Example #1
0
// start dials the remote addr and commences gossip once connected. Upon exit,
// the client is sent on the disconnected channel. This method starts client
// processing in a goroutine and returns immediately.
func (c *client) start(
	g *Gossip,
	disconnected chan *client,
	rpcCtx *rpc.Context,
	stopper *stop.Stopper,
	nodeID roachpb.NodeID,
	breaker *circuit.Breaker,
) {
	stopper.RunWorker(func() {
		ctx, cancel := context.WithCancel(c.AnnotateCtx(context.Background()))
		var wg sync.WaitGroup
		defer func() {
			// This closes the outgoing stream, causing any attempt to send or
			// receive to return an error.
			//
			// Note: it is still possible for incoming gossip to be processed after
			// this point.
			cancel()

			// The stream is closed, but there may still be some incoming gossip
			// being processed. Wait until that is complete to avoid racing the
			// client's removal against the discovery of its remote's node ID.
			wg.Wait()
			disconnected <- c
		}()

		consecFailures := breaker.ConsecFailures()
		var stream Gossip_GossipClient
		if err := breaker.Call(func() error {
			// Note: avoid using `grpc.WithBlock` here. This code is already
			// asynchronous from the caller's perspective, so the only effect of
			// `WithBlock` here is blocking shutdown - at the time of this writing,
			// that ends ups up making `kv` tests take twice as long.
			conn, err := rpcCtx.GRPCDial(c.addr.String())
			if err != nil {
				return err
			}
			if stream, err = NewGossipClient(conn).Gossip(ctx); err != nil {
				return err
			}
			return c.requestGossip(g, stream)
		}, 0); err != nil {
			if consecFailures == 0 {
				log.Warningf(ctx, "node %d: failed to start gossip client: %s", nodeID, err)
			}
			return
		}

		// Start gossiping.
		log.Infof(ctx, "node %d: started gossip client to %s", nodeID, c.addr)
		if err := c.gossip(ctx, g, stream, stopper, &wg); err != nil {
			if !grpcutil.IsClosedConnection(err) {
				g.mu.Lock()
				if c.peerID != 0 {
					log.Infof(ctx, "node %d: closing client to node %d (%s): %s", nodeID, c.peerID, c.addr, err)
				} else {
					log.Infof(ctx, "node %d: closing client to %s: %s", nodeID, c.addr, err)
				}
				g.mu.Unlock()
			}
		}
	})
}