gRPC is an effective way of implementing service-to-service APIs. However, there are limited tools available for mocking and testing gRPC services out of the box. One option is to set up a live test server, although this comes with its own challenges and costs. In this blog post we will demonstrate a more lightweight solution using the bufconn package and a hand-built mock. Don’t worry, this is easier than it seems!
Before we get started, the full tutorial code is available here.
The API
We usually start by defining the API for our service, as it can inform many decisions related to implementation. For simplicity, we’ll create an “agenda” service in our .proto file, which contains two RPC definitions and some protocol buffers. The Schedule RPC will allow the client to schedule items on the agenda and the Daily RPC will return a list of scheduled items to the client.
service Agenda {
rpc Schedule(Item) returns (Item) {}
rpc Daily(Day) returns (Docket) {}
}
message Item {
string id = 1;
string title = 2;
string date = 3;
string start = 4;
string end = 5;
string description = 6;
}
message Day {
string date = 1;
string start = 2;
string end = 3;
}
message Docket {
string date = 1;
repeated Item items = 2;
}
Mocking Communication
In order to test our gRPC service, we will need to inject a mocked connection. Fortunately, there is a package called bufconn1, which does most of the heavy lifting for us. bufconn gives us two things, a way to start an in-memory buffered listener and a way to connect to that listener using a grpc.DialOption. We have found it useful to abstract this functionality into a “utils” package.
const bufsize = 1024 * 1024
type Listener struct {
sock *bufconn.Listener
}
func New() *Listener {
return &Listener{
sock: bufconn.Listen(bufsize),
}
}
func (l *Listener) Sock() net.Listener {
return l.sock
}
func (l *Listener) Close() error {
return l.sock.Close()
}
func (l *Listener) Connect(ctx context.Context, opts ...grpc.DialOption) (cc *grpc.ClientConn, err error) {
opts = append([]grpc.DialOption{grpc.WithContextDialer(l.Dialer)}, opts...)
if cc, err = grpc.DialContext(ctx, "bufnet", opts...); err != nil {
return nil, err
}
return cc, nil
}
func (l *Listener) Dialer(context.Context, string) (net.Conn, error) {
return l.sock.Dial()
}
Creating the Server
Our server setup is fairly idiomatic aside from one important detail: we don’t want our testing code to start a “real” server since we are mocking that with bufconn. Therefore, our Serve function is separated into Serve and Run. This will allow our tests to pass in a net.Listener object obtained from bufconn rather than a connection address. The exciting part about this is that it completely isolates the mock code from the production server code!
type Server struct {
api.UnimplementedAgendaServer
srv *grpc.Server
echan chan error
}
func New() (*Server, error) {
s := &Server{
echan: make(chan error, 1),
}
// Create the gRPC Server
s.srv = grpc.NewServer()
api.RegisterAgendaServer(s.srv, s)
return s, nil
}
// The test code will not call this function
func (s *Server) Serve(addr string) (err error) {
// Catch OS signals for graceful shutdowns
quit := make(chan os.Signal, 1)
signal.Notify(quit, os.Interrupt)
go func() {
<-quit
s.echan <- s.Shutdown()
}()
// Listen for TCP requests on the specified address and port
var sock net.Listener
if sock, err = net.Listen("tcp", addr); err != nil {
return fmt.Errorf("could not listen on %q", addr)
}
// Run the server
go s.Run(sock)
log.Info().Str("listen", addr).Str("version", pkg.Version()).Msg("agenda server started")
// Block and wait for an error either from shutdown or grpc.
if err = <-s.echan; err != nil {
return err
}
return nil
}
// The test code can pass in a bufconn listener here
func (s *Server) Run(sock net.Listener) {
defer sock.Close()
if err := s.srv.Serve(sock); err != nil {
s.echan <- err
}
}
Creating the Client
For the client implementation, we take advantage of variadic arguments to implement a similar isolation strategy. When initializing the client, the caller can pass in any number of grpc.DialOption parameters (even zero). In production, we may want to pass in mTLS or other options here. For testing, we will pass in the dialer from bufconn.
type Client struct {
cc *grpc.ClientConn
rpc api.AgendaClient
}
func New(endpoint string, opts ...grpc.DialOption) (c *Client, err error) {
c = &Client{}
if len(opts) == 0 {
// Production connection opts (TLS, etc.)
opts = make([]grpc.DialOption, 0)
opts = append(opts, grpc.WithTransportCredentials(insecure.NewCredentials()))
}
if c.cc, err = grpc.Dial(endpoint, opts...); err != nil {
return nil, err
}
c.rpc = api.NewAgendaClient(c.cc)
return c, nil
}
func (c *Client) Close() error {
return c.cc.Close()
}
The Server Mock
In previous blog posts, we have discussed the importance of isolation and configurability of mocks. We will now demonstrate these concepts by creating an independent mock for the AgendaServer in a different package that we can easily utilize from the tests. The key is to use the bufconn utilities that we created earlier to start the server by calling Run and create an accessor for the tests to access the client side of the bufconn connection.
const (
DailyRPC = "agenda.v1.Agenda/Daily"
ScheduleRPC = "agenda.v1.Agenda/Schedule"
)
// New creates a new mock AgendaServer. If bufnet is nil, one is created for the user.
func New(bufnet *utils.Listener) *AgendaServer {
if bufnet == nil {
bufnet = utils.New()
}
remote := &AgendaServer{
bufnet: bufnet,
srv: grpc.NewServer(),
Calls: make(map[string]int),
}
api.RegisterAgendaServer(remote.srv, remote)
go remote.srv.Run(remote.bufnet.Sock())
return remote
}
type AgendaServer struct {
api.UnimplementedAgendaServer
bufnet *utils.Listener
srv *grpc.Server
Calls map[string]int
OnSchedule func(context.Context, *api.Item) (*api.Item, error)
OnDaily func(context.Context, *api.Day) (*api.Docket, error)
}
// Tests can call this to get access to the bufconn
func (s *AgendaServer) Channel() *utils.Listener {
return s.bufnet
}
func (s *AgendaServer) Shutdown() {
s.srv.GracefulStop()
s.bufnet.Close()
}
Note that we have included a call counter on the AgendaServer. This allows us to verify from the tests what methods have been called via the gRPC server.
func (s *AgendaServer) Daily(ctx context.Context, in *api.Day) (out *api.Docket, err error) {
s.Calls[DailyRPC]++
return s.OnDaily(ctx, in)
}
func (s *AgendaServer) Schedule(ctx context.Context, in *api.Item) (out *api.Item, err error) {
s.Calls[ScheduleRPC]++
return s.OnSchedule(ctx, in)
}
The mock also gives us the ability to configure how the gRPC handlers will behave. For example, we might want to test how our code deals with errors or specific responses from the handlers.
// UseFixture loads a a JSON fixture from disk (usually in a testdata folder) to use as
// the protocol buffer response to the specified RPC, simplifying handler mocking.
func (s *AgendaServer) UseFixture(rpc, path string) (err error) {
var data []byte
if data, err = ioutil.ReadFile(path); err != nil {
return fmt.Errorf("could not read fixture: %v", err)
}
jsonpb := &protojson.UnmarshalOptions{
AllowPartial: true,
DiscardUnknown: true,
}
switch rpc {
case ScheduleRPC:
out := &api.Item{}
if err = jsonpb.Unmarshal(data, out); err != nil {
return fmt.Errorf("could not unmarshal json into %T: %v", out, err)
}
s.OnSchedule = func(context.Context, *api.Item) (*api.Item, error) {
return out, nil
}
case DailyRPC:
out := &api.Docket{}
if err = jsonpb.Unmarshal(data, out); err != nil {
return fmt.Errorf("could not unmarshal json into %T: %v", out, err)
}
s.OnDaily = func(context.Context, *api.Day) (*api.Docket, error) {
return out, nil
}
default:
return fmt.Errorf("unknown RPC %q", rpc)
}
return nil
}
// UseError allows you to specify a gRPC status error to return from the specified RPC.
func (s *AgendaServer) UseError(rpc string, code codes.Code, msg string) error {
switch rpc {
case ScheduleRPC:
s.OnSchedule = func(context.Context, *api.Item) (*api.Item, error) {
return nil, status.Error(code, msg)
}
case DailyRPC:
s.OnDaily = func(context.Context, *api.Day) (*api.Docket, error) {
return nil, status.Error(code, msg)
}
default:
return fmt.Errorf("unknown RPC %q", rpc)
}
return nil
}
Putting It All Together
With our mock package, writing gRPC tests becomes more of a matter of what we want to test rather than how to test it. The test below verifies that our mock works by simulting an error path and a “happy” path.
func TestSchedule(t *testing.T) {
// Setup AgendaServer mock
srv := mock.New(nil)
defer srv.Shutdown()
// Create a client to test that is connected to the mock server.
dialer := grpc.WithContextDialer(srv.Channel().Dialer)
creds := grpc.WithTransportCredentials(insecure.NewCredentials())
agenda, err := client.New("bufconn", dialer, creds)
require.NoError(t, err, "could not connect to remote agenda via bufconn")
// Test the case where the server returns an error
srv.UseError(mock.ScheduleRPC, codes.DataLoss, "something bad happened")
err = agenda.Schedule("hello", "world", time.Now(), time.Now().Add(5*time.Minute))
require.Error(t, err, "expected an error returned from the server")
require.Equal(t, 1, srv.Calls[mock.ScheduleRPC])
// Test the case where server returns a response
srv.UseFixture(mock.ScheduleRPC, "testdata/item.json")
err = agenda.Schedule("hello", "world", time.Now(), time.Now().Add(5*time.Minute))
require.NoError(t, err, "expected no error in happy path")
require.Equal(t, 2, srv.Calls[mock.ScheduleRPC])
}
Conclusion
Mocking gRPC services for the purposes of testing isn’t always straightforward. However, with tools like bufconn and some clever code factoring, we can create an in-memory mock that is completely isolated from production code. Introducing some measure of configurability into the mocks can also help us focus on what we’re testing rather than how we’re testing it.
Check out the full tutorial code here.
References
Photo by JJ Ying via Unsplash
