BF: CS-1054 with pe setting ign_sreq_on_mhost=true a globally defined resource which is requested for slave scope is not booked into the resource diagram

jgabler-hpc · jgabler-hpc · commit ebeca95576ae · 2025-03-03T10:46:54.000+01:00
diff --git a/source/daemons/qmaster/setup_qmaster.cc b/source/daemons/qmaster/setup_qmaster.cc
@@ -1278,6 +1278,7 @@ static void debit_all_jobs_from_qs() {
          const lListElem *pe = lGetObject(jatep, JAT_pe_object);
 
          /* don't look at states - we only trust in "granted destin. ident. list" */
+         bool do_per_global_host_booking = true;
          const char *last_hostname = nullptr;
          for_each_ep(gdi, lGetList(jatep, JAT_granted_destin_identifier_list)) {
             u_long32 ar_id = lGetUlong(jep, JB_ar);
@@ -1300,7 +1301,7 @@ static void debit_all_jobs_from_qs() {
                debit_host_consumable(jep, jatep, pe,
                                      host_list_locate(*ocs::DataStore::get_master_list(SGE_TYPE_EXECHOST),
                                                       SGE_GLOBAL_NAME), master_centry_list, slots,
-                                     master_task, do_per_host_booking, nullptr);
+                                     master_task, do_per_global_host_booking, nullptr);
                debit_host_consumable(jep, jatep, pe,
                                      host_list_locate(*ocs::DataStore::get_master_list(SGE_TYPE_EXECHOST),
                                                       lGetHost(qep, QU_qhostname)), master_centry_list,
@@ -1329,6 +1330,7 @@ static void debit_all_jobs_from_qs() {
                }
             }
             master_task = false;
+            do_per_global_host_booking = false;
          }
       }
    }
diff --git a/source/daemons/qmaster/sge_centry_qmaster.cc b/source/daemons/qmaster/sge_centry_qmaster.cc
@@ -528,6 +528,7 @@ void centry_redebit_consumables(const lList *centries, u_long64 gdi_version) {
          const char *last_hostname = nullptr;
          const lListElem *pe = lGetObject(jatep, JAT_pe_object);
 
+         bool do_per_global_host_booking = true;
          for_each_ep(gdil, lGetList(jatep, JAT_granted_destin_identifier_list)) {
             int qslots;
 
@@ -539,6 +540,9 @@ void centry_redebit_consumables(const lList *centries, u_long64 gdi_version) {
 
             qslots = lGetUlong(gdil, JG_slots);
 
+            debit_host_consumable(jep, jatep, pe, host_list_locate(master_ehost_list, SGE_GLOBAL_NAME),
+                                  master_centry_list, slots, master_task, do_per_global_host_booking, nullptr);
+
             bool do_per_host_booking = host_do_per_host_booking(&last_hostname, lGetHost(gdil, JG_qhostname));
             debit_host_consumable(jep, jatep, pe, host_list_locate(master_ehost_list,
                                                                         lGetHost(qep, QU_qhostname)),
@@ -548,9 +552,8 @@ void centry_redebit_consumables(const lList *centries, u_long64 gdi_version) {
                                        nullptr);
             slots += qslots;
             master_task = false;
+            do_per_global_host_booking = false;
          }
-         debit_host_consumable(jep, jatep, pe, host_list_locate(master_ehost_list, SGE_GLOBAL_NAME),
-                               master_centry_list, slots, true, true, nullptr);
       }
    }
 
diff --git a/source/daemons/qmaster/sge_give_jobs.cc b/source/daemons/qmaster/sge_give_jobs.cc
@@ -965,6 +965,7 @@ sge_commit_job(lListElem *jep, lListElem *jatep, lListElem *jr, sge_commit_mode_
                                   MSG_LOG_SENT2EXECD);
 
          global_host_ep = host_list_locate(master_exechost_list, "global");
+         bool do_per_global_host_booking = true;
          const char *last_hostname = nullptr;
          const lListElem *gdil_ep;
          const lList *gdil = lGetList(jatep, JAT_granted_destin_identifier_list);
@@ -1009,7 +1010,7 @@ sge_commit_job(lListElem *jep, lListElem *jatep, lListElem *jr, sge_commit_mode_
 
                /* debit consumable resources */
                if (debit_host_consumable(jep, jatep, pe, global_host_ep, master_centry_list, tmp_slot, master_task,
-                                         do_per_host_booking,
+                                         do_per_global_host_booking,
                                          nullptr) > 0) {
                   /* this info is not spooled */
                   sge_add_event(0, sgeE_EXECHOST_MOD, 0, 0,
@@ -1056,6 +1057,7 @@ sge_commit_job(lListElem *jep, lListElem *jatep, lListElem *jr, sge_commit_mode_
                }
             }
             master_task = false;
+            do_per_global_host_booking = false;
          }
 
          lSetUlong64(jatep, JAT_wallclock_limit, task_wallclock);
@@ -1535,6 +1537,7 @@ sge_clear_granted_resources(lListElem *job, lListElem *ja_task, int incslots, mo
    cqueue_list_x_on_subordinate_gdil(master_cqueue_list, false, gdi_list, monitor, gdi_session);
 
    global_host_ep = host_list_locate(master_exechost_list, SGE_GLOBAL_NAME);
+   bool do_per_global_host_booking = true;
 
    const char *pe_name = lGetString(ja_task, JAT_granted_pe);
    lListElem *pe = lGetObject(ja_task, JAT_pe_object);
@@ -1566,7 +1569,7 @@ sge_clear_granted_resources(lListElem *job, lListElem *ja_task, int incslots, mo
 
             /* undebit consumable resources */
             if (debit_host_consumable(job, ja_task, pe, global_host_ep, master_centry_list, -tmp_slot, master_task,
-                                      do_per_host_booking, nullptr) > 0) {
+                                      do_per_global_host_booking, nullptr) > 0) {
                /* this info is not spooled */
                sge_add_event(0, sgeE_EXECHOST_MOD, 0, 0, SGE_GLOBAL_NAME, nullptr, nullptr, global_host_ep, gdi_session);
                ocs::ReportingFileWriter::create_host_consumable_records(&answer_list, global_host_ep, job, now);
@@ -1613,6 +1616,7 @@ sge_clear_granted_resources(lListElem *job, lListElem *ja_task, int incslots, mo
          }
       }
       master_task = false;
+      do_per_global_host_booking = false;
    }
 
    /* free granted resources of the parallel environment */
diff --git a/source/libs/sched/debit.cc b/source/libs/sched/debit.cc
@@ -282,6 +282,7 @@ debit_job_from_hosts(lListElem *job, lListElem *ja_task, const lListElem *pe, lL
    so = lParseSortOrderVarArg(lGetListDescr(host_list), "%I+", EH_sort_value);
 
    global = host_list_locate(host_list, "global");
+   bool do_per_global_host_booking = true;
 
    load_formula = sconf_get_load_formula();
 
@@ -304,9 +305,10 @@ debit_job_from_hosts(lListElem *job, lListElem *ja_task, const lListElem *pe, lL
       }
 
       debit_host_consumable(job, ja_task, pe, host_list_locate(host_list, SGE_GLOBAL_NAME), centry_list, slots,
-                            is_master_task, do_per_host_booking, nullptr);
+                            is_master_task, do_per_global_host_booking, nullptr);
       debit_host_consumable(job, ja_task, pe, hep, centry_list, slots, is_master_task, do_per_host_booking, nullptr);
       is_master_task = false;
+      do_per_global_host_booking = false;
 
       /* compute new combined load for this host and put it into the host */
       old_sort_value = lGetDouble(hep, EH_sort_value);
diff --git a/source/libs/sched/sge_resource_utilization.cc b/source/libs/sched/sge_resource_utilization.cc
@@ -269,7 +269,6 @@ int utilization_add(lListElem *cr, u_long64 start_time, u_long64 duration, doubl
    lList *resource_diagram;
    lListElem *thiz, *prev, *start, *end;
    const char *name = lGetString(cr, RUE_name);
-   char level_char = CENTRY_LEVEL_TO_CHAR(level);
    u_long64 end_time;
    int nm;
    double util_prev;
@@ -299,7 +298,7 @@ int utilization_add(lListElem *cr, u_long64 start_time, u_long64 duration, doubl
    end_time = utilization_endtime(start_time, duration);
 
    serf_record_entry(job_id, ja_taskid, (type!=nullptr)?type:"<unknown>", start_time, end_time,
-         level_char, object_name, name, utilization);
+                     level, object_name, name, utilization);
 
    /* ensure resource diagram is initialized */
    if (resource_diagram == nullptr) {
@@ -705,14 +704,10 @@ int add_job_utilization(const sge_assignment_t *a, const char *type, bool for_jo
                a->job_id, a->ja_task_id, PE_TAG, lGetString(a->pe, PE_name), type, for_job_scheduling, false);
       }
 
-      /* global */
-      rc_add_job_utilization(a->job, a->pe, a->ja_task_id, type, a->gep, a->centry_list, a->slots,
-                             EH_consumable_config_list, EH_resource_utilization, SGE_GLOBAL_NAME,
-                             a->start, a->duration, GLOBAL_TAG, for_job_scheduling, true, true);
-
       bool is_master_task = true;
       const lListElem *gdil_ep;
       const char *last_eh_name = nullptr;
+      bool do_per_global_host_booking = true;
       for_each_ep(gdil_ep, a->gdil) {
          int slots = lGetUlong(gdil_ep, JG_slots);
          const char *eh_name = lGetHost(gdil_ep, JG_qhostname);
@@ -723,14 +718,20 @@ int add_job_utilization(const sge_assignment_t *a, const char *type, bool for_jo
          const lListElem *rqs = nullptr;
          bool do_per_host_booking = host_do_per_host_booking(&last_eh_name, eh_name);
 
-         /* hosts */
+         // global
+         // we really need to do it per gdil_ep, because we have to consider is_master_task and ign_sreq_on_mhost
+         rc_add_job_utilization(a->job, a->pe, a->ja_task_id, type, a->gep, a->centry_list, slots,
+                                EH_consumable_config_list, EH_resource_utilization, SGE_GLOBAL_NAME,
+                                a->start, a->duration, GLOBAL_TAG, for_job_scheduling, is_master_task, do_per_global_host_booking);
+
+         // host
          if ((hep = host_list_locate(a->host_list, eh_name)) != nullptr) {
             rc_add_job_utilization(a->job, a->pe, a->ja_task_id, type, hep, a->centry_list, slots,
                                    EH_consumable_config_list, EH_resource_utilization, eh_name, a->start,
                                    a->duration, HOST_TAG, for_job_scheduling, is_master_task, do_per_host_booking);
          }
 
-         /* queues */
+         // queue
          if ((qep = qinstance_list_locate2(a->queue_list, qname)) != nullptr) {
             /* 
              * The nullptr case happens in case of queues that were sorted out b/c they
@@ -767,6 +768,7 @@ int add_job_utilization(const sge_assignment_t *a, const char *type, bool for_jo
 
          sge_free(&queue);
          is_master_task = false;
+         do_per_global_host_booking = false;
       }
 
       sge_dstring_free(&rue_name);
diff --git a/source/libs/sched/sge_serf.cc b/source/libs/sched/sge_serf.cc
@@ -39,6 +39,7 @@
 
 #include "cull/cull.h"
 
+#include "sgeobj/sge_centry.h"
 #include "sgeobj/sge_schedd_conf.h"
 
 #include "sge_serf.h"
@@ -87,7 +88,7 @@ void serf_init(record_schedule_entry_func_t write, new_schedule_func_t newline)
 *
 *  SYNOPSIS
 *     void serf_record_entry(u_long32 job_id, u_long32 ja_taskid, const char 
-*     *state, u_long32 start_time, u_long32 end_time, char level_char, const 
+*     *state, u_long32 start_time, u_long32 end_time, u_long32 level, const
 *     char *object_name, const char *name, double utilization) 
 *
 *  FUNCTION
@@ -132,11 +133,13 @@ void serf_init(record_schedule_entry_func_t write, new_schedule_func_t newline)
 *     MT-NOTE:     MT safety of registered recording function
 *******************************************************************************/
 void serf_record_entry(u_long32 job_id, u_long32 ja_taskid,
-      const char *type, u_long64 start_time, u_long64 end_time, char level_char,
-      const char *object_name, const char *name, double utilization)
+                       const char *type, u_long64 start_time, u_long64 end_time, u_long32 level,
+                       const char *object_name, const char *name, double utilization)
 {
    DENTER(TOP_LAYER);
 
+   char level_char = CENTRY_LEVEL_TO_CHAR(level);
+
    /* human-readable format */
    if (DPRINTF_IS_ACTIVE) {
       DSTRING_STATIC(dstr_s, 64);
diff --git a/source/libs/sched/sge_serf.h b/source/libs/sched/sge_serf.h
@@ -77,7 +77,7 @@ typedef void (*new_schedule_func_t)();
 
 void serf_init(record_schedule_entry_func_t, new_schedule_func_t);
 void serf_record_entry(u_long32 job_id, u_long32 ja_taskid,
-      const char *state, u_long64 start_time, u_long64 end_time, char level_char,
-            const char *object_name, const char *name, double utilization);
+                       const char *state, u_long64 start_time, u_long64 end_time, u_long32 level,
+                       const char *object_name, const char *name, double utilization);
 void serf_new_interval(u_long64 time);
 void serf_exit();

Original file line number	Diff line number	Diff line change
`@@ -1278,6 +1278,7 @@ static void debit_all_jobs_from_qs() {`
`1278`	`1278`	`const lListElem *pe = lGetObject(jatep, JAT_pe_object);`
`1279`	`1279`
`1280`	`1280`	`/* don't look at states - we only trust in "granted destin. ident. list" */`
	`1281`	`+ bool do_per_global_host_booking = true;`
`1281`	`1282`	`const char *last_hostname = nullptr;`
`1282`	`1283`	`for_each_ep(gdi, lGetList(jatep, JAT_granted_destin_identifier_list)) {`
`1283`	`1284`	`u_long32 ar_id = lGetUlong(jep, JB_ar);`
`@@ -1300,7 +1301,7 @@ static void debit_all_jobs_from_qs() {`
`1300`	`1301`	`debit_host_consumable(jep, jatep, pe,`
`1301`	`1302`	`host_list_locate(*ocs::DataStore::get_master_list(SGE_TYPE_EXECHOST),`
`1302`	`1303`	`SGE_GLOBAL_NAME), master_centry_list, slots,`
`1303`		`- master_task, do_per_host_booking, nullptr);`
	`1304`	`+ master_task, do_per_global_host_booking, nullptr);`
`1304`	`1305`	`debit_host_consumable(jep, jatep, pe,`
`1305`	`1306`	`host_list_locate(*ocs::DataStore::get_master_list(SGE_TYPE_EXECHOST),`
`1306`	`1307`	`lGetHost(qep, QU_qhostname)), master_centry_list,`
`@@ -1329,6 +1330,7 @@ static void debit_all_jobs_from_qs() {`
`1329`	`1330`	`}`
`1330`	`1331`	`}`
`1331`	`1332`	`master_task = false;`
	`1333`	`+ do_per_global_host_booking = false;`
`1332`	`1334`	`}`
`1333`	`1335`	`}`
`1334`	`1336`	`}`