When it comes to the tests, one of the main questions is how to keep balance between expressivity, extendability, repeatability and performance. There are countless discussions if you should write integration or unit tests, should your mock or not, should you write tests first or after and etc. Since you will never find the right answer, on a growing code base, multiple people start introducing own methodology and distinct test helpers. Even looking at our quite small code base, I can find like 3-4 ways to write the same test. Additionally expressivity of our tests are quite bad: it is quite hard to understand what actually get tested, lot of boilerplate code not related to test logic, and amount of copy-paste growing with each test.
In order to fix issues described above, I think it is important to have an official guide on writing the tests.
This idea behind this framework is not new, and we already had pieces of it around the code. My goal was to unify all the patterns we used previously, and design a small layer on top of it, to streamline process of writing the tests.
Main points of the new framework are:
- All tests run HTTP requests though the full HTTP stack, same as user will do
- Test definition logic separated from test runner.
- Official mocks for the Dashboard, RPC, and Bundler
Let’s learn by example:
func genAuthHeader(username, password string) string {
toEncode := strings.Join([]string{username, password}, ":")
encodedPass := base64.StdEncoding.EncodeToString([]byte(toEncode))
return fmt.Sprintf("Basic %s", encodedPass)
}
func TestBasicAuth(t *testing.T) {
// Start the test server
ts := newTykTestServer()
defer ts.Close()
// Configure and load API definition
buildAndLoadAPI(func(spec *APISpec) {
spec.UseBasicAuth = true
spec.UseKeylessAccess = false
spec.Proxy.ListenPath = "/"
spec.OrgID = "default"
})
// Prepare data which will be used in tests
session := createStandardSession()
session.BasicAuthData.Password = "password"
session.AccessRights = map[string]user.AccessDefinition{"test": {APIID: "test", Versions: []string{"v1"}}}
validPassword := map[string]string{"Authorization": genAuthHeader("user", "password")}
wrongPassword := map[string]string{"Authorization": genAuthHeader("user", "wrong")}
wrongFormat := map[string]string{"Authorization": genAuthHeader("user", "password:more")}
malformed := map[string]string{"Authorization": "not base64"}
// Running tests one by one, based on our definition
ts.Run(t, []test.TestCase{
// Create base auth based key
{Method: "POST", Path: "/tyk/keys/defaultuser", Data: session, AdminAuth: true, Code: 200},
{Method: "GET", Path: "/", Code: 401, BodyMatch: `Authorization field missing`},
{Method: "GET", Path: "/", Headers: validPassword, Code: 200},
{Method: "GET", Path: "/", Headers: wrongPassword, Code: 401},
{Method: "GET", Path: "/", Headers: wrongFormat, Code: 400, BodyMatch: `Attempted access with malformed header, values not in basic auth format`},
{Method: "GET", Path: "/", Headers: malformed, Code: 400, BodyMatch: `Attempted access with malformed header, auth data not encoded correctly`},
}...)
}And now compare it with previous Go style approach:
func TestBasicAuthWrongPassword(t *testing.T) {
spec := createSpecTest(t, basicAuthDef)
session := createBasicAuthSession()
username := "4321"
// Basic auth sessions are stored as {org-id}{username}, so we need to append it here when we create the session.
spec.SessionManager.UpdateSession("default4321", session, 60)
to_encode := strings.Join([]string{username, "WRONGPASSTEST"}, ":")
encodedPass := base64.StdEncoding.EncodeToString([]byte(to_encode))
recorder := httptest.NewRecorder()
req := testReq(t, "GET", "/", nil)
req.Header.Set("Authorization", fmt.Sprintf("Basic %s", encodedPass))
chain := getBasicAuthChain(spec)
chain.ServeHTTP(recorder, req)
if recorder.Code == 200 {
t.Error("Request should have failed and returned non-200 code!: \n", recorder.Code)
}
if recorder.Code != 401 {
t.Error("Request should have returned 401 code!: \n", recorder.Code)
}
if recorder.Header().Get("WWW-Authenticate") == "" {
t.Error("Request should have returned WWW-Authenticate header!: \n")
}
}Note that in the last “classic” way we defined only 1 test case, while in with our new framework we defined 6, all of them repeatable, and share the same assertion and test runner logic provided by framework.
Now lets review tests written with a new framework piece by piece.
One of the core ideas, is that tests should be as close as possible to real users. In order to implement it, framework provides you a way to programmatically start and stop full Gateway HTTP stack using tykTestServer object, like this:
ts := newTykTestServer()
defer ts.Close()When you create a new server, it initialize gateway itself, starts listener on random port, setup required global variables and etc. It is very similar to what happens when you start gateway process, but in this case you can start and stop it on demand.
You can configure server behavior using few variable, like setting control API on a separate port, by providing tykTestServerConfig object, to newTykTestServer as argument. Here is the list of all possible arguments:
ts := newTykTestServer(tykTestServerConfig{
// Run control API on a separate port
sepatateControlAPI: true,
// Add delay after each test case, if you code depend on timing
// Bad practice, but sometimes needed
delay: 10 * time.Millisecond,
// Emulate that Gateway restarted using SIGUSR2
hotReload: true,
// Emulate that listener will
overrideDefaults, true,
})To close the server simply call tykTestServer#Close method, which will ensure that all the listeners will be properly closed.
buildAndLoadAPI(func(spec *APISpec) {
spec.UseBasicAuth = true
spec.UseKeylessAccess = false
spec.Proxy.ListenPath = "/"
spec.OrgID = "default"
})Basic idea that you have default bare minimum API definition, which you can configure using generator function, to set state required for the test. API then will be loaded into the Gateway, and will be ready to be used inside tests.
If you need to load multiple APIs at the same time, buildAndLoadAPI support variadic number of arguments: buildAndLoadAPI(<fn1>, <fn2>, ...)
You can also call it without arguments at all, in this case it will load default API definition: buildAndLoadAPI()
In fact, this function is mashup of 2 lower level functions: buildAPI and loadAPI, both returning []*APISpec array. In some cases you may need to build API template, and with some smaller modifications load it on demand in different tests. So it can look like:
spec := buildAPI(<fn>)
...
spec.SomeField = "Case1"
loadAPI(spec)
...
spec.SomeField = "Case2"
loadAPI(spec)Updating variables inside API version can be tricky, because API version object is inside Versions map, and direct manipulations with map value is prohibited. To simplify this process, there is special helper updateAPIVersion, which can be used like this:
updateAPIVersion(spec, "v1", func(v *apidef.VersionInfo) {
v.Paths.BlackList = []string{"/blacklist/literal", "/blacklist/{id}/test"}
v.UseExtendedPaths = false
})In some cases updating API definition via Go structures can be a bit complex, and you may want to update API definition directly via JSON unmarshaling:
updateAPIVersion(spec, "v1", func(v *apidef.VersionInfo) {
json.Unmarshal([]byte(`[
{
"path": "/ignored/literal",
"method_actions": {"GET": {"action": "no_action"}}
},
{
"path": "/ignored/{id}/test",
"method_actions": {"GET": {"action": "no_action"}}
}
]`), &v.ExtendedPaths.Ignored)
})ts.Run(t, []test.TestCase{
// Create base auth based key
{Method: "POST", Path: "/tyk/keys/defaultuser", Data: session, AdminAuth: true, Code: 200},
{Method: "GET", Path: "/", Code: 401, BodyMatch: `Authorization field missing`},
{Method: "GET", Path: "/", Headers: validPassword, Code: 200},
{Method: "GET", Path: "/", Headers: wrongPassword, Code: 401},
{Method: "GET", Path: "/", Headers: wrongFormat, Code: 400, BodyMatch: `Attempted access with malformed header, values not in basic auth format`},
{Method: "GET", Path: "/", Headers: malformed, Code: 400, BodyMatch: `Attempted access with malformed header, auth data not encoded correctly`},
}...)
}Tests are defined using new test package TestCase structure, which allows you to define both http request details and response assertions. For example {Method: "GET", Path: "/", Headers: validPassword, Code: 200} tells to make a GET request to / path, with specified headers. After request is made, it will assert response status code with given value.
type TestCase struct {
Method string
Path string
Domain string
// Request body, can be any object. If not string, It will be automatically serialized
Data interface{}
Headers map[string]string
PathParams map[string]string
Cookies []*http.Cookie
BeforeFn func()
// If need custom http client settings, like SSL client certificate or custom timeouts
Client *http.Client
// If true, pass valid admin-auth header
AdminAuth bool
// if Control API run on separate port (or domain), tell to run request on this listener
ControlRequest bool
// Assertions
Code int
BodyMatch string
BodyNotMatch string
HeadersMatch map[string]string
HeadersNotMatch map[string]string
// If http request returns non http error, like TLS or Timeout
ErrorMatch string
}tykTestServer provides a test runner, which generate HTTP requests based on specification and does assertions. Most of the time you going to use tykTestServer#Run(t *testing.T, test.TestCase...) (*http.Response, error)function. Note that it use variadic number of arguments, so if you need to pass multiple test cases, pass it like in example above: []test.TestCase{<tc1>,<tc2>}..., with 3 dots in the end.
Additionally there is RunEx function, with exactly same definition, but internally it runs test cases multiple times (4 right now) with different combinations of overrideDefaults and hotReload options. This can be handy if you need to test functionality that tightly depends hot reload functionality, like reloading APIs, loading plugin bundles or listener itself.
Both Run and RunEx also return response and error of the last test case, in case if you need it.
In lot of cases tests depend on various config variables. You can can update them directly on config.Global object, and restore default config using resetTestConfig function.
config.Global.HttpServerOptions.OverrideDefaults = true
config.Global.HttpServerOptions.SkipURLCleaning = true
defer resetTestConfig()You may notice that default API already targets some upstream mock, created for testing purpose. Url of the upstream hold in testHttpAny variable, but in most cases you do not need it, because APIs created by default already embed it. By default this upstream mock will successfully respond to any url, and response will contain details of the request in the following format:
type testHttpResponse struct {
Method string
Url string
Headers map[string]string
Form map[string]string
}Note that it include final request details, so, for example if you need to test URL rewriting functionality, URL of original request will differ from URL in response of upstream mock, and you can assert it with: BodyMatch: "Url":"". Also notice how we used simple BodyMatch string assertion to validation JSON response.
There is also few special URLs with specific behavior:
/getaccepts onlyGETrequests/postaccepts onlyPOSTrequests/jwk.jsonused for cases when JWK token downloaded from upsteram/wsused for testing WebSockets/bundlesbuilt in plugin bundle web server, more details below
If you want use Python, Lua or GRPC plugins, you need bundle manifest file and scripts to ZIP file, upload them somewhere on external file webserver, and point Gateway to bundle location.
Our test framework include built-in bundle file server, and for simplicity, you provide only content of the of the bundle files, and it will automatically server it as ZIP file.
- Create
map[string]stringobject with file contents, where key is file name - Call
registerBundle("<unique-plugin-id>", <map-with-files>)which will return unique bundle ID. - When creating API set
spec.CustomMiddlewareBundleto bundle ID returned byregisterBundle
Example of loading python auth plugin:
var pythonBundleWithAuthCheck = map[string]string{
"manifest.json": `
{
"file_list": [
"middleware.py"
],
"custom_middleware": {
"driver": "python",
"auth_check": {
"name": "MyAuthHook"
}
}
}
`,
"middleware.py": `
from tyk.decorators import *
from gateway import TykGateway as tyk
@Hook
def MyAuthHook(request, session, metadata, spec):
print("MyAuthHook is called")
auth_header = request.get_header('Authorization')
if auth_header == 'valid_token':
session.rate = 1000.0
session.per = 1.0
metadata["token"] = "valid_token"
return request, session, metadata
`,
}
func TestPython(t *testing.T) {
ts := newTykTestServer()
defer ts.Close()
bundleID := registerBundle("python_with_auth_check", pythonBundleWithAuthCheck)
buildAndLoadAPI(func(spec *APISpec) {
spec.UseKeylessAccess = false
spec.EnableCoProcessAuth = true
spec.CustomMiddlewareBundle = bundleID
})
// test code goes here
}You can create a user session, similar to API, by calling createSession function:
key := createSession(func(s *user.SessionState) {
s.QuotaMax = 2
})You can call it without arguments as well, if you are ok with default settings createSession()
If you need to create session object without adding it to database, for example if you need to create key explicitly via API, you can use createStandardSession() function, which returns *user.SessionState object.
If you need to create custom upstream test server, for example if you need custom TLS settings for Mutual TLS testing, the easiest way is to use standard Go net/http/httptest package and override spec.Proxy.TargetURL API URL to test server.
upstream := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
// custom logic
}))
buildAndLoadAPI(func(spec *APISpec) {
spec.Proxy.TargetURL = upstream.URL
})There is no any specific object to mock the dashboard (yet), but since Dashboard is a standard HTTP server, you can use approach similar to described in Custom upstream mock section:
dashboard := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
if r.URL.Path == "/system/apis" {
w.Write([]byte(`{"Status": "OK", "Nonce": "1", "Message": [{"api_definition": {}}]}`))
} else {
t.Fatal("Unknown dashboard API request", r)
}
}))
config.Global.UseDBAppConfigs = true
config.Global.AllowInsecureConfigs = true
config.Global.DBAppConfOptions.ConnectionString = dashboard.URLWhen Gateway works in Hybrid mode, it talks with MDCB instance via RPC channel using gorpc library. You can use startRPCMock and stopRPCMock functions to mock RPC server. startRPCMock internally sets required config variables to enable RPC mode.
func TestSyncAPISpecsRPCSuccess(t *testing.T) {
// Mock RPC
dispatcher := gorpc.NewDispatcher()
dispatcher.AddFunc("GetApiDefinitions", func(clientAddr string, dr *DefRequest) (string, error) {
return "[{}]", nil
})
dispatcher.AddFunc("Login", func(clientAddr, userKey string) bool {
return true
})
rpc := startRPCMock(dispatcher)
defer stopRPCMock(rpc)
count := syncAPISpecs()
if count != 1 {
t.Error("Should return array with one spec", apiSpecs)
}
}Inside tests we override default network resolver to use custom DNS server mock, creating using awesome github.com/miekg/dns library. Domain -> IP mapping set via map inside helpers_test.go file. By default you have access to domains: localhost, host1.local, host2.local and host3.local. Access to all unknown domains will cause panic.
Using DNS mock means that you are able to create tests with APIs on multiple domains, without modifying machine /etc/hosts file.