Fix sources of test instability (#368)

- Wait up to 1 second for the server to become healthy before assuming
  it's failed. After the server starts accepting connections, we were
  assuming it should be healthy immediately, but that's not always true.

- Make sure the server finishes shutting down before exiting the CLI,
  because otherwise it will hold open files, and this intermittently
  breaks tests on Windows (which, unlike POSIXy OSes, will refuse to
  delete open files). Closes #338. I hope.

app/app_test.go

@@ -320,8 +320,16 @@ func assertServerHealth(ctx context.Context, t *testing.T, opts sdkclient.Option
 		t.Error(clientErr)
 	}
 
-	if _, err := c.CheckHealth(ctx, nil); err != nil {
-		t.Error(err)
+	// Give the server 1 second to become healthy.
+	for i := 0; i < 10; i++ {
+		_, clientErr = c.CheckHealth(ctx, nil)
+		if clientErr == nil {
+			break
+		}
+		time.Sleep(100 * time.Millisecond)
+	}
+	if clientErr != nil {
+		t.Error(clientErr)
 	}
 
 	// Check for pollers on a system task queue to ensure that the worker service is running.

server/commands.go

@@ -337,6 +337,7 @@ func NewServerCommands(defaultCfg *sconfig.Config) []*cli.Command {
 				if err := s.Start(); err != nil {
 					return cli.Exit(fmt.Sprintf("Unable to start server. Error: %v", err), 1)
 				}
+				s.Stop()
 				return cli.Exit("All services are stopped.", 0)
 			},
 		},