void reply_sid( struct batch_request *preq, /* M */ long sid, /* I */ int which) /* I */ { char sid_text[MAXLINE]; if (preq->rq_reply.brp_choice != BATCH_REPLY_CHOICE_NULL) { /* in case another reply was being built up, clean it out */ reply_free(&preq->rq_reply); } sprintf(sid_text,"%lu",sid); preq->rq_reply.brp_choice = which; preq->rq_reply.brp_un.brp_txt.brp_str = strdup(sid_text); preq->rq_reply.brp_un.brp_txt.brp_txtlen = strlen(sid_text); preq->rq_reply.brp_code = 0; preq->rq_reply.brp_auxcode = 0; reply_send_mom(preq); } /* END reply_text() */
void reply_text( batch_request *preq, int code, const char *text) /* I */ { if (preq->rq_reply.brp_choice != BATCH_REPLY_CHOICE_NULL) { /* in case another reply was being built up, clean it out */ reply_free(&preq->rq_reply); } preq->rq_reply.brp_code = code; preq->rq_reply.brp_auxcode = 0; if ((text != NULL) && (text[0] != '\0')) { preq->rq_reply.brp_choice = BATCH_REPLY_CHOICE_Text; preq->rq_reply.brp_un.brp_txt.brp_str = strdup(text); preq->rq_reply.brp_un.brp_txt.brp_txtlen = strlen(text); } else { preq->rq_reply.brp_choice = BATCH_REPLY_CHOICE_NULL; } reply_send(preq); return; } /* END reply_text() */
void req_stat_que(struct batch_request *preq) { char *name; pbs_queue *pque; struct batch_reply *preply; int rc = 0; int type = 0; /* * first, validate the name of the requested object, either * a queue, or null for all queues */ name = preq->rq_ind.rq_status.rq_id; if ((*name == '\0') || (*name =='@')) type = 1; else { pque = find_queuebyname(name); #ifdef NAS /* localmod 075 */ if (pque == NULL) pque = find_resvqueuebyname(name); #endif /* localmod 075 */ if (pque == NULL) { req_reject(PBSE_UNKQUE, 0, preq); return; } } preply = &preq->rq_reply; preply->brp_choice = BATCH_REPLY_CHOICE_Status; CLEAR_HEAD(preply->brp_un.brp_status); if (type == 0) { /* get status of the one named queue */ rc = status_que(pque, preq, &preply->brp_un.brp_status); } else { /* get status of queues */ pque = (pbs_queue *)GET_NEXT(svr_queues); while (pque) { rc = status_que(pque, preq, &preply->brp_un.brp_status); if (rc != 0) { if (rc == PBSE_PERM) rc = 0; else break; } pque = (pbs_queue *)GET_NEXT(pque->qu_link); } } if (rc) { (void)reply_free(preply); req_reject(rc, bad, preq); } else { (void)reply_send(preq); } }
void req_stat_svr(struct batch_request *preq) { svrattrl *pal; struct batch_reply *preply; struct brp_status *pstat; /* update count and state counts from sv_numjobs and sv_jobstates */ server.sv_attr[(int)SRV_ATR_TotalJobs].at_val.at_long = server.sv_qs.sv_numjobs; server.sv_attr[(int)SRV_ATR_TotalJobs].at_flags |= ATR_VFLAG_SET|ATR_VFLAG_MODCACHE; update_state_ct(&server.sv_attr[(int)SRV_ATR_JobsByState], server.sv_jobstates, server.sv_jobstbuf); update_license_ct(&server.sv_attr[(int)SRV_ATR_license_count], server.sv_license_ct_buf); /* allocate a reply structure and a status sub-structure */ preply = &preq->rq_reply; preply->brp_choice = BATCH_REPLY_CHOICE_Status; CLEAR_HEAD(preply->brp_un.brp_status); pstat = (struct brp_status *)malloc(sizeof(struct brp_status)); if (pstat == NULL) { reply_free(preply); req_reject(PBSE_SYSTEM, 0, preq); return; } CLEAR_LINK(pstat->brp_stlink); (void)strcpy(pstat->brp_objname, server_name); pstat->brp_objtype = MGR_OBJ_SERVER; CLEAR_HEAD(pstat->brp_attr); append_link(&preply->brp_un.brp_status, &pstat->brp_stlink, pstat); /* add attributes to the status reply */ bad = 0; pal = (svrattrl *)GET_NEXT(preq->rq_ind.rq_status.rq_attr); if (status_attrib(pal, svr_attr_def, server.sv_attr, SRV_ATR_LAST, preq->rq_perm, &pstat->brp_attr, &bad)) reply_badattr(PBSE_NOATTR, bad, pal, preq); else (void)reply_send(preq); }
static void connection_free_link (struct connection *c) { PacketPassFifoQueue_PrepareFree(&c->send_queue); LinkedList0Node *ln; while (ln = LinkedList0_GetFirst(&c->replies_list)) { struct reply *r = UPPER_OBJECT(ln, struct reply, replies_list_node); ASSERT(r->con == c) reply_free(r); } PacketPassFifoQueue_Free(&c->send_queue); PacketStreamSender_Free(&c->send_pss); PacketProtoDecoder_Free(&c->recv_decoder); PacketPassInterface_Free(&c->recv_if); BConnection_RecvAsync_Free(&c->con); BConnection_SendAsync_Free(&c->con); BConnection_Free(&c->con); }
int req_stat_svr( struct batch_request *preq) /* ptr to the decoded request */ { svrattrl *pal; struct batch_reply *preply; struct brp_status *pstat; int bad = 0; char nc_buf[128]; int numjobs; int netrates[3]; memset(netrates, 0, sizeof(netrates)); /* update count and state counts from sv_numjobs and sv_jobstates */ lock_sv_qs_mutex(server.sv_qs_mutex, __func__); numjobs = server.sv_qs.sv_numjobs; unlock_sv_qs_mutex(server.sv_qs_mutex, __func__); pthread_mutex_lock(server.sv_attr_mutex); server.sv_attr[SRV_ATR_TotalJobs].at_val.at_long = numjobs; server.sv_attr[SRV_ATR_TotalJobs].at_flags |= ATR_VFLAG_SET; pthread_mutex_lock(server.sv_jobstates_mutex); update_state_ct( &server.sv_attr[SRV_ATR_JobsByState], server.sv_jobstates, server.sv_jobstbuf); pthread_mutex_unlock(server.sv_jobstates_mutex); netcounter_get(netrates); snprintf(nc_buf, 127, "%d %d %d", netrates[0], netrates[1], netrates[2]); if (server.sv_attr[SRV_ATR_NetCounter].at_val.at_str != NULL) free(server.sv_attr[SRV_ATR_NetCounter].at_val.at_str); server.sv_attr[SRV_ATR_NetCounter].at_val.at_str = strdup(nc_buf); if (server.sv_attr[SRV_ATR_NetCounter].at_val.at_str != NULL) server.sv_attr[SRV_ATR_NetCounter].at_flags |= ATR_VFLAG_SET; pthread_mutex_unlock(server.sv_attr_mutex); /* allocate a reply structure and a status sub-structure */ preply = &preq->rq_reply; preply->brp_choice = BATCH_REPLY_CHOICE_Status; CLEAR_HEAD(preply->brp_un.brp_status); pstat = (struct brp_status *)calloc(1, sizeof(struct brp_status)); if (pstat == NULL) { reply_free(preply); req_reject(PBSE_SYSTEM, 0, preq, NULL, NULL); pthread_mutex_unlock(server.sv_attr_mutex); return(PBSE_SYSTEM); } CLEAR_LINK(pstat->brp_stlink); strcpy(pstat->brp_objname, server_name); pstat->brp_objtype = MGR_OBJ_SERVER; CLEAR_HEAD(pstat->brp_attr); append_link(&preply->brp_un.brp_status, &pstat->brp_stlink, pstat); /* add attributes to the status reply */ pal = (svrattrl *)GET_NEXT(preq->rq_ind.rq_status.rq_attr); if (status_attrib( pal, svr_attr_def, server.sv_attr, SRV_ATR_LAST, preq->rq_perm, &pstat->brp_attr, &bad, 1)) /* IsOwner == TRUE */ { reply_badattr(PBSE_NOATTR, bad, pal, preq); } else { reply_send_svr(preq); } return(PBSE_NONE); } /* END req_stat_svr() */
int req_stat_que( batch_request *preq) { char *name; pbs_queue *pque = NULL; struct batch_reply *preply; int rc = 0; int type = 0; char log_buf[LOCAL_LOG_BUF_SIZE+1]; /* * first, validate the name of the requested object, either * a queue, or null for all queues */ name = preq->rq_ind.rq_status.rq_id; if ((*name == '\0') || (*name == '@')) { type = 1; } else { pque = find_queuebyname(name); if (pque == NULL) { rc = PBSE_UNKQUE; snprintf(log_buf, LOCAL_LOG_BUF_SIZE, "cannot locate queue %s", name); req_reject(rc, 0, preq, NULL, log_buf); return rc; } } preply = &preq->rq_reply; preply->brp_choice = BATCH_REPLY_CHOICE_Status; CLEAR_HEAD(preply->brp_un.brp_status); if (type == 0) { /* get status of the named queue */ mutex_mgr pque_mutex = mutex_mgr(pque->qu_mutex, true); rc = status_que(pque, preq, &preply->brp_un.brp_status); /* pque_qu_mutex will be unlocked in the destructor when we leave this scope */ } else { /* pque == NULL before next_queue */ svr_queues.lock(); all_queues_iterator *iter = svr_queues.get_iterator(); svr_queues.unlock(); /* get status of all queues */ while ((pque = next_queue(&svr_queues,iter)) != NULL) { mutex_mgr pque_mutex = mutex_mgr(pque->qu_mutex, true); rc = status_que(pque, preq, &preply->brp_un.brp_status); if (rc != 0) { if (rc != PBSE_PERM) { break; } rc = 0; } } delete iter; } if (rc != PBSE_NONE) { reply_free(preply); req_reject(PBSE_NOATTR, rc, preq, NULL, "status_queue failed"); } else { reply_send_svr(preq); } return rc; } /* END req_stat_que() */
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_GSS_Context: if (preq->rq_ind.rq_gssdata.rq_data) free(preq->rq_ind.rq_gssdata.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; #endif /* PBS_MOM */ } if (preq->rppcmd_msgid) free(preq->rppcmd_msgid); (void)free(preq); }
static void req_stat_job_step2( struct stat_cntl *cntl) /* I/O (free'd on return) */ { svrattrl *pal; job *pjob = NULL; struct batch_request *preq; struct batch_reply *preply; int rc = 0; enum TJobStatTypeEnum type; pbs_queue *pque = NULL; int exec_only = 0; int bad = 0; long DTime; /* delta time - only report full pbs_attribute list if J->MTime > DTime */ static svrattrl *dpal = NULL; int job_array_index = 0; job_array *pa = NULL; char log_buf[LOCAL_LOG_BUF_SIZE]; int iter; time_t time_now = time(NULL); long poll_jobs = 0; char job_id[PBS_MAXSVRJOBID+1]; int job_substate = -1; time_t job_momstattime = -1; preq = cntl->sc_origrq; type = (enum TJobStatTypeEnum)cntl->sc_type; preply = &preq->rq_reply; /* See pbs_server_attributes(1B) for details on "poll_jobs" behaviour */ if (dpal == NULL) { /* build 'delta' pbs_attribute list */ svrattrl *tpal; tlist_head dalist; int aindex; int atrlist[] = { JOB_ATR_jobname, JOB_ATR_resc_used, JOB_ATR_LAST }; CLEAR_LINK(dalist); for (aindex = 0;atrlist[aindex] != JOB_ATR_LAST;aindex++) { if ((tpal = attrlist_create("", "", 23)) == NULL) { return; } tpal->al_valln = atrlist[aindex]; if (dpal == NULL) dpal = tpal; append_link(&dalist, &tpal->al_link, tpal); } } /* END if (dpal == NULL) */ if (type == tjstArray) { pa = get_array(preq->rq_ind.rq_status.rq_id); if (pa == NULL) { req_reject(PBSE_UNKARRAYID, 0, preq, NULL, "unable to find array"); return; } } iter = -1; get_svr_attr_l(SRV_ATR_PollJobs, &poll_jobs); if (!poll_jobs) { /* polljobs not set - indicates we may need to obtain fresh data from MOM */ if (cntl->sc_jobid[0] == '\0') pjob = NULL; else pjob = svr_find_job(cntl->sc_jobid, FALSE); while (1) { if (pjob == NULL) { /* start from the first job */ if (type == tjstJob) { pjob = svr_find_job(preq->rq_ind.rq_status.rq_id, FALSE); } else if (type == tjstQueue) { pjob = next_job(cntl->sc_pque->qu_jobs,&iter); } else if (type == tjstArray) { job_array_index = 0; /* increment job_array_index until we find a non-null pointer or hit the end */ while (job_array_index < pa->ai_qs.array_size) { if (pa->job_ids[job_array_index] != NULL) { if ((pjob = svr_find_job(pa->job_ids[job_array_index], FALSE)) != NULL) { unlock_ji_mutex(pjob, __func__, "2", LOGLEVEL); break; } } job_array_index++; } } else { pjob = next_job(&alljobs,&iter); } } /* END if (pjob == NULL) */ else { strcpy(job_id, pjob->ji_qs.ji_jobid); unlock_ji_mutex(pjob, __func__, "3", LOGLEVEL); if (type == tjstJob) break; if (type == tjstQueue) pjob = next_job(cntl->sc_pque->qu_jobs,&iter); else if (type == tjstArray) { pjob = NULL; /* increment job_array_index until we find a non-null pointer or hit the end */ while (++job_array_index < pa->ai_qs.array_size) { if (pa->job_ids[job_array_index] != NULL) { if ((pjob = svr_find_job(pa->job_ids[job_array_index], FALSE)) != NULL) { unlock_ji_mutex(pjob, __func__, "3", LOGLEVEL); break; } } } } else pjob = next_job(&alljobs,&iter); } if (pjob == NULL) break; strcpy(job_id, pjob->ji_qs.ji_jobid); job_substate = pjob->ji_qs.ji_substate; job_momstattime = pjob->ji_momstat; strcpy(cntl->sc_jobid, job_id); unlock_ji_mutex(pjob, __func__, "4", LOGLEVEL); pjob = NULL; /* PBS_RESTAT_JOB defaults to 30 seconds */ if ((job_substate == JOB_SUBSTATE_RUNNING) && ((time_now - job_momstattime) > JobStatRate)) { /* go to MOM for status */ if ((rc = stat_to_mom(job_id, cntl)) == PBSE_MEM_MALLOC) break; if (rc != 0) { pjob = svr_find_job(job_id, FALSE); rc = 0; continue; } if (pa != NULL) unlock_ai_mutex(pa, __func__, "1", LOGLEVEL); return; /* will pick up after mom replies */ } } /* END while(1) */ if (rc != 0) { if (pa != NULL) unlock_ai_mutex(pa, __func__, "2", LOGLEVEL); reply_free(preply); req_reject(rc, 0, preq, NULL, "cannot get update from mom"); return; } } /* END if (!server.sv_attr[SRV_ATR_PollJobs].at_val.at_long) */ /* * now ready for part 3, building the status reply, * loop through again */ if ((type == tjstSummarizeArraysQueue) || (type == tjstSummarizeArraysServer)) { /* No array can be owned for these options */ update_array_statuses(); } if (type == tjstJob) pjob = svr_find_job(preq->rq_ind.rq_status.rq_id, FALSE); else if (type == tjstQueue) pjob = next_job(cntl->sc_pque->qu_jobs,&iter); else if (type == tjstSummarizeArraysQueue) pjob = next_job(cntl->sc_pque->qu_jobs_array_sum,&iter); else if (type == tjstSummarizeArraysServer) pjob = next_job(&array_summary,&iter); else if (type == tjstArray) { job_array_index = -1; pjob = NULL; /* increment job_array_index until we find a non-null pointer or hit the end */ while (++job_array_index < pa->ai_qs.array_size) { if (pa->job_ids[job_array_index] != NULL) { if ((pjob = svr_find_job(pa->job_ids[job_array_index], FALSE)) != NULL) { break; } } } } else pjob = next_job(&alljobs,&iter); DTime = 0; if (preq->rq_extend != NULL) { char *ptr; /* FORMAT: { EXECQONLY | DELTA:<EPOCHTIME> } */ if (strstr(preq->rq_extend, EXECQUEONLY)) exec_only = 1; ptr = strstr(preq->rq_extend, "DELTA:"); if (ptr != NULL) { ptr += strlen("delta:"); DTime = strtol(ptr, NULL, 10); } } if ((type == tjstTruncatedServer) || (type == tjstTruncatedQueue)) { long sentJobCounter; long qjcounter; long qmaxreport; int iter = -1; /* loop through all queues */ while ((pque = next_queue(&svr_queues,&iter)) != NULL) { qjcounter = 0; if ((exec_only == 1) && (pque->qu_qs.qu_type != QTYPE_Execution)) { /* ignore routing queues */ unlock_queue(pque, __func__, "ignore queue", LOGLEVEL); continue; } if (((pque->qu_attr[QA_ATR_MaxReport].at_flags & ATR_VFLAG_SET) != 0) && (pque->qu_attr[QA_ATR_MaxReport].at_val.at_long >= 0)) { qmaxreport = pque->qu_attr[QA_ATR_MaxReport].at_val.at_long; } else { qmaxreport = TMAX_JOB; } if (LOGLEVEL >= 5) { sprintf(log_buf,"giving scheduler up to %ld idle jobs in queue %s\n", qmaxreport, pque->qu_qs.qu_name); log_event(PBSEVENT_SYSTEM,PBS_EVENTCLASS_QUEUE,pque->qu_qs.qu_name,log_buf); } sentJobCounter = 0; /* loop through jobs in queue */ if (pjob != NULL) unlock_ji_mutex(pjob, __func__, "5", LOGLEVEL); iter = -1; while ((pjob = next_job(pque->qu_jobs,&iter)) != NULL) { if ((qjcounter >= qmaxreport) && (pjob->ji_qs.ji_state == JOB_STATE_QUEUED)) { /* max_report of queued jobs reached for queue */ unlock_ji_mutex(pjob, __func__, "6", LOGLEVEL); continue; } pal = (svrattrl *)GET_NEXT(preq->rq_ind.rq_status.rq_attr); rc = status_job( pjob, preq, (pjob->ji_wattr[JOB_ATR_mtime].at_val.at_long >= DTime) ? pal : dpal, &preply->brp_un.brp_status, &bad); if ((rc != 0) && (rc != PBSE_PERM)) { req_reject(rc, bad, preq, NULL, NULL); if (pa != NULL) { unlock_ai_mutex(pa, __func__, "1", LOGLEVEL); } unlock_ji_mutex(pjob, __func__, "7", LOGLEVEL); unlock_queue(pque, __func__, "perm", LOGLEVEL); return; } sentJobCounter++; if (pjob->ji_qs.ji_state == JOB_STATE_QUEUED) qjcounter++; unlock_ji_mutex(pjob, __func__, "8", LOGLEVEL); } /* END foreach (pjob from pque) */ if (LOGLEVEL >= 5) { sprintf(log_buf,"sent scheduler %ld total jobs for queue %s\n", sentJobCounter, pque->qu_qs.qu_name); log_event(PBSEVENT_SYSTEM,PBS_EVENTCLASS_QUEUE,pque->qu_qs.qu_name,log_buf); } unlock_queue(pque, __func__, "end while", LOGLEVEL); } /* END for (pque) */ if (pa != NULL) unlock_ai_mutex(pa, __func__, "1", LOGLEVEL); reply_send_svr(preq); return; } /* END if ((type == tjstTruncatedServer) || ...) */ while (pjob != NULL) { /* go ahead and build the status reply for this job */ if (exec_only) { if (cntl->sc_pque != NULL) { if (cntl->sc_pque->qu_qs.qu_type != QTYPE_Execution) goto nextjob; } else { if (pa != NULL) pthread_mutex_unlock(pa->ai_mutex); pque = get_jobs_queue(&pjob); if (pa != NULL) pthread_mutex_lock(pa->ai_mutex); if ((pjob == NULL) || (pque == NULL)) goto nextjob; if (pque->qu_qs.qu_type != QTYPE_Execution) { unlock_queue(pque, __func__, "not exec", LOGLEVEL); goto nextjob; } unlock_queue(pque, __func__, "exec", LOGLEVEL); } } pal = (svrattrl *)GET_NEXT(preq->rq_ind.rq_status.rq_attr); rc = status_job( pjob, preq, pal, &preply->brp_un.brp_status, &bad); if ((rc != 0) && (rc != PBSE_PERM)) { if (pa != NULL) { unlock_ai_mutex(pa, __func__, "1", LOGLEVEL); } unlock_ji_mutex(pjob, __func__, "9", LOGLEVEL); req_reject(rc, bad, preq, NULL, NULL); return; } /* get next job */ nextjob: if (pjob != NULL) unlock_ji_mutex(pjob, __func__, "10", LOGLEVEL); if (type == tjstJob) break; if (type == tjstQueue) pjob = next_job(cntl->sc_pque->qu_jobs,&iter); else if (type == tjstSummarizeArraysQueue) pjob = next_job(cntl->sc_pque->qu_jobs_array_sum,&iter); else if (type == tjstSummarizeArraysServer) pjob = next_job(&array_summary,&iter); else if (type == tjstArray) { pjob = NULL; /* increment job_array_index until we find a non-null pointer or hit the end */ while (++job_array_index < pa->ai_qs.array_size) { if (pa->job_ids[job_array_index] != NULL) { if ((pjob = svr_find_job(pa->job_ids[job_array_index], FALSE)) != NULL) { break; } } } } else pjob = next_job(&alljobs,&iter); rc = 0; } /* END while (pjob != NULL) */ if (pa != NULL) { unlock_ai_mutex(pa, __func__, "1", LOGLEVEL); } reply_send_svr(preq); if (LOGLEVEL >= 7) { log_event(PBSEVENT_SYSTEM, PBS_EVENTCLASS_JOB, "req_statjob", "Successfully returned the status of queued jobs\n"); } return; } /* END req_stat_job_step2() */
int req_stat_que( struct batch_request *preq) { char *name; pbs_queue *pque = NULL; struct batch_reply *preply; int rc = 0; int type = 0; char log_buf[LOCAL_LOG_BUF_SIZE+1]; /* * first, validate the name of the requested object, either * a queue, or null for all queues */ name = preq->rq_ind.rq_status.rq_id; if ((*name == '\0') || (*name == '@')) { type = 1; } else { pque = find_queuebyname(name); if (pque == NULL) { rc = PBSE_UNKQUE; snprintf(log_buf, LOCAL_LOG_BUF_SIZE, "can not locate queue %s", name); req_reject(rc, 0, preq, NULL, log_buf); return rc; } } preply = &preq->rq_reply; preply->brp_choice = BATCH_REPLY_CHOICE_Status; CLEAR_HEAD(preply->brp_un.brp_status); if (type == 0) { /* get status of the named queue */ rc = status_que(pque, preq, &preply->brp_un.brp_status); unlock_queue(pque, "req_stat_que", "type == 0", LOGLEVEL); } else { /* pque == NULL before next_queue */ int iter = -1; /* get status of all queues */ while ((pque = next_queue(&svr_queues,&iter)) != NULL) { rc = status_que(pque, preq, &preply->brp_un.brp_status); if (rc != 0) { if (rc != PBSE_PERM) { unlock_queue(pque, "req_stat_que", "break", LOGLEVEL); break; } rc = 0; } unlock_queue(pque, "req_stat_que", "end while", LOGLEVEL); } } if (rc != PBSE_NONE) { reply_free(preply); req_reject(PBSE_NOATTR, rc, preq, NULL, "status_queue failed"); } else { reply_send_svr(preq); } return rc; } /* END req_stat_que() */
void free_br( struct batch_request *preq) { if (preq->rq_id != NULL) { remove_batch_request(preq->rq_id); free(preq->rq_id); preq->rq_id = NULL; } reply_free(&preq->rq_reply); if (preq->rq_extend) { free(preq->rq_extend); preq->rq_extend = NULL; } 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); preq->rq_ind.rq_jobcred.rq_data = NULL; } 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); preq->rq_ind.rq_jobfile.rq_data = NULL; } 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); preq->rq_ind.rq_message.rq_text = NULL; } 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; 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: free_rescrq(&preq->rq_ind.rq_rescq); break; case PBS_BATCH_SelectJobs: case PBS_BATCH_SelStat: free_attrlist(&preq->rq_ind.rq_select); break; case PBS_BATCH_RunJob: case PBS_BATCH_AsyrunJob: if (preq->rq_ind.rq_run.rq_destin) { free(preq->rq_ind.rq_run.rq_destin); preq->rq_ind.rq_run.rq_destin = NULL; } break; default: /* NO-OP */ break; } /* END switch (preq->rq_type) */ free(preq); return; } /* END free_br() */
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; #ifndef PBS_MOM /* Server Only */ 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: free_rescrq(&preq->rq_ind.rq_rescq); break; case PBS_BATCH_SelectJobs: case PBS_BATCH_SelStat: free_attrlist(&preq->rq_ind.rq_select); break; case PBS_BATCH_RunJob: case PBS_BATCH_AsyrunJob: if (preq->rq_ind.rq_run.rq_destin) free(preq->rq_ind.rq_run.rq_destin); break; #endif /* !PBS_MOM */ default: /* NO-OP */ break; } /* END switch (preq->rq_type) */ free(preq); return; } /* END free_br() */
static void req_stat_job_step2( struct stat_cntl *cntl) /* I/O (freed on return) */ { svrattrl *pal; job *pjob = NULL; struct batch_request *preq; struct batch_reply *preply; int rc = 0; enum TJobStatTypeEnum type; pbs_queue *pque = NULL; int exec_only = 0; int IsTruncated = 0; long DTime; /* delta time - only report full attribute list if J->MTime > DTime */ static svrattrl *dpal = NULL; int job_array_index = 0; job_array *pa = NULL; preq = cntl->sc_origrq; type = (enum TJobStatTypeEnum)cntl->sc_type; preply = &preq->rq_reply; /* See pbs_server_attributes(1B) for details on "poll_jobs" behaviour */ /* NOTE: If IsTruncated is true, should walk all queues and walk jobs in each queue until max_reported is reached (NYI) */ if (dpal == NULL) { /* build 'delta' attribute list */ svrattrl *tpal; tlist_head dalist; int aindex; int atrlist[] = { JOB_ATR_jobname, JOB_ATR_resc_used, JOB_ATR_LAST }; CLEAR_LINK(dalist); for (aindex = 0;atrlist[aindex] != JOB_ATR_LAST;aindex++) { if ((tpal = attrlist_create("", "", 23)) == NULL) { return; } tpal->al_valln = atrlist[aindex]; if (dpal == NULL) dpal = tpal; append_link(&dalist, &tpal->al_link, tpal); } } /* END if (dpal == NULL) */ if (type == tjstArray) { pa = get_array(preq->rq_ind.rq_status.rq_id); } if (!server.sv_attr[(int)SRV_ATR_PollJobs].at_val.at_long) { /* polljobs not set - indicates we may need to obtain fresh data from MOM */ if (cntl->sc_jobid[0] == '\0') pjob = NULL; else pjob = find_job(cntl->sc_jobid); while (1) { if (pjob == NULL) { /* start from the first job */ if (type == tjstJob) { pjob = find_job(preq->rq_ind.rq_status.rq_id); } else if (type == tjstQueue) { pjob = (job *)GET_NEXT(cntl->sc_pque->qu_jobs); } else if (type == tjstArray) { job_array_index = 0; /* increment job_array_index until we find a non-null pointer or hit the end */ while (job_array_index < pa->ai_qs.array_size && (pjob = pa->jobs[job_array_index]) == NULL) job_array_index++; } else { if ((type == tjstTruncatedServer) || (type == tjstTruncatedQueue)) IsTruncated = TRUE; pjob = (job *)GET_NEXT(svr_alljobs); } } /* END if (pjob == NULL) */ else { /* get next job */ if (type == tjstJob) break; if (type == tjstQueue) pjob = (job *)GET_NEXT(pjob->ji_jobque); else pjob = (job *)GET_NEXT(pjob->ji_alljobs); if (type == tjstArray) { pjob = NULL; /* increment job_array_index until we find a non-null pointer or hit the end */ while (++job_array_index < pa->ai_qs.array_size && (pjob = pa->jobs[job_array_index]) == NULL) ; } } if (pjob == NULL) break; /* PBS_RESTAT_JOB defaults to 30 seconds */ if ((pjob->ji_qs.ji_substate == JOB_SUBSTATE_RUNNING) && ((time_now - pjob->ji_momstat) > JobStatRate)) { /* go to MOM for status */ strcpy(cntl->sc_jobid, pjob->ji_qs.ji_jobid); if ((rc = stat_to_mom(pjob, cntl)) == PBSE_SYSTEM) { break; } if (rc != 0) { rc = 0; continue; } return; /* will pick up after mom replies */ } } /* END while(1) */ if (cntl->sc_conn >= 0) svr_disconnect(cntl->sc_conn); /* close connection to MOM */ if (rc != 0) { free(cntl); reply_free(preply); req_reject(rc, 0, preq, NULL, "cannot get update from mom"); return; } } /* END if (!server.sv_attr[(int)SRV_ATR_PollJobs].at_val.at_long) */ /* * now ready for part 3, building the status reply, * loop through again */ if (type == tjstSummarizeArraysQueue || type == tjstSummarizeArraysServer) { update_array_statuses(); } if (type == tjstJob) pjob = find_job(preq->rq_ind.rq_status.rq_id); else if (type == tjstQueue) pjob = (job *)GET_NEXT(cntl->sc_pque->qu_jobs); else if (type == tjstSummarizeArraysQueue) pjob = (job *)GET_NEXT(cntl->sc_pque->qu_jobs_array_sum); else if (type == tjstSummarizeArraysServer) pjob = (job *)GET_NEXT(svr_jobs_array_sum); else if (type == tjstArray) { job_array_index = 0; pjob = NULL; /* increment job_array_index until we find a non-null pointer or hit the end */ while (job_array_index < pa->ai_qs.array_size && (pjob = pa->jobs[job_array_index]) == NULL) job_array_index++; } else pjob = (job *)GET_NEXT(svr_alljobs); DTime = 0; if (preq->rq_extend != NULL) { char *ptr; /* FORMAT: { EXECQONLY | DELTA:<EPOCHTIME> } */ if (strstr(preq->rq_extend, EXECQUEONLY)) exec_only = 1; ptr = strstr(preq->rq_extend, "DELTA:"); if (ptr != NULL) { ptr += strlen("delta:"); DTime = strtol(ptr, NULL, 10); } } free(cntl); if ((type == tjstTruncatedServer) || (type == tjstTruncatedQueue)) { long sentJobCounter; long qjcounter; long qmaxreport; /* loop through all queues */ for (pque = (pbs_queue *)GET_NEXT(svr_queues); pque != NULL; pque = (pbs_queue *)GET_NEXT(pque->qu_link)) { qjcounter = 0; if ((exec_only == 1) && (pque->qu_qs.qu_type != QTYPE_Execution)) { /* ignore routing queues */ continue; } if (((pque->qu_attr[QA_ATR_MaxReport].at_flags & ATR_VFLAG_SET) != 0) && (pque->qu_attr[QA_ATR_MaxReport].at_val.at_long >= 0)) { qmaxreport = pque->qu_attr[QA_ATR_MaxReport].at_val.at_long; } else { qmaxreport = TMAX_JOB; } if (LOGLEVEL >= 5) { sprintf(log_buffer,"giving scheduler up to %ld idle jobs in queue %s\n", qmaxreport, pque->qu_qs.qu_name); log_event( PBSEVENT_SYSTEM, PBS_EVENTCLASS_QUEUE, pque->qu_qs.qu_name, log_buffer); } sentJobCounter = 0; /* loop through jobs in queue */ for (pjob = (job *)GET_NEXT(pque->qu_jobs); pjob != NULL; pjob = (job *)GET_NEXT(pjob->ji_jobque)) { if ((qjcounter >= qmaxreport) && (pjob->ji_qs.ji_state == JOB_STATE_QUEUED)) { /* max_report of queued jobs reached for queue */ continue; } pal = (svrattrl *)GET_NEXT(preq->rq_ind.rq_status.rq_attr); rc = status_job( pjob, preq, (pjob->ji_wattr[(int)JOB_ATR_mtime].at_val.at_long >= DTime) ? pal : dpal, &preply->brp_un.brp_status, &bad); if ((rc != 0) && (rc != PBSE_PERM)) { req_reject(rc, bad, preq, NULL, NULL); return; } sentJobCounter++; if (pjob->ji_qs.ji_state == JOB_STATE_QUEUED) qjcounter++; } /* END for (pjob) */ if (LOGLEVEL >= 5) { sprintf(log_buffer,"sent scheduler %ld total jobs for queue %s\n", sentJobCounter, pque->qu_qs.qu_name); log_event( PBSEVENT_SYSTEM, PBS_EVENTCLASS_QUEUE, pque->qu_qs.qu_name, log_buffer); } } /* END for (pque) */ reply_send(preq); return; } /* END if ((type == tjstTruncatedServer) || ...) */ while (pjob != NULL) { /* go ahead and build the status reply for this job */ if (exec_only) { pque = find_queuebyname(pjob->ji_qs.ji_queue); if (pque->qu_qs.qu_type != QTYPE_Execution) goto nextjob; } pal = (svrattrl *)GET_NEXT(preq->rq_ind.rq_status.rq_attr); rc = status_job( pjob, preq, pal, &preply->brp_un.brp_status, &bad); if ((rc != 0) && (rc != PBSE_PERM)) { req_reject(rc, bad, preq, NULL, NULL); return; } /* get next job */ nextjob: if (type == tjstJob) break; if (type == tjstQueue) pjob = (job *)GET_NEXT(pjob->ji_jobque); else if (type == tjstSummarizeArraysQueue) pjob = (job *)GET_NEXT(pjob->ji_jobque_array_sum); else if (type == tjstSummarizeArraysServer) pjob = (job *)GET_NEXT(pjob->ji_jobs_array_sum); else if (type == tjstArray) { pjob = NULL; /* increment job_array_index until we find a non-null pointer or hit the end */ while (++job_array_index < pa->ai_qs.array_size && (pjob = pa->jobs[job_array_index]) == NULL) ; } else pjob = (job *)GET_NEXT(pjob->ji_alljobs); rc = 0; } /* END while (pjob != NULL) */ reply_send(preq); if (LOGLEVEL >= 7) { log_event(PBSEVENT_SYSTEM, PBS_EVENTCLASS_JOB, "req_statjob", "Successfully returned the status of queued jobs\n"); } return; } /* END req_stat_job_step2() */
void req_stat_svr( struct batch_request *preq) /* ptr to the decoded request */ { svrattrl *pal; struct batch_reply *preply; struct brp_status *pstat; int *nc; static char nc_buf[128]; /* update count and state counts from sv_numjobs and sv_jobstates */ server.sv_attr[(int)SRV_ATR_TotalJobs].at_val.at_long = server.sv_qs.sv_numjobs; server.sv_attr[(int)SRV_ATR_TotalJobs].at_flags |= ATR_VFLAG_SET; update_state_ct( &server.sv_attr[(int)SRV_ATR_JobsByState], server.sv_jobstates, server.sv_jobstbuf); nc = netcounter_get(); sprintf(nc_buf, "%d %d %d", *nc, *(nc + 1), *(nc + 2)); server.sv_attr[(int)SRV_ATR_NetCounter].at_val.at_str = nc_buf; server.sv_attr[(int)SRV_ATR_NetCounter].at_flags |= ATR_VFLAG_SET; /* allocate a reply structure and a status sub-structure */ preply = &preq->rq_reply; preply->brp_choice = BATCH_REPLY_CHOICE_Status; CLEAR_HEAD(preply->brp_un.brp_status); pstat = (struct brp_status *)malloc(sizeof(struct brp_status)); if (pstat == NULL) { reply_free(preply); req_reject(PBSE_SYSTEM, 0, preq, NULL, NULL); return; } CLEAR_LINK(pstat->brp_stlink); strcpy(pstat->brp_objname, server_name); pstat->brp_objtype = MGR_OBJ_SERVER; CLEAR_HEAD(pstat->brp_attr); append_link(&preply->brp_un.brp_status, &pstat->brp_stlink, pstat); /* add attributes to the status reply */ bad = 0; pal = (svrattrl *)GET_NEXT(preq->rq_ind.rq_status.rq_attr); if (status_attrib( pal, svr_attr_def, server.sv_attr, SRV_ATR_LAST, preq->rq_perm, &pstat->brp_attr, &bad, 1)) /* IsOwner == TRUE */ { reply_badattr(PBSE_NOATTR, bad, pal, preq); } else { reply_send(preq); } return; } /* END req_stat_svr() */
void req_stat_que( struct batch_request *preq) /* ptr to the decoded request */ { char *name; pbs_queue *pque = NULL; struct batch_reply *preply; int rc = 0; int type = 0; /* * first, validate the name of the requested object, either * a queue, or null for all queues */ name = preq->rq_ind.rq_status.rq_id; if ((*name == '\0') || (*name == '@')) { type = 1; } else { pque = find_queuebyname(name); if (pque == NULL) { req_reject(PBSE_UNKQUE, 0, preq, NULL, "cannot locate queue"); return; } } preply = &preq->rq_reply; preply->brp_choice = BATCH_REPLY_CHOICE_Status; CLEAR_HEAD(preply->brp_un.brp_status); if (type == 0) { /* get status of the named queue */ rc = status_que(pque, preq, &preply->brp_un.brp_status); } else { /* get status of all queues */ pque = (pbs_queue *)GET_NEXT(svr_queues); while (pque != NULL) { rc = status_que(pque, preq, &preply->brp_un.brp_status); if (rc != 0) { if (rc != PBSE_PERM) break; rc = 0; } pque = (pbs_queue *)GET_NEXT(pque->qu_link); } } if (rc != 0) { reply_free(preply); req_reject(rc, bad, preq, NULL, "status_queue failed"); } else { reply_send(preq); } return; } /* END req_stat_que() */
void free_br( batch_request *preq) { if (preq == NULL) return; 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() */