add test for 0 runnable workers and treat as-if overcommit queue

Author: dgrove-oss

src/event/workqueue.c

@@ -226,27 +226,39 @@ _dispatch_workq_monitor_thread_pool(void *context DISPATCH_UNUSED)
 	dispatch_workq_manager_t mgr = _dispatch_workq_get_default_manager();
 	dispatch_queue_t dq = _dispatch_workq_get_default_root_queue();
 	bool work_available = _dispatch_workq_root_queue_has_work(dq);
-	if (work_available) {
-		_dispatch_workq_count_runnable_workers(mgr);
-		int32_t count = _dispatch_pthread_root_queue_thread_pool_size(dq);
-
-		_dispatch_debug("workq: %s is non-empty with pool count %d (%d runnable)",
-				dq->dq_label, count,  mgr->runnable_workers);
-
-		if (mgr->runnable_workers < mgr->target_runnable_workers) {
-			int32_t allowed_over = WORKQ_OVERSUBSCRIBE_FACTOR * mgr->target_runnable_workers;
-			allowed_over = MIN(allowed_over, WORKQ_MAX_TRACKED_WORKERS - mgr->target_runnable_workers);
-			if (count + allowed_over > 0) {
-				_dispatch_debug("workq: %s has count %d; requesting 1 additional worker",
-						dq->dq_label, count);
-				_dispatch_pthread_root_queue_oversubscribe(dq, 1);
-			} else {
-				_dispatch_debug("workq: %s already oversubscribed by %d; taking no action",
-						dq->dq_label, -count);
-			}
+
+	if (!work_available) {
+		_dispatch_debug("workq: %s is empty; doing nothing", dq->dq_label);
+		return;
+	}
+
+	_dispatch_workq_count_runnable_workers(mgr);
+	int32_t count = _dispatch_pthread_root_queue_thread_pool_size(dq);
+
+	_dispatch_debug("workq: %s is non-empty with pool count %d (%d runnable)",
+			dq->dq_label, count,  mgr->runnable_workers);
+
+	if (mgr->runnable_workers < mgr->target_runnable_workers) {
+		// If we are below target there are two cases to consider:
+		//   (a) We are below target, but some workers are still runnable.
+		//       We want to oversubscribe to hit the target, but this
+		//       may be transitory so only go up to a small multiple
+		//       of threads per core.
+		//   (b) We are below target, and no worker is runnable.
+		//       It is likely the program is stalled. Therefore treat
+		//       this as dq was an overcommit queue and create
+		//       another worker unless we have already hit the hard
+		//       limit on the maximum number of workers for dq.
+		int32_t oversubscribe_limit = WORKQ_OVERSUBSCRIBE_FACTOR * mgr->target_runnable_workers;
+		int32_t stalled_limit = WORKQ_MAX_TRACKED_WORKERS - mgr->target_runnable_workers;
+		int32_t limit = mgr->runnable_workers == 0 ? stalled_limit : MIN(stalled_limit, oversubscribe_limit);
+		if (count + limit > 0) {
+			_dispatch_debug("workq: adding 1 additional worker to %s", dq->dq_label);
+			_dispatch_pthread_root_queue_oversubscribe(dq, 1);
+		} else {
+			_dispatch_debug("workq: %s already oversubscribed by %d; taking no action",
+					dq->dq_label, -count);
 		}
-	} else {
-		_dispatch_debug("workq: %s is empty; skipping analysis of /proc", dq->dq_label);
 	}
 }