[threadpool] Remove assertion on RequestWorkerThread failure (#4167)

We observed that this assertion would sometimes be triggered on CI in case of intensive load. This would be possible because we wouldn't necessarily decrease the outstanding_request in a new worker. To decrease resource usage, we also want to limit the amount of new worker we request.

diff --git a/mono/metadata/threadpool.c b/mono/metadata/threadpool.c
index 07db889f26c197b15d71ab64d6992797c6381859..bef76e578159788083c0fbbc1fec9b2ea8b64258 100644 (file)
--- a/mono/metadata/threadpool.c
+++ b/mono/metadata/threadpool.c
@@ -801,18 +801,18 @@ ves_icall_System_Threading_ThreadPool_RequestWorkerThread (void)
        tpdomain->outstanding_request ++;
        g_assert (tpdomain->outstanding_request >= 1);
 
-       mono_refcount_inc (threadpool);
+       domains_unlock ();
 
        COUNTER_ATOMIC (threadpool, counter, {
-               if (!(counter._.starting < 32767 /* G_MAXINT16 */))
-                       g_error ("%s: counter._.starting = %d, but should be < 32767", __func__, counter._.starting);
+               if (counter._.starting == 16)
+                       return TRUE;
 
                counter._.starting ++;
        });
 
-       mono_threadpool_worker_enqueue (threadpool->worker, worker_callback, NULL);
+       mono_refcount_inc (threadpool);
 
-       domains_unlock ();
+       mono_threadpool_worker_enqueue (threadpool->worker, worker_callback, NULL);
 
        return TRUE;
 }