diff --git a/go/vt/vtgate/fakerpcvtgateconn/conn.go b/go/vt/vtgate/fakerpcvtgateconn/conn.go
index 8f25b26ca8..2d6409c215 100644
--- a/go/vt/vtgate/fakerpcvtgateconn/conn.go
+++ b/go/vt/vtgate/fakerpcvtgateconn/conn.go
@@ -215,6 +215,11 @@ func (conn *FakeVTGateConn) StreamExecute(ctx context.Context, query string, bin
 	return resultChan, func() error { return nil }, nil
 }
 
+// StreamExecute2 please see vtgateconn.Impl.StreamExecute2
+func (conn *FakeVTGateConn) StreamExecute2(ctx context.Context, query string, bindVars map[string]interface{}, tabletType topo.TabletType) (<-chan *mproto.QueryResult, vtgateconn.ErrFunc, error) {
+	panic("not implemented")
+}
+
 // StreamExecuteShard please see vtgateconn.Impl.StreamExecuteShard
 func (conn *FakeVTGateConn) StreamExecuteShard(ctx context.Context, query string, keyspace string, shards []string, bindVars map[string]interface{}, tabletType topo.TabletType) (<-chan *mproto.QueryResult, vtgateconn.ErrFunc, error) {
 	panic("not implemented")
diff --git a/go/vt/vtgate/gorpcvtgateconn/conn.go b/go/vt/vtgate/gorpcvtgateconn/conn.go
index 7824894e4a..4e73b29761 100644
--- a/go/vt/vtgate/gorpcvtgateconn/conn.go
+++ b/go/vt/vtgate/gorpcvtgateconn/conn.go
@@ -257,6 +257,19 @@ func (conn *vtgateConn) StreamExecute(ctx context.Context, query string, bindVar
 	return sendStreamResults(c, sr)
 }
 
+func (conn *vtgateConn) StreamExecute2(ctx context.Context, query string, bindVars map[string]interface{}, tabletType topo.TabletType) (<-chan *mproto.QueryResult, vtgateconn.ErrFunc, error) {
+	req := &proto.Query{
+		CallerID:      getEffectiveCallerID(ctx),
+		Sql:           query,
+		BindVariables: bindVars,
+		TabletType:    tabletType,
+		Session:       nil,
+	}
+	sr := make(chan *proto.QueryResult, 10)
+	c := conn.rpcConn.StreamGo("VTGate.StreamExecute2", req, sr)
+	return sendStreamResults(c, sr)
+}
+
 func (conn *vtgateConn) StreamExecuteShard(ctx context.Context, query string, keyspace string, shards []string, bindVars map[string]interface{}, tabletType topo.TabletType) (<-chan *mproto.QueryResult, vtgateconn.ErrFunc, error) {
 	req := &proto.QueryShard{
 		CallerID:      getEffectiveCallerID(ctx),
@@ -304,13 +317,29 @@ func (conn *vtgateConn) StreamExecuteKeyspaceIds(ctx context.Context, query stri
 
 func sendStreamResults(c *rpcplus.Call, sr chan *proto.QueryResult) (<-chan *mproto.QueryResult, vtgateconn.ErrFunc, error) {
 	srout := make(chan *mproto.QueryResult, 1)
+	var vtErr error
 	go func() {
 		defer close(srout)
 		for r := range sr {
-			srout <- r.Result
+			vtErr = vterrors.FromRPCError(r.Err)
+			// If we get a QueryResult with an RPCError, that was an extra QueryResult sent by
+			// the server specifically to indicate an error, and we shouldn't surface it to clients.
+			if vtErr == nil {
+				srout <- r.Result
+			}
 		}
 	}()
-	return srout, func() error { return c.Error }, nil
+	// errFunc will return either an RPC-layer error or an application error, if one exists.
+	// It will only return the most recent application error (i.e, from the QueryResult that
+	// most recently contained an error). It will prioritize an RPC-layer error over an apperror,
+	// if both exist.
+	errFunc := func() error {
+		if c.Error != nil {
+			return c.Error
+		}
+		return vtErr
+	}
+	return srout, errFunc, nil
 }
 
 func (conn *vtgateConn) Begin(ctx context.Context) (interface{}, error) {
diff --git a/go/vt/vtgate/gorpcvtgateservice/server.go b/go/vt/vtgate/gorpcvtgateservice/server.go
index 7c5dbba718..f8f74c2182 100644
--- a/go/vt/vtgate/gorpcvtgateservice/server.go
+++ b/go/vt/vtgate/gorpcvtgateservice/server.go
@@ -152,6 +152,32 @@ func (vtg *VTGate) StreamExecute(ctx context.Context, request *proto.Query, send
 	})
 }
 
+// StreamExecute2 is the RPC version of vtgateservice.VTGateService method
+func (vtg *VTGate) StreamExecute2(ctx context.Context, request *proto.Query, sendReply func(interface{}) error) (err error) {
+	defer vtg.server.HandlePanic(&err)
+	ctx = callerid.NewContext(ctx,
+		callerid.GoRPCEffectiveCallerID(request.CallerID),
+		callerid.NewImmediateCallerID("gorpc client"))
+	vtgErr := vtg.server.StreamExecute(ctx, request, func(value *proto.QueryResult) error {
+		return sendReply(value)
+	})
+	if vtgErr == nil {
+		return nil
+	}
+	if *vtgate.RPCErrorOnlyInReply {
+		// If there was an app error, send a QueryResult back with it.
+		qr := new(proto.QueryResult)
+		vtgate.AddVtGateErrorToQueryResult(vtgErr, qr)
+		// Sending back errors this way is not backwards compatible. If a (new) server sends an additional
+		// QueryResult with an error, and the (old) client doesn't know how to read it, it will cause
+		// problems where the client will get out of sync with the number of QueryResults sent.
+		// That's why this the error is only sent this way when the --rpc_errors_only_in_reply flag is set
+		// (signalling that all clients are able to handle new-style errors).
+		return sendReply(qr)
+	}
+	return vtgErr
+}
+
 // StreamExecuteShard is the RPC version of vtgateservice.VTGateService method
 func (vtg *VTGate) StreamExecuteShard(ctx context.Context, request *proto.QueryShard, sendReply func(interface{}) error) (err error) {
 	defer vtg.server.HandlePanic(&err)
diff --git a/go/vt/vtgate/grpcvtgateconn/conn.go b/go/vt/vtgate/grpcvtgateconn/conn.go
index c75e7de439..52b561d91f 100644
--- a/go/vt/vtgate/grpcvtgateconn/conn.go
+++ b/go/vt/vtgate/grpcvtgateconn/conn.go
@@ -244,6 +244,10 @@ func (conn *vtgateConn) StreamExecute(ctx context.Context, query string, bindVar
 	}, nil
 }
 
+func (conn *vtgateConn) StreamExecute2(ctx context.Context, query string, bindVars map[string]interface{}, tabletType topo.TabletType) (<-chan *mproto.QueryResult, vtgateconn.ErrFunc, error) {
+	return conn.StreamExecute(ctx, query, bindVars, tabletType)
+}
+
 func (conn *vtgateConn) StreamExecuteShard(ctx context.Context, query string, keyspace string, shards []string, bindVars map[string]interface{}, tabletType topo.TabletType) (<-chan *mproto.QueryResult, vtgateconn.ErrFunc, error) {
 	req := &pb.StreamExecuteShardsRequest{
 		CallerId:   callerid.EffectiveCallerIDFromContext(ctx),
diff --git a/go/vt/vtgate/vtgateconn/vtgateconn.go b/go/vt/vtgate/vtgateconn/vtgateconn.go
index 86d25980ba..957e1a6028 100644
--- a/go/vt/vtgate/vtgateconn/vtgateconn.go
+++ b/go/vt/vtgate/vtgateconn/vtgateconn.go
@@ -101,6 +101,15 @@ func (conn *VTGateConn) StreamExecute(ctx context.Context, query string, bindVar
 	return conn.impl.StreamExecute(ctx, query, bindVars, tabletType)
 }
 
+// StreamExecute2 executes a streaming query on vtgate. It returns a
+// channel, an ErrFunc, and error. First check the error. Then you can
+// pull values from the channel till it's closed. Following this, you
+// can call ErrFunc to see if the stream ended normally or due to a
+// failure.
+func (conn *VTGateConn) StreamExecute2(ctx context.Context, query string, bindVars map[string]interface{}, tabletType topo.TabletType) (<-chan *mproto.QueryResult, ErrFunc, error) {
+	return conn.impl.StreamExecute2(ctx, query, bindVars, tabletType)
+}
+
 // StreamExecuteShard executes a streaming query on vtgate, on a set
 // of shards.  It returns a channel, an ErrFunc, and error. First
 // check the error. Then you can pull values from the channel till
@@ -322,6 +331,9 @@ type Impl interface {
 	// StreamExecute executes a streaming query on vtgate.
 	StreamExecute(ctx context.Context, query string, bindVars map[string]interface{}, tabletType topo.TabletType) (<-chan *mproto.QueryResult, ErrFunc, error)
 
+	// StreamExecute2 executes a streaming query on vtgate.
+	StreamExecute2(ctx context.Context, query string, bindVars map[string]interface{}, tabletType topo.TabletType) (<-chan *mproto.QueryResult, ErrFunc, error)
+
 	// StreamExecuteShard executes a streaming query on vtgate, on a set of shards.
 	StreamExecuteShard(ctx context.Context, query string, keyspace string, shards []string, bindVars map[string]interface{}, tabletType topo.TabletType) (<-chan *mproto.QueryResult, ErrFunc, error)
 
diff --git a/go/vt/vtgate/vtgateconntest/client.go b/go/vt/vtgate/vtgateconntest/client.go
index 7de694f053..1c0c9e93f0 100644
--- a/go/vt/vtgate/vtgateconntest/client.go
+++ b/go/vt/vtgate/vtgateconntest/client.go
@@ -541,6 +541,7 @@ func TestSuite(t *testing.T, impl vtgateconn.Impl, fakeServer vtgateservice.VTGa
 	testExecuteBatchShardError(t, conn)
 	testExecuteBatchKeyspaceIdsError(t, conn)
 	testStreamExecuteError(t, conn, fs)
+	testStreamExecute2Error(t, conn, fs)
 	testStreamExecuteShardError(t, conn, fs)
 	testStreamExecuteKeyRangesError(t, conn, fs)
 	testStreamExecuteKeyspaceIdsError(t, conn, fs)
@@ -949,6 +950,32 @@ func testStreamExecuteError(t *testing.T, conn *vtgateconn.VTGateConn, fake *fak
 	verifyError(t, err, "StreamExecute")
 }
 
+func testStreamExecute2Error(t *testing.T, conn *vtgateconn.VTGateConn, fake *fakeVTGateService) {
+	ctx := newContext()
+	execCase := execMap["request1"]
+	stream, errFunc, err := conn.StreamExecute2(ctx, execCase.execQuery.Sql, execCase.execQuery.BindVariables, execCase.execQuery.TabletType)
+	if err != nil {
+		t.Fatalf("StreamExecute2 failed: %v", err)
+	}
+	qr, ok := <-stream
+	if !ok {
+		t.Fatalf("StreamExecute2 failed: cannot read result1")
+	}
+
+	if !reflect.DeepEqual(qr, &streamResult1) {
+		t.Errorf("Unexpected result from StreamExecute2: got %#v want %#v", qr, &streamResult1)
+	}
+	// signal to the server that the first result has been received
+	close(fake.errorWait)
+	// After 1 result, we expect to get an error (no more results).
+	qr, ok = <-stream
+	if ok {
+		t.Fatalf("StreamExecute2 channel wasn't closed")
+	}
+	err = errFunc()
+	verifyError(t, err, "StreamExecute2")
+}
+
 func testStreamExecutePanic(t *testing.T, conn *vtgateconn.VTGateConn) {
 	ctx := newContext()
 	execCase := execMap["request1"]