fix: a race condition in Host.resetAllServices (#182)

Fix a race condition in Host.resetAllServices() caused by calling
channel.close() as a callback of grpc.Client.waitForReady().

The root cause is gRPC subchannel pool state could become into
invalid state when global subchannel pool state is used, and there are
multiple channels used the same subchannel from the pool.

Closing the channel inside of waitForReady callback prevent all the
remaining callbacks for the subchannel from the correct state transition
handling.

Closing the channel inside of waitForReady callback prevents
the subchannel's remaining callbacks from the correct state transition handling.

The fix is to use setImmediate to allow the remaining callbacks for the
subchannel to process the state transition.

src/connection-pool.ts

@@ -215,7 +215,7 @@ export class Host {
         // workaround: https://github.com/grpc/grpc-node/issues/1487
         const state = service.getChannel().getConnectivityState(false);
         if (state === grpc.connectivityState.CONNECTING) {
-          service.waitForReady(Date.now() + 10_00, () => service.close());
+          service.waitForReady(Date.now() + 10_00, () => setImmediate(() => service.close()));
         } else {
           service.close();
         }