erts: Honor CPU quotas when deciding number of online schedulers

Using more cores than we have quota for is likely to hurt
performance as any busy-waiting we might do (waiting for new jobs
to arrive in a queue, waiting for a lock, etc) will burn the quota
without doing useful work, leaving us with fewer resources to work
with. This is especially bad when the +sbwt option has been used to
increase the wait time.

This commit checks our current CPU quota (if possible) and tries to
limit the number of online schedulers accordingly on startup. The
schedulers can be brought online later on, or overridden with the
`+S` option.

Author: jhogberg

erts/doc/src/erl.xml

@@ -944,14 +944,19 @@
         <c><![CDATA[+S Schedulers:SchedulerOnline]]></c></tag>
       <item>
         <p>Sets the number of scheduler threads to create and scheduler threads
-          to set online. The maximum for both
-          values is 1024. If the Erlang runtime system is able to determine the
-          number of logical processors configured and logical processors
-          available, <c>Schedulers</c> defaults to logical processors
-          configured, and <c>SchedulersOnline</c> defaults to logical processors
-          available; otherwise the default values are 1. <c>Schedulers</c> can
-          be omitted if <c>:SchedulerOnline</c> is not and conversely. The
-          number of schedulers online can be changed at runtime through
+          to set online. The maximum for both values is 1024. If the Erlang
+          runtime system is able to determine the number of logical processors
+          configured and logical processors available, <c>Schedulers</c>
+          defaults to logical processors configured, and
+          <c>SchedulersOnline</c> defaults to logical processors available;
+          otherwise the default values are 1. If the emulator detects that it
+          is subject to a <seealso marker="erlang#system_info_cpu_quota">CPU
+          quota</seealso>, the default value for <c>SchedulersOnline</c> will
+          be limited accordingly.</p>
+        <p>
+          <c>Schedulers</c> can be omitted if <c>:SchedulerOnline</c> is not
+          and conversely. The number of schedulers online can be changed at
+          runtime through
           <seealso marker="erlang#system_flag_schedulers_online">
           <c>erlang:system_flag(schedulers_online,
           SchedulersOnline)</c></seealso>.</p>

erts/doc/src/erlang.xml

@@ -8153,6 +8153,15 @@ Metadata = #{ pid => pid(),
               <seealso marker="#system_info_logical_processors">logical processors
               configured</seealso>.</p>
           </item>
+          <tag><marker id="system_info_cpu_quota"/>
+            <c>cpu_quota</c></tag>
+          <item>
+            <p>Returns the detected CPU quota the emulator is limited by. The
+              return value is an integer saying how many processors' worth of
+              runtime we get (between 1 and the number of logical processors),
+              or the atom <c>unknown</c> if the emulator cannot detect a
+              quota.</p>
+          </item>
           <tag><marker id="system_info_update_cpu_info"/>
             <c>update_cpu_info</c></tag>
           <item>
@@ -8162,8 +8171,9 @@ Metadata = #{ pid => pid(),
               CPU topology</seealso> and the number of logical processors
               <seealso marker="#system_info_logical_processors">configured</seealso>,
               <seealso marker="#system_info_logical_processors_online">online</seealso>,
-              and <seealso marker="#system_info_logical_processors_available">
-              available</seealso>.</p>
+              <seealso marker="#system_info_logical_processors_available">available</seealso>,
+              and <seealso marker="#system_info_cpu_quota">cpu
+              quota</seealso>.</p>
             <p>If the CPU information has changed since the last time
               it was read, the atom <c>changed</c> is returned, otherwise
               the atom <c>unchanged</c>. If the CPU information has changed,

erts/emulator/beam/erl_bif_info.c

@@ -2825,7 +2825,7 @@ BIF_RETTYPE system_info_1(BIF_ALIST_1)
     /* Arguments that are unusual follow ... */
     else if (ERTS_IS_ATOM_STR("logical_processors", BIF_ARG_1)) {
 	int no;
-	erts_get_logical_processors(&no, NULL, NULL);
+	erts_get_logical_processors(&no, NULL, NULL, NULL);
 	if (no > 0)
 	    BIF_RET(make_small((Uint) no));
 	else {
@@ -2835,7 +2835,7 @@ BIF_RETTYPE system_info_1(BIF_ALIST_1)
     }
     else if (ERTS_IS_ATOM_STR("logical_processors_online", BIF_ARG_1)) {
 	int no;
-	erts_get_logical_processors(NULL, &no, NULL);
+	erts_get_logical_processors(NULL, &no, NULL, NULL);
 	if (no > 0)
 	    BIF_RET(make_small((Uint) no));
 	else {
@@ -2845,7 +2845,17 @@ BIF_RETTYPE system_info_1(BIF_ALIST_1)
     }
     else if (ERTS_IS_ATOM_STR("logical_processors_available", BIF_ARG_1)) {
 	int no;
-	erts_get_logical_processors(NULL, NULL, &no);
+	erts_get_logical_processors(NULL, NULL, &no, NULL);
+	if (no > 0)
+	    BIF_RET(make_small((Uint) no));
+	else {
+	    DECL_AM(unknown);
+	    BIF_RET(AM_unknown);
+	}
+    }
+    else if (ERTS_IS_ATOM_STR("cpu_quota", BIF_ARG_1)) {
+	int no;
+	erts_get_logical_processors(NULL, NULL, NULL, &no);
 	if (no > 0)
 	    BIF_RET(make_small((Uint) no));
 	else {

erts/emulator/beam/erl_cpu_topology.c

@@ -1632,21 +1632,23 @@ erts_get_cpu_topology_term(Process *c_p, Eterm which)
 }
 
 static void
-get_logical_processors(int *conf, int *onln, int *avail)
+get_logical_processors(int *conf, int *onln, int *avail, int *quota)
 {
     if (conf)
 	*conf = erts_get_cpu_configured(cpuinfo);
     if (onln)
 	*onln = erts_get_cpu_online(cpuinfo);
     if (avail)
 	*avail = erts_get_cpu_available(cpuinfo);
+    if (quota)
+	*quota = erts_get_cpu_quota(cpuinfo);
 }
 
 void
-erts_get_logical_processors(int *conf, int *onln, int *avail)
+erts_get_logical_processors(int *conf, int *onln, int *avail, int *quota)
 {
     erts_rwmtx_rlock(&cpuinfo_rwmtx);
-    get_logical_processors(conf, onln, avail);
+    get_logical_processors(conf, onln, avail, quota);
     erts_rwmtx_runlock(&cpuinfo_rwmtx);
 }
 
@@ -1655,14 +1657,15 @@ erts_pre_early_init_cpu_topology(int *max_dcg_p,
                                  int *max_rg_p,
 				 int *conf_p,
 				 int *onln_p,
-				 int *avail_p)
+				 int *avail_p,
+				 int *quota_p)
 {
     cpu_groups_maps = NULL;
     no_cpu_groups_callbacks = 0;
     *max_rg_p = ERTS_MAX_READER_GROUPS;
     *max_dcg_p = ERTS_MAX_FLXCTR_GROUPS;
     cpuinfo = erts_cpu_info_create();
-    get_logical_processors(conf_p, onln_p, avail_p);
+    get_logical_processors(conf_p, onln_p, avail_p, quota_p);
 }
 
 void

erts/emulator/beam/erl_cpu_topology.h

@@ -32,7 +32,8 @@ erts_pre_early_init_cpu_topology(int *max_dcg_p,
                                  int *max_rg_p,
 				 int *conf_p,
 				 int *onln_p,
-				 int *avail_p);
+				 int *avail_p,
+				 int *quota_p);
 void
 erts_early_init_cpu_topology(int no_schedulers,
 			     int *max_main_threads_p,
@@ -81,7 +82,7 @@ Eterm erts_set_cpu_topology(Process *c_p, Eterm term);
 Eterm erts_get_cpu_topology_term(Process *c_p, Eterm which);
 
 int erts_update_cpu_info(void);
-void erts_get_logical_processors(int *conf, int *onln, int *avail);
+void erts_get_logical_processors(int *conf, int *onln, int *avail, int *quota);
 
 int erts_sched_bind_atthrcreate_prepare(void);
 int erts_sched_bind_atthrcreate_child(int unbind);

erts/emulator/beam/erl_init.c

@@ -774,6 +774,7 @@ early_init(int *argc, char **argv) /*
     int ncpu;
     int ncpuonln;
     int ncpuavail;
+    int ncpuquota;
     int schdlrs;
     int schdlrs_onln;
     int schdlrs_percentage = 100;
@@ -811,7 +812,8 @@ early_init(int *argc, char **argv) /*
                                      &max_reader_groups,
 				     &ncpu,
 				     &ncpuonln,
-				     &ncpuavail);
+				     &ncpuavail,
+				     &ncpuquota);
 
     ignore_break = 0;
     replace_intr = 0;
@@ -838,9 +840,18 @@ early_init(int *argc, char **argv) /*
      * can initialize the allocators.
      */
     no_schedulers = (Uint) (ncpu > 0 ? ncpu : 1);
-    no_schedulers_online = (ncpuavail > 0
-			    ? ncpuavail
-			    : (ncpuonln > 0 ? ncpuonln : no_schedulers));
+
+    if (ncpuavail > 0) {
+        if (ncpuquota > 0) {
+            no_schedulers_online = MIN(ncpuquota, ncpuavail);
+        } else {
+            no_schedulers_online = ncpuavail;
+        }
+    } else if (ncpuonln > 0) {
+        no_schedulers_online = ncpuonln;
+    } else {
+        no_schedulers_online = no_schedulers;
+    }
 
     schdlrs = no_schedulers;
     schdlrs_onln = no_schedulers_online;