x/ref/runtime/internal/rpc: fix proxymgr shutdown race (#410)

Ensure that the proxymanager's monitoring goroutine terminates before the proxymanager itself terminates. This avoids a race between the monitoring goroutine issuing new RPCs to the mounttable and the runtime being shutdown. This fixes #409.

x/ref/runtime/internal/rpc/proxymgr.go

@@ -244,7 +244,8 @@ func drainNotifyChan(ch chan struct{}) {
 	}
 }
 
-func (pm *proxyManager) watchForChanges(ctx *context.T, ch chan struct{}) {
+func (pm *proxyManager) watchForChanges(ctx *context.T, ch, doneCh chan struct{}) {
+	defer close(doneCh)
 	for {
 		select {
 		case <-ctx.Done():
@@ -260,12 +261,17 @@ func (pm *proxyManager) watchForChanges(ctx *context.T, ch chan struct{}) {
 
 func (pm *proxyManager) manageProxyConnections(ctx *context.T) {
 	notifyCh := make(chan struct{}, 10)
+	doneCh := make(chan struct{})
+
+	defer func() {
+		<-doneCh
+	}()
 	pm.updateAvailableProxies(ctx)
 	// Watch for changes in the set of available proxies so that for the
 	// 'all' policy, the server will connect to new proxies as they appear.
 	// For other policies reconnection may be little faster since the
 	// new set of proxies is already available.
-	go pm.watchForChanges(ctx, notifyCh)
+	go pm.watchForChanges(ctx, notifyCh, doneCh)
 
 	for {
 		select {