gdt - The Go Declarative Testing framework

gdt is a testing library that allows test authors to cleanly describe tests in a YAML file. gdt reads YAML files that describe a test's assertions and then builds a set of Go structures that the standard Go testing package and standard go test tool can execute.

Introduction

Writing functional tests in Go can be overly verbose and tedious. When the code that tests some part of an application is verbose or tedious, then it becomes difficult to read the tests and quickly understand the assertions the test is making.

The more difficult it is to understand the test assertions or the test setups and assumptions, the greater the chance that the test improperly validates the application behaviour. Furthermore, test code that is cumbersome to read is prone to bit-rot due to its high maintenance cost. This is particularly true for code that verifies an application's integration points with other applications via an API.

The idea behind gdt is to allow test authors to cleanly and clearly describe a functional test's assumptions and assertions in a declarative format.

Separating the description of a test's assumptions (setup) and assertions from the Go code that actually performs the test assertions leads to tests that are easier to read and understand. This allows developers to spend more time writing code and less time copy/pasting boilerplate test code. Due to the easier test comprehension, gdt also encourages writing greater quality and coverage of functional tests.

Instead of developers writing code that looks like this:

var _ = Describe("Books API - GET /books failures", func() {
    var response *http.Response
    var err error
    var testPath = "/books/nosuchbook"

    BeforeEach(func() {
        response, err = http.Get(apiPath(testPath))
        Ω(err).Should(BeZero())
    })

    Describe("failure modes", func() {
        Context("when no such book was found", func() {
            It("should not include JSON in the response", func() {
                Ω(respJSON(response)).Should(BeZero())
            })
            It("should return 404", func() {
                Ω(response.StatusCode).Should(Equal(404))
            })
        })
    })
})

they can instead have a test that looks like this:

fixtures:
 - books_api
tests:
 - name: no such book was found
   GET: /books/nosuchbook
   response:
     json:
       len: 0
     status: 404

Coming from Ginkgo

When using Ginkgo, developers create tests for a particular module (say, the books module) by creating a books_test.go file and calling some Ginkgo functions in a BDD test style. A sample Ginkgo test might look something like this (types_test.go):

package api_test

import (
    "github.com/gdt-dev/examples/books/api"
    . "github.com/onsi/ginkgo"
    . "github.com/onsi/gomega"
)

var _ = Describe("Books API Types", func() {
    var (
        longBook  api.Book
        shortBook api.Book
    )

    BeforeEach(func() {
        longBook = api.Book{
            Title: "Les Miserables",
            Pages: 1488,
            Author: &api.Author{
                Name: "Victor Hugo",
            },
        }

        shortBook = api.Book{
            Title: "Fox In Socks",
            Pages: 24,
            Author: &api.Author{
                Name: "Dr. Seuss",
            },
        }
    })

    Describe("Categorizing book length", func() {
        Context("With more than 300 pages", func() {
            It("should be a novel", func() {
                Expect(longBook.CategoryByLength()).To(Equal("NOVEL"))
            })
        })

        Context("With fewer than 300 pages", func() {
            It("should be a short story", func() {
                Expect(shortBook.CategoryByLength()).To(Equal("SHORT STORY"))
            })
        })
    })
})

This is perfectly fine for simple unit tests of Go code. However, once the tests begin to call multiple APIs or packages, the Ginkgo Go tests start to get cumbersome. Consider the following example of functionally testing the failure modes for a simple HTTP REST API endpoint (failure_test.go):

package api_test

import (
    "io/ioutil"
    "log"
    "net/http"
    "net/http/httptest"
    "os"
    "strings"

    . "github.com/onsi/ginkgo"
    . "github.com/onsi/gomega"

    "github.com/gdt-dev/examples/http/api"
)

var (
    server *httptest.Server
)

// respJSON returns a string if the supplied HTTP response body is JSON,
// otherwise the empty string
func respJSON(r *http.Response) string {
    if r == nil {
        return ""
    }
    if !strings.HasPrefix(r.Header.Get("content-type"), "application/json") {
        return ""
    }
    bodyStr, _ := ioutil.ReadAll(r.Body)
    return string(bodyStr)
}

// respText returns a string if the supplied HTTP response has a text/plain
// content type and a body, otherwise the empty string
func respText(r *http.Response) string {
    if r == nil {
        return ""
    }
    if !strings.HasPrefix(r.Header.Get("content-type"), "text/plain") {
        return ""
    }
    bodyStr, _ := ioutil.ReadAll(r.Body)
    return string(bodyStr)
}

func apiPath(path string) string {
    return strings.TrimSuffix(server.URL, "/") + "/" + strings.TrimPrefix(path, "/")
}

// Register an HTTP server fixture that spins up the API service on a
// random port on localhost
var _ = BeforeSuite(func() {
    logger := log.New(os.Stdout, "http: ", log.LstdFlags)
    c := api.NewControllerWithBooks(logger, nil)
    server = httptest.NewServer(c.Router())
})

var _ = AfterSuite(func() {
    server.Close()
})

var _ = Describe("Books API - GET /books failures", func() {
    var response *http.Response
    var err error
    var testPath string

    BeforeEach(func() {
        response, err = http.Get(apiPath(testPath))
        Ω(err).Should(BeZero())
    })

    Describe("failure modes", func() {
        AssertZeroJSONLength := func() {
            It("should not include JSON in the response", func() {
                Ω(respJSON(response)).Should(BeZero())
            })
        }

        Context("when no such book was found", func() {
            JustBeforeEach(func() {
                testPath = "/books/nosuchbook"
            })

            AssertZeroJSONLength()

            It("should return 404", func() {
                Ω(response.StatusCode).Should(Equal(404))
            })
        })

        Context("when an invalid query parameter is supplied", func() {
            JustBeforeEach(func() {
                testPath = "/books?invalidparam=1"
            })

            AssertZeroJSONLength()

            It("should return 400", func() {
                Ω(response.StatusCode).Should(Equal(400))
            })
            It("should indicate invalid query parameter", func() {
                Ω(respText(response)).Should(ContainSubstring("invalid parameter"))
            })
        })
    })
})

The above test code obscures what is being tested by cluttering the test assertions with the Go closures and accessor code. Compare the above with how gdt allows the test author to describe the same assertions (failures.yaml):

fixtures:
 - books_api
tests:
 - name: no such book was found
   GET: /books/nosuchbook
   response:
     json:
       len: 0
     status: 404
 - name: invalid query parameter is supplied
   GET: /books?invalidparam=1
   response:
     json:
       len: 0
     status: 400
     strings:
       - invalid parameter

No more closures and boilerplate function code getting in the way of expressing the assertions, which should be the focus of the test.

The more intricate the assertions being verified by the test, generally the more verbose and cumbersome the Go test code tends to become. First and foremost, tests should be readable. If they are not readable, then the test's assertions are not understandable. And tests that cannot easily be understood are often the source of bit rot and technical debt. Worse, tests that aren't understandable stand a greater chance of having an improper assertion go undiscovered, leading to tests that validate the wrong behaviour or don't validate the correct behaviour.

Consider a Ginkgo test case that checks the following behaviour:

• When a book is created via a call to POST /books, we are able to get book information from the link returned in the HTTP response's Location header
• The newly-created book's author name should be set to a known value
• The newly-created book's ID field is a valid UUID
• The newly-created book's publisher has an address containing a known state code

A typical implementation of a Ginkgo test might look like this (create_then_get_test.go):

package api_test

import (
    "bytes"
    "encoding/json"
    "net/http"

    . "github.com/onsi/ginkgo"
    . "github.com/onsi/gomega"

    "github.com/gdt-dev/examples/http/api"
)

var _ = Describe("Books API - POST /books -> GET /books from Location", func() {

    var err error
    var resp *http.Response
    var locURL string
    var authorID, publisherID string

    Describe("proper HTTP GET after POST", func() {

        Context("when creating a single book resource", func() {
            It("should be retrievable via GET {location header}", func() {
                // See https://github.com/onsi/ginkgo/issues/457 for why this
                // needs to be here instead of in the outer Describe block.
                authorID = getAuthorByName("Ernest Hemingway").ID
                publisherID = getPublisherByName("Charles Scribner's Sons").ID
                req := api.CreateBookRequest{
                    Title:       "For Whom The Bell Tolls",
                    AuthorID:    authorID,
                    PublisherID: publisherID,
                    PublishedOn: "1940-10-21",
                    Pages:       480,
                }
                var payload []byte
                payload, err = json.Marshal(&req)
                if err != nil {
                    Fail("Failed to serialize JSON in setup")
                }
                resp, err = http.Post(apiPath("/books"), "application/json", bytes.NewBuffer(payload))
                Ω(err).Should(BeNil())

                // See https://github.com/onsi/ginkgo/issues/70 for why this
                // has to be one giant It() block. The GET tests rely on the
                // result of an earlier POST response (for the Location header)
                // and therefore all of the assertions below must be in a
                // single It() block. :(

                Ω(resp.StatusCode).Should(Equal(201))
                Ω(resp.Header).Should(HaveKey("Location"))

                locURL = resp.Header["Location"][0]

                resp, err = http.Get(apiPath(locURL))
                Ω(err).Should(BeNil())

                Ω(resp.StatusCode).Should(Equal(200))

                var book api.Book

                err := json.Unmarshal([]byte(respJSON(resp)), &book)
                Ω(err).Should(BeNil())

                Ω(IsValidUUID4(book.ID)).Should(BeTrue())
                Ω(book.Author).ShouldNot(BeNil())
                Ω(book.Author.Name).Should(Equal("Ernest Hemingway"))
                Ω(book.Publisher).ShouldNot(BeNil())
                Ω(book.Publisher.Address).ShouldNot(BeNil())
                Ω(book.Publisher.Address.State).Should(Equal("NY"))
            })
        })
    })
})

Compare the above test code to the following YAML document that a gdt user might create to describe the same assertions (create_then_get.yaml):

fixtures:
 - books_api
 - books_data
tests:
 - name: create a new book
   POST: /books
   data:
     title: For Whom The Bell Tolls
     published_on: 1940-10-21
     pages: 480
     author_id: $.authors.by_name["Ernest Hemingway"].id
     publisher_id: $.publishers.by_name["Charles Scribner's Sons"].id
   response:
     status: 201
     headers:
      - Location
 - name: look up that created book
   GET: $$LOCATION
   response:
     status: 200
     json:
       paths:
         $.author.name: Ernest Hemingway
         $.publisher.address.state: New York
       path-formats:
         $.id: uuid4

gdt test scenario structure

A gdt test scenario (or just "scenario") is simply a YAML file.

All gdt scenarios have the following fields:

• name: (optional) string describing the contents of the test file. If missing or empty, the filename is used as the name
• description: (optional) string with longer description of the test file contents
• defaults: (optional) is a map of default options and configuration values
• fixtures: (optional) list of strings indicating named fixtures that will be started before any of the tests in the file are run
• skip-if: (optional) list of Spec specializations that will be evaluated before running any test in the scenario. If any of these conditions evaluates successfully, the test scenario will be skipped.
• tests: list of Spec specializations that represent the runnable test units in the test scenario.

The scenario's tests field is the most important and the Spec objects that it contains are the meat of a test scenario.

gdt test spec structure

A spec represents a single action that is taken and zero or more assertions that represent what you expect to see resulting from that action.

gdt plugins each define a specialized subclass of the base Spec that contains fields that are specific to that type of test.

For example, there is an exec plugin that allows you to execute arbitrary commands and assert expected result codes and output. There is an http that allows you to call an HTTP URL and assert that the response looks like what you expect. There is a kube plugin that allows you to interact with a Kubernetes API, etc.

gdt examines the YAML file that defines your test scenario and uses these plugins to parse individual test specs.

All test specs have the following fields:

• name: (optional) string describing the test unit.
• description: (optional) string with longer description of the test unit.
• timeout: (optional) a string duration of time the test unit is expected to complete within.
• retry: (optional) an object containing retry configurationu for the test unit. Some plugins will automatically attempt to retry the test action when an assertion fails. This field allows you to control this retry behaviour for each individual test.
• retry.interval: (optional) a string duration of time that the test plugin will retry the test action in the event assertions fail. The default interval for retries is plugin-dependent.
• retry.attempts: (optional) an integer indicating the number of times that a plugin will retry the test action in the event assertions fail. The default number of attempts for retries is plugin-dependent.
• retry.exponential: (optional) a boolean indicating an exponential backoff should be applied to the retry interval. The default is is plugin-dependent.
• wait (optional) an object containing wait information for the test unit.
• wait.before: a string duration of time that gdt should wait before executing the test unit's action.
• wait.after: a string duration of time that gdt should wait after executing the test unit's action.
• on: (optional) an object describing actions to take upon certain conditions.
• on.fail: (optional) an object describing an action to take when any assertion fails for the test action.
• on.fail.exec: a string with the exact command to execute upon test assertion failure. You may execute more than one command but must include the on.fail.shell field to indicate that the command should be run in a shell.
• on.fail.shell: (optional) a string with the specific shell to use in executing the command to run upon test assertion failure. If empty (the default), no shell is used to execute the command and instead the operating system's exec family of calls is used.

exec test spec structure

The exec plugin's test spec allows test authors to execute arbitrary commands and assert that the command results in an expected result code or output.

In addition to all the base Spec fields listed above, the exec plugin's test spec also contains these fields:

• exec: a string with the exact command to execute. You may execute more than one command but must include the shell field to indicate that the command should be run in a shell. It is best practice, however, to simply use multiple exec specs instead of executing multiple commands in a single shell call.
• shell: (optional) a string with the specific shell to use in executing the command. If empty (the default), no shell is used to execute the command and instead the operating system's exec family of calls is used.
• assert: (optional) an object describing the conditions that will be asserted about the test action.
• assert.exit-code: (optional) an integer with the expected exit code from the executed command. The default successful exit code is 0 and therefore you do not need to specify this if you expect a successful exit code.
• assert.out: (optional) a PipeExpect object containing assertions about content in stdout.
• assert.out.is: (optional) a string with the exact contents of stdout you expect to get.
• assert.out.all: (optional) a string or list of strings that all must be present in stdout.
• assert.out.any: (optional) a string or list of strings of which at least one must be present in stdout.
• assert.out.none: (optional) a string or list of strings of which none should be present in stdout.
• assert.err: (optional) a PipeAssertions object containing assertions about content in stderr.
• assert.err.is: (optional) a string with the exact contents of stderr you expect to get.
• assert.err.all: (optional) a string or list of strings that all must be present in stderr.
• assert.err.any: (optional) a string or list of strings of which at least one must be present in stderr.
• assert.err.none: (optional) a string or list of strings of which none should be present in stderr.

Timeouts and retrying assertions

When evaluating assertions for a test spec, gdt inspects the test's timeout value to determine how long to retry the get call and recheck the assertions.

If a test's timeout is empty, gdt inspects the scenario's defaults.timeout value. If both of those values are empty, gdt will look for any default timeout value that the plugin uses.

If you're interested in seeing the individual results of gdt's assertion-checks for a single get call, you can use the gdt.WithDebug() function, like this test function demonstrates:

file: testdata/matches.yaml:

name: matches
description: create a deployment and check the matches condition succeeds
fixtures:
  - kind
tests:
  - name: create-deployment
    kube:
      create: testdata/manifests/nginx-deployment.yaml
  - name: deployment-exists
    kube:
      get: deployments/nginx
    assert:
      matches:
        spec:
          replicas: 2
          template:
            metadata:
              labels:
                app: nginx
        status:
          readyReplicas: 2
  - name: delete-deployment
    kube:
      delete: deployments/nginx

file: matches_test.go

import (
    "github.com/gdt-dev/gdt"
    _ "github.com/gdt-dev/kube"
    kindfix "github.com/gdt-dev/kube/fixture/kind"
)

func TestMatches(t *testing.T) {
	fp := filepath.Join("testdata", "matches.yaml")

	kfix := kindfix.New()

	s, err := gdt.From(fp)

	ctx := gdt.NewContext(gdt.WithDebug())
	ctx = gdt.RegisterFixture(ctx, "kind", kfix)
	s.Run(ctx, t)
}

Here's what running go test -v matches_test.go would look like:

$ go test -v matches_test.go
=== RUN   TestMatches
=== RUN   TestMatches/matches
=== RUN   TestMatches/matches/create-deployment
=== RUN   TestMatches/matches/deployment-exists
deployment-exists (try 1 after 1.303µs) ok: false, terminal: false
deployment-exists (try 1 after 1.303µs) failure: assertion failed: match field not equal: $.status.readyReplicas not present in subject
deployment-exists (try 2 after 595.62786ms) ok: false, terminal: false
deployment-exists (try 2 after 595.62786ms) failure: assertion failed: match field not equal: $.status.readyReplicas not present in subject
deployment-exists (try 3 after 1.020003807s) ok: false, terminal: false
deployment-exists (try 3 after 1.020003807s) failure: assertion failed: match field not equal: $.status.readyReplicas not present in subject
deployment-exists (try 4 after 1.760006109s) ok: false, terminal: false
deployment-exists (try 4 after 1.760006109s) failure: assertion failed: match field not equal: $.status.readyReplicas had different values. expected 2 but found 1
deployment-exists (try 5 after 2.772416449s) ok: true, terminal: false
=== RUN   TestMatches/matches/delete-deployment
--- PASS: TestMatches (3.32s)
    --- PASS: TestMatches/matches (3.30s)
        --- PASS: TestMatches/matches/create-deployment (0.01s)
        --- PASS: TestMatches/matches/deployment-exists (2.78s)
        --- PASS: TestMatches/matches/delete-deployment (0.02s)
PASS
ok  	command-line-arguments	3.683s

You can see from the debug output above that gdt created the Deployment and then did a kube.get for the deployments/nginx Deployment. Initially (attempt 1), the assert.matches assertion failed because the status.readyReplicas field was not present in the returned resource. gdt retried the kube.get call 4 more times (attempts 2-5), with attempts 2 and 3 failed the existence check for the status.readyReplicas field and attempt 4 failing the value check for the status.readyReplicas field being 1 instead of the expected 2. Finally, when the Deployment was completely rolled out, attempt 5 succeeded in all the assert.matches assertions.

Contributing and acknowledgements

gdt was inspired by Gabbi, the excellent Python declarative testing framework. gdt tries to bring the same clear, concise test definitions to the world of Go functional testing.

The Go gopher logo, from which gdt's logo was derived, was created by Renee French.

Contributions to gdt are welcomed! Feel free to open a Github issue or submit a pull request.