Make sure scheduler threads don't ACK the quiescence protocol CNF messages if they have an actor waiting to be run

.release-notes/4583.md

@@ -0,0 +1,3 @@
+## Make sure scheduler threads don't ACK the quiescence protocol CNF messages if they have an actor waiting to be run
+
+Prior to this, the pinned actor thread could cause early termination/quiescence of a pony program if there were only pinned actors active. This change fixes the issue to ensure that pony programs with only pinned actors active will no longer terminate too early.

src/libponyrt/sched/scheduler.c

@@ -463,7 +463,7 @@ static void handle_sched_unblock(scheduler_t* sched)
   pony_assert(sched->block_count <= scheduler_count);
 }
 
-static bool read_msg(scheduler_t* sched)
+static bool read_msg(scheduler_t* sched, pony_actor_t* actor)
 {
 #ifdef USE_RUNTIMESTATS
     uint64_t used_cpu = ponyint_sched_cpu_used(&sched->ctx);
@@ -506,8 +506,11 @@ static bool read_msg(scheduler_t* sched)
       {
         pony_assert(PONY_UNKNOWN_SCHEDULER_INDEX != sched->index);
 
-        // Echo the token back as ACK(token).
-        send_msg(sched->index, 0, SCHED_ACK, m->i);
+        if(NULL == actor)
+        {
+          // Echo the token back as ACK(token) only if we don't have an actor to run.
+          send_msg(sched->index, 0, SCHED_ACK, m->i);
+        }
         break;
       }
 
@@ -758,7 +761,7 @@ static pony_actor_t* suspend_scheduler(scheduler_t* sched,
       if(actor != NULL)
         break;
 
-      if(read_msg(sched))
+      if(read_msg(sched, actor))
       {
         // An actor was unmuted and added to our run queue. Pop it and return.
         // Effectively, we are "stealing" from ourselves. We need to verify that
@@ -925,7 +928,7 @@ static pony_actor_t* steal(scheduler_t* sched)
 
     uint64_t tsc2 = ponyint_cpu_tick();
 
-    if(read_msg(sched))
+    if(read_msg(sched, actor))
     {
       // An actor was unmuted and added to our run queue. Pop it and return.
       // Effectively, we are "stealing" from ourselves. We need to verify that
@@ -1154,7 +1157,7 @@ static void run(scheduler_t* sched)
     // In response to reading a message, we might have unmuted an actor and
     // added it back to our queue. if we don't have an actor to run, we want
     // to pop from our queue to check for a recently unmuted actor
-    if(read_msg(sched) && actor == NULL)
+    if(read_msg(sched, actor) && actor == NULL)
     {
       actor = pop_global(sched);
     }
@@ -1419,7 +1422,7 @@ static void run_pinned_actors()
     // scheduler should be handled by the pinned actor scheduler but for the moment
     // that is how things work and the actor will eventually come back to this thread
     // to be run anyways.
-    read_msg(sched);
+    read_msg(sched, actor);
 
     // Termination. all the normal scheduler threads have decided there is no
     // more work to do so we can shutdown

test/full-program-tests/regression-4582/expected-exit-code.txt

@@ -0,0 +1 @@
+100

test/full-program-tests/regression-4582/main.pony

@@ -0,0 +1,40 @@
+use @pony_exitcode[None](code: I32)
+use @usleep[I32](micros: U32) if not windows
+use @Sleep[None](millis: U32) if windows
+use "actor_pinning"
+
+actor Main
+  let _env: Env
+  let _auth: PinUnpinActorAuth
+
+  new create(env: Env) =>
+    _env = env
+    _auth = PinUnpinActorAuth(env.root)
+    ActorPinning.request_pin(_auth)
+    check_pinned()
+
+  be check_pinned() =>
+    if ActorPinning.is_successfully_pinned(_auth) then
+      do_stuff(100)
+    else
+      check_pinned()
+    end
+
+  be do_stuff(i: I32) =>
+    // sleep for a while so that the quiescence CNF/ACK protocol can happen
+    ifdef windows then
+      @Sleep(10)
+    else
+      @usleep(10000)
+    end
+    if i < 0 then
+      // set the exit code if this behavior has been run enough times
+      // issue 4582 identified early quiescence/termination if only pinned
+      // actors remained active
+      @pony_exitcode(100)
+    else
+      do_stuff(i - 1)
+    end
+
+  be done() =>
+    ActorPinning.request_unpin(_auth)