BATCH_CopyHookFile:
if (preq->rq_ind.rq_hookfile.rq_data)
(void)free(preq->rq_ind.rq_hookfile.rq_data);
break;
case PBS_BATCH_HoldJob:
freebr_manage(&preq->rq_ind.rq_hold.rq_orig);
break;
case PBS_BATCH_MessJob:
if (preq->rq_ind.rq_message.rq_text)
(void)free(preq->rq_ind.rq_message.rq_text);
break;
case PBS_BATCH_RelnodesJob:
if (preq->rq_ind.rq_relnodes.rq_node_list)
(void)free(preq->rq_ind.rq_relnodes.rq_node_list);
break;
case PBS_BATCH_PySpawn:
arrayfree(preq->rq_ind.rq_py_spawn.rq_argv);
arrayfree(preq->rq_ind.rq_py_spawn.rq_envp);
break;
case PBS_BATCH_ModifyJob:
case PBS_BATCH_ModifyResv:
freebr_manage(&preq->rq_ind.rq_modify);
break;
case PBS_BATCH_RunJob:
case PBS_BATCH_AsyrunJob:
case PBS_BATCH_StageIn:
case PBS_BATCH_ConfirmResv:
if (preq->rq_ind.rq_run.rq_destin)
(void)free(preq->rq_ind.rq_run.rq_destin);
break;
case PBS_BATCH_StatusJob:
case PBS_BATCH_StatusQue:
case PBS_BATCH_StatusNode:
case PBS_BATCH_StatusSvr:
case PBS_BATCH_StatusSched:
case PBS_BATCH_StatusHook:
case PBS_BATCH_StatusRsc:
case PBS_BATCH_StatusResv:
if (preq->rq_ind.rq_status.rq_id)
free(preq->rq_ind.rq_status.rq_id);
free_attrlist(&preq->rq_ind.rq_status.rq_attr);
break;
case PBS_BATCH_CopyFiles:
case PBS_BATCH_DelFiles:
freebr_cpyfile(&preq->rq_ind.rq_cpyfile);
break;
case PBS_BATCH_CopyFiles_Cred:
case PBS_BATCH_DelFiles_Cred:
freebr_cpyfile_cred(&preq->rq_ind.rq_cpyfile_cred);
break;
case PBS_BATCH_MvJobFile:
if (preq->rq_ind.rq_jobfile.rq_data)
free(preq->rq_ind.rq_jobfile.rq_data);
break;
#ifndef PBS_MOM /* Server Only */
case PBS_BATCH_SubmitResv:
free_attrlist(&preq->rq_ind.rq_queuejob.rq_attr);
break;
case PBS_BATCH_Manager:
freebr_manage(&preq->rq_ind.rq_manager);
break;
case PBS_BATCH_ReleaseJob:
freebr_manage(&preq->rq_ind.rq_release);
break;
case PBS_BATCH_Rescq:
case PBS_BATCH_ReserveResc:
case PBS_BATCH_ReleaseResc:
free_rescrq(&preq->rq_ind.rq_rescq);
break;
case PBS_BATCH_DefSchReply:
free(preq->rq_ind.rq_defrpy.rq_id);
free(preq->rq_ind.rq_defrpy.rq_txt);
break;
case PBS_BATCH_SelectJobs:
case PBS_BATCH_SelStat:
free_attrlist(&preq->rq_ind.rq_select.rq_selattr);
free_attrlist(&preq->rq_ind.rq_select.rq_rtnattr);
break;
case PBS_BATCH_PreemptJobs:
free(preq->rq_ind.rq_preempt.ppj_list);
break;
#endif /* PBS_MOM */
}
if (preq->rppcmd_msgid)
free(preq->rppcmd_msgid);
(void)free(preq);
}
