ringhash: fix bug where ring hash can be stuck in transient failure d…

…espite having available endpoints (#7364)
grpc · Aug 20, 2024 · ee5cbce · ee5cbce
1 parent 1e2bb71
commit ee5cbce
Show file tree

Hide file tree

Showing 5 changed files with 490 additions and 42 deletions.
diff --git a/internal/testutils/blocking_context_dialer.go b/internal/testutils/blocking_context_dialer.go
@@ -127,3 +127,13 @@ func (h *Hold) Fail(err error) {
 	h.blockCh <- err
 	close(h.blockCh)
 }
+
+// IsStarted returns true if this hold has received a connection attempt.
+func (h *Hold) IsStarted() bool {
+	select {
+	case <-h.waitCh:
+		return true
+	default:
+		return false
+	}
+}
diff --git a/internal/testutils/blocking_context_dialer_test.go b/internal/testutils/blocking_context_dialer_test.go
@@ -59,6 +59,10 @@ func (s) TestBlockingDialer_HoldWaitResume(t *testing.T) {
 	d := NewBlockingDialer()
 	h := d.Hold(lis.Addr().String())
 
+	if h.IsStarted() {
+		t.Fatalf("hold.IsStarted() = true, want false")
+	}
+
 	done := make(chan struct{})
 	ctx, cancel := context.WithTimeout(context.Background(), testTimeout)
 	defer cancel()
@@ -69,13 +73,22 @@ func (s) TestBlockingDialer_HoldWaitResume(t *testing.T) {
 			t.Errorf("BlockingDialer.DialContext() got error: %v, want success", err)
 			return
 		}
+
+		if !h.IsStarted() {
+			t.Errorf("hold.IsStarted() = false, want true")
+		}
 		conn.Close()
 	}()
 
 	// This should block until the goroutine above is scheduled.
 	if !h.Wait(ctx) {
 		t.Fatalf("Timeout while waiting for a connection attempt to %q", h.addr)
 	}
+
+	if !h.IsStarted() {
+		t.Errorf("hold.IsStarted() = false, want true")
+	}
+
 	select {
 	case <-done:
 		t.Fatalf("Expected dialer to be blocked.")