replication/driver: automatic retries on connectivity-related errors

rpc/dataconn/dataconn_client.go

@@ -213,3 +213,23 @@ func (c *Client) ReqRecv(ctx context.Context, req *pdu.ReceiveReq, streamCopier
 
 	return res.res, cause
 }
+
+
+func (c *Client) ReqPing(ctx context.Context, req *pdu.PingReq) (*pdu.PingRes, error) {
+	conn, err := c.getWire(ctx)
+	if err != nil {
+		return nil, err
+	}
+			defer c.putWire(conn)
+
+	if err := c.send(ctx, conn, EndpointPing, req, nil); err != nil {
+		return nil, err
+	}
+
+	var res pdu.PingRes
+	if err := c.recv(ctx, conn, &res); err != nil {
+		return nil, err
+	}
+
+	return &res, nil
+}

rpc/dataconn/dataconn_server.go

@@ -25,6 +25,8 @@ type Handler interface {
 	// It is guaranteed that Server calls Receive with a stream that holds the IdleConnTimeout
 	// configured in ServerConfig.Shared.IdleConnTimeout.
 	Receive(ctx context.Context, r *pdu.ReceiveReq, receive zfs.StreamCopier) (*pdu.ReceiveRes, error)
+	// PingDataconn handles a PingReq
+	PingDataconn(ctx context.Context, r *pdu.PingReq) (*pdu.PingRes, error)
 }
 
 type Logger = logger.Logger
@@ -125,6 +127,13 @@ func (s *Server) serveConn(nc *transport.AuthConn) {
 			return
 		}
 		res, handlerErr = s.h.Receive(ctx, &req, &streamCopier{streamConn: c, closeStreamOnClose: false}) // SHADOWING
+	case EndpointPing:
+		var req pdu.PingReq
+		if err := proto.Unmarshal(reqStructured, &req); err != nil {
+			s.log.WithError(err).Error("cannot unmarshal ping request")
+			return
+		}
+		res, handlerErr = s.h.PingDataconn(ctx, &req) // SHADOWING
 	default:
 		s.log.WithField("endpoint", endpoint).Error("unknown endpoint")
 		handlerErr = fmt.Errorf("requested endpoint does not exist")

rpc/dataconn/dataconn_shared.go

@@ -10,6 +10,7 @@ import (
 )
 
 const (
+	EndpointPing string = "/v1/ping"
 	EndpointSend string = "/v1/send"
 	EndpointRecv string = "/v1/recv"
 )

rpc/dataconn/microbenchmark/microbenchmark.go

@@ -77,6 +77,12 @@ func (devNullHandler) Receive(ctx context.Context, r *pdu.ReceiveReq, stream zfs
 	return &res, err
 }
 
+func (devNullHandler) PingDataconn(ctx context.Context, r *pdu.PingReq) (*pdu.PingRes, error) {
+	return &pdu.PingRes{
+		Echo: r.GetMessage(),
+	}, nil
+}
+
 type tcpConnecter struct {
 	addr string
 }