Skip to content

Latest commit

 

History

History
1416 lines (1124 loc) · 66.4 KB

running-and-writing-acceptance-tests.md

File metadata and controls

1416 lines (1124 loc) · 66.4 KB
Raw

Running and Writing Acceptance Tests

Terraform includes an acceptance test harness that does most of the repetitive work involved in testing a resource. For additional information about testing Terraform Providers, see the Extending Terraform documentation.

Acceptance Tests Often Cost Money to Run

Because acceptance tests create real resources, they often cost money to run. Because the resources only exist for a short period of time, the total amount of money required is usually a relatively small. Nevertheless, we don't want financial limitations to be a barrier to contribution, so if you are unable to pay to run acceptance tests for your contribution, mention this in your pull request. We will happily accept "best effort" implementations of acceptance tests and run them for you on our side. This might mean that your PR takes a bit longer to merge, but it most definitely is not a blocker for contributions.

Running an Acceptance Test

Acceptance tests can be run using the testacc target in the Terraform Makefile. The individual tests to run can be controlled using a regular expression. Prior to running the tests provider configuration details such as access keys must be made available as environment variables.

For example, to run an acceptance test against the Amazon Web Services provider, the following environment variables must be set:

# Using a profile
export AWS_PROFILE=...
# Otherwise
export AWS_ACCESS_KEY_ID=...
export AWS_SECRET_ACCESS_KEY=...
export AWS_DEFAULT_REGION=...

Please note that the default region for the testing is us-west-2 and must be overridden via the AWS_DEFAULT_REGION environment variable, if necessary. This is especially important for testing AWS GovCloud (US), which requires:

export AWS_DEFAULT_REGION=us-gov-west-1

Tests can then be run by specifying a regular expression defining the tests to run:

$ make testacc TESTARGS='-run=TestAccAWSCloudWatchDashboard_update'
==> Checking that code complies with gofmt requirements...
TF_ACC=1 go test ./aws -v -run=TestAccAWSCloudWatchDashboard_update -timeout 120m
=== RUN   TestAccAWSCloudWatchDashboard_update
--- PASS: TestAccAWSCloudWatchDashboard_update (26.56s)
PASS
ok  	github.com/hashicorp/terraform-provider-aws/aws	26.607s

Entire resource test suites can be targeted by using the naming convention to write the regular expression. For example, to run all tests of the aws_cloudwatch_dashboard resource rather than just the update test, you can start testing like this:

$ make testacc TESTARGS='-run=TestAccAWSCloudWatchDashboard'
==> Checking that code complies with gofmt requirements...
TF_ACC=1 go test ./aws -v -run=TestAccAWSCloudWatchDashboard -timeout 120m
=== RUN   TestAccAWSCloudWatchDashboard_importBasic
--- PASS: TestAccAWSCloudWatchDashboard_importBasic (15.06s)
=== RUN   TestAccAWSCloudWatchDashboard_basic
--- PASS: TestAccAWSCloudWatchDashboard_basic (12.70s)
=== RUN   TestAccAWSCloudWatchDashboard_update
--- PASS: TestAccAWSCloudWatchDashboard_update (27.81s)
PASS
ok  	github.com/hashicorp/terraform-provider-aws/aws	55.619s

Running acceptance tests requires version 0.12.26 or higher of the Terraform CLI to be installed.

For advanced developers, the acceptance testing framework accepts some additional environment variables that can be used to control Terraform CLI binary selection, logging, and other behaviors. See the Extending Terraform documentation for more information.

Please Note: On macOS 10.14 and later (and some Linux distributions), the default user open file limit is 256. This may cause unexpected issues when running the acceptance testing since this can prevent various operations from occurring such as opening network connections to AWS. To view this limit, the ulimit -n command can be run. To update this limit, run ulimit -n 1024 (or higher).

Running Cross-Account Tests

Certain testing requires multiple AWS accounts. This additional setup is not typically required and the testing will return an error (shown below) if your current setup does not have the secondary AWS configuration:

$ make testacc TEST=./aws TESTARGS='-run=TestAccAWSDBInstance_DbSubnetGroupName_RamShared'
=== RUN   TestAccAWSDBInstance_DbSubnetGroupName_RamShared
=== PAUSE TestAccAWSDBInstance_DbSubnetGroupName_RamShared
=== CONT  TestAccAWSDBInstance_DbSubnetGroupName_RamShared
    TestAccAWSDBInstance_DbSubnetGroupName_RamShared: provider_test.go:386: AWS_ALTERNATE_ACCESS_KEY_ID or AWS_ALTERNATE_PROFILE must be set for acceptance tests
--- FAIL: TestAccAWSDBInstance_DbSubnetGroupName_RamShared (2.22s)
FAIL
FAIL	github.com/hashicorp/terraform-provider-aws/aws	4.305s
FAIL

Running these acceptance tests is the same as before, except the following additional AWS credential information is required:

# Using a profile
export AWS_ALTERNATE_PROFILE=...
# Otherwise
export AWS_ALTERNATE_ACCESS_KEY_ID=...
export AWS_ALTERNATE_SECRET_ACCESS_KEY=...

Running Cross-Region Tests

Certain testing requires multiple AWS regions. Additional setup is not typically required because the testing defaults the second AWS region to us-east-1 and the third AWS region to us-east-2.

Running these acceptance tests is the same as before, but if you wish to override the second and third regions:

export AWS_ALTERNATE_REGION=...
export AWS_THIRD_REGION=...

Writing an Acceptance Test

Terraform has a framework for writing acceptance tests which minimises the amount of boilerplate code necessary to use common testing patterns. This guide is meant to augment the general Extending Terraform documentation with Terraform AWS Provider specific conventions and helpers.

Anatomy of an Acceptance Test

This section describes in detail how the Terraform acceptance testing framework operates with respect to the Terraform AWS Provider. We recommend those unfamiliar with this provider, or Terraform resource testing in general, take a look here first to generally understand how we interact with AWS and the resource code to verify functionality.

The entry point to the framework is the resource.ParallelTest() function. This wraps our testing to work with the standard Go testing framework, while also preventing unexpected usage of AWS by requiring the TF_ACC=1 environment variable. This function accepts a TestCase parameter, which has all the details about the test itself. For example, this includes the test steps (TestSteps) and how to verify resource deletion in the API after all steps have been run (CheckDestroy).

Each TestStep proceeds by applying some Terraform configuration using the provider under test, and then verifying that results are as expected by making assertions using the provider API. It is common for a single test function to exercise both the creation of and updates to a single resource. Most tests follow a similar structure.

  1. Pre-flight checks are made to ensure that sufficient provider configuration is available to be able to proceed - for example in an acceptance test targeting AWS, AWS_ACCESS_KEY_ID and AWS_SECRET_ACCESS_KEY must be set prior to running acceptance tests. This is common to all tests exercising a single provider.

Most assertion functions are defined out of band with the tests. This keeps the tests readable, and allows reuse of assertion functions across different tests of the same type of resource. The definition of a complete test looks like this:

func TestAccAWSCloudWatchDashboard_basic(t *testing.T) {
	var dashboard cloudwatch.GetDashboardOutput
	rInt := acctest.RandInt()
	resource.ParallelTest(t, resource.TestCase{
		PreCheck:     func() { testAccPreCheck(t) },
		Providers:    testAccProviders,
		CheckDestroy: testAccCheckAWSCloudWatchDashboardDestroy,
		Steps: []resource.TestStep{
			{
				Config: testAccAWSCloudWatchDashboardConfig(rInt),
				Check: resource.ComposeTestCheckFunc(
					testAccCheckCloudWatchDashboardExists("aws_cloudwatch_dashboard.foobar", &dashboard),
					resource.TestCheckResourceAttr("aws_cloudwatch_dashboard.foobar", "dashboard_name", testAccAWSCloudWatchDashboardName(rInt)),
				),
			},
		},
	})
}

When executing the test, the following steps are taken for each TestStep:

  1. The Terraform configuration required for the test is applied. This is responsible for configuring the resource under test, and any dependencies it may have. For example, to test the aws_cloudwatch_dashboard resource, a valid configuration with the requisite fields is required. This results in configuration which looks like this:

    resource "aws_cloudwatch_dashboard" "foobar" {
  dashboard_name = "terraform-test-dashboard-%d"
  dashboard_body = <<EOF
  {
    "widgets": [{
      "type": "text",
      "x": 0,
      "y": 0,
      "width": 6,
      "height": 6,
      "properties": {
        "markdown": "Hi there from Terraform: CloudWatch"
      }
    }]
  }
  EOF
}

  2. Assertions are run using the provider API. These use the provider API directly rather than asserting against the resource state. For example, to verify that the aws_cloudwatch_dashboard described above was created successfully, a test function like this is used:

    func testAccCheckCloudWatchDashboardExists(n string, dashboard *cloudwatch.GetDashboardOutput) resource.TestCheckFunc {
  return func(s *terraform.State) error {
    rs, ok := s.RootModule().Resources[n]
    if !ok {
      return fmt.Errorf("Not found: %s", n)
    }

    conn := testAccProvider.Meta().(*AWSClient).cloudwatchconn
    params := cloudwatch.GetDashboardInput{
      DashboardName: aws.String(rs.Primary.ID),
    }

    resp, err := conn.GetDashboard(&params)
    if err != nil {
      return err
    }

    *dashboard = *resp

    return nil
  }
}

    Notice that the only information used from the Terraform state is the ID of the resource. For computed properties, we instead assert that the value saved in the Terraform state was the expected value if possible. The testing framework provides helper functions for several common types of check - for example:

    resource.TestCheckResourceAttr("aws_cloudwatch_dashboard.foobar", "dashboard_name", testAccAWSCloudWatchDashboardName(rInt)),

  3. The resources created by the test are destroyed. This step happens automatically, and is the equivalent of calling terraform destroy.

  4. Assertions are made against the provider API to verify that the resources have indeed been removed. If these checks fail, the test fails and reports "dangling resources". The code to ensure that the aws_cloudwatch_dashboard shown above has been destroyed looks like this:

    func testAccCheckAWSCloudWatchDashboardDestroy(s *terraform.State) error {
  conn := testAccProvider.Meta().(*AWSClient).cloudwatchconn

  for _, rs := range s.RootModule().Resources {
    if rs.Type != "aws_cloudwatch_dashboard" {
      continue
    }

    params := cloudwatch.GetDashboardInput{
      DashboardName: aws.String(rs.Primary.ID),
    }

    _, err := conn.GetDashboard(&params)
    if err == nil {
      return fmt.Errorf("Dashboard still exists: %s", rs.Primary.ID)
    }
    if !isCloudWatchDashboardNotFoundErr(err) {
      return err
    }
  }

  return nil
}

    These functions usually test only for the resource directly under test.

Resource Acceptance Testing

Most resources that implement standard Create, Read, Update, and Delete functionality should follow the pattern below. Each test type has a section that describes them in more detail:

  • basic: This represents the bare minimum verification that the resource can be created, read, deleted, and optionally imported.
  • disappears: A test that verifies Terraform will offer to recreate a resource if it is deleted outside of Terraform (e.g. via the Console) instead of returning an error that it cannot be found.
  • Per Attribute: A test that verifies the resource with a single additional argument can be created, read, optionally updated (or force resource recreation), deleted, and optionally imported.

The leading sections below highlight additional recommended patterns.

Test Configurations

Most of the existing test configurations you will find in the Terraform AWS Provider are written in the following function-based style:

func TestAccAwsExampleThing_basic(t *testing.T) {
  // ... omitted for brevity ...

  resource.ParallelTest(t, resource.TestCase{
    // ... omitted for brevity ...
    Steps: []resource.TestStep{
      {
        Config: testAccAwsExampleThingConfig(),
        // ... omitted for brevity ...
      },
    },
  })
}

func testAccAwsExampleThingConfig() string {
  return `
resource "aws_example_thing" "test" {
  # ... omitted for brevity ...
}
`
}

Even when no values need to be passed in to the test configuration, we have found this setup to be the most flexible for allowing that to be easily implemented. Any configurable values are handled via fmt.Sprintf(). Using text/template or other templating styles is explicitly forbidden.

For consistency, resources in the test configuration should be named resource "..." "test" unless multiple of that resource are necessary.

We discourage re-using test configurations across test files (except for some common configuration helpers we provide) as it is much harder to discover potential testing regressions.

Please also note that the newline on the first line of the configuration (before resource) and the newline after the last line of configuration (after }) are important to allow test configurations to be easily combined without generating Terraform configuration language syntax errors.

Combining Test Configurations

We include a helper function, composeConfig() for iteratively building and chaining test configurations together. It accepts any number of configurations to combine them. This simplifies a single resource's testing by allowing the creation of a "base" test configuration for all the other test configurations (if necessary) and also allows the maintainers to curate common configurations. Each of these is described in more detail in below sections.

Please note that we do discourage excessive chaining of configurations such as implementing multiple layers of "base" configurations. Usually these configurations are harder for maintainers and other future readers to understand due to the multiple levels of indirection.

Base Test Configurations

If a resource requires the same Terraform configuration as a prerequisite for all test configurations, then a common pattern is implementing a "base" test configuration that is combined with each test configuration.

For example:

func testAccAwsExampleThingConfigBase() string {
  return `
resource "aws_iam_role" "test" {
  # ... omitted for brevity ...
}

resource "aws_iam_role_policy" "test" {
  # ... omitted for brevity ...
}
`
}

func testAccAwsExampleThingConfig() string {
  return composeConfig(
    testAccAwsExampleThingConfigBase(),
    `
resource "aws_example_thing" "test" {
  # ... omitted for brevity ...
}
`)
}
Available Common Test Configurations

These test configurations are typical implementations we have found or allow testing to implement best practices easier, since the Terraform AWS Provider testing is expected to run against various AWS Regions and Partitions.

  • testAccAvailableEc2InstanceTypeForRegion("type1", "type2", ...): Typically used to replace hardcoded EC2 Instance Types. Uses aws_ec2_instance_type_offering data source to return an available EC2 Instance Type in preferred ordering. Reference the instance type via: data.aws_ec2_instance_type_offering.available.instance_type
  • testAccLatestAmazonLinuxHvmEbsAmiConfig(): Typically used to replace hardcoded EC2 Image IDs (ami-12345678). Uses aws_ami data source to find the latest Amazon Linux image. Reference the AMI ID via: data.aws_ami.amzn-ami-minimal-hvm-ebs.id

Randomized Naming

For AWS resources that require unique naming, the tests should implement a randomized name, typically coded as a rName variable in the test and passed as a parameter to creating the test configuration.

For example:

func TestAccAwsExampleThing_basic(t *testing.T) {
  rName := acctest.RandomWithPrefix("tf-acc-test")
  // ... omitted for brevity ...

  resource.ParallelTest(t, resource.TestCase{
    // ... omitted for brevity ...
    Steps: []resource.TestStep{
      {
        Config: testAccAwsExampleThingConfigName(rName),
        // ... omitted for brevity ...
      },
    },
  })
}

func testAccAwsExampleThingConfigName(rName string) string {
  return fmt.Sprintf(`
resource "aws_example_thing" "test" {
  name = %[1]q
}
`, rName)
}

Typically the rName is always the first argument to the test configuration function, if used, for consistency.

Other Recommended Variables

We also typically recommend saving a resourceName variable in the test that contains the resource reference, e.g. aws_example_thing.test, which is repeatedly used in the checks.

For example:

func TestAccAwsExampleThing_basic(t *testing.T) {
  // ... omitted for brevity ...
  resourceName := "aws_example_thing.test"

  resource.ParallelTest(t, resource.TestCase{
    // ... omitted for brevity ...
    Steps: []resource.TestStep{
      {
        // ... omitted for brevity ...
        Check: resource.ComposeTestCheckFunc(
          testAccCheckAwsExampleThingExists(resourceName),
          testAccCheckResourceAttrRegionalARN(resourceName, "arn", "example", fmt.Sprintf("thing/%s", rName)),
          resource.TestCheckResourceAttr(resourceName, "description", ""),
          resource.TestCheckResourceAttr(resourceName, "name", rName),
        ),
      },
      {
        ResourceName:      resourceName,
        ImportState:       true,
        ImportStateVerify: true,
      },
    },
  })
}

// below all TestAcc functions

func testAccAwsExampleThingConfigName(rName string) string {
  return fmt.Sprintf(`
resource "aws_example_thing" "test" {
  name = %[1]q
}
`, rName)
}

Basic Acceptance Tests

Usually this test is implemented first. The test configuration should contain only required arguments (Required: true attributes) and it should check the values of all read-only attributes (Computed: true without Optional: true). If the resource supports it, it verifies import. It should NOT perform other TestStep such as updates or verify recreation.

These are typically named TestAccAws{SERVICE}{THING}_basic, e.g. TestAccAwsCloudWatchDashboard_basic

For example:

func TestAccAwsExampleThing_basic(t *testing.T) {
  rName := acctest.RandomWithPrefix("tf-acc-test")
  resourceName := "aws_example_thing.test"

  resource.ParallelTest(t, resource.TestCase{
    PreCheck:     func() { testAccPreCheck(t) },
    Providers:    testAccProviders,
    CheckDestroy: testAccCheckAwsExampleThingDestroy,
    Steps: []resource.TestStep{
      {
        Config: testAccAwsExampleThingConfigName(rName),
        Check: resource.ComposeTestCheckFunc(
          testAccCheckAwsExampleThingExists(resourceName),
          testAccCheckResourceAttrRegionalARN(resourceName, "arn", "example", fmt.Sprintf("thing/%s", rName)),
          resource.TestCheckResourceAttr(resourceName, "description", ""),
          resource.TestCheckResourceAttr(resourceName, "name", rName),
        ),
      },
      {
        ResourceName:      resourceName,
        ImportState:       true,
        ImportStateVerify: true,
      },
    },
  })
}

// below all TestAcc functions

func testAccAwsExampleThingConfigName(rName string) string {
  return fmt.Sprintf(`
resource "aws_example_thing" "test" {
  name = %[1]q
}
`, rName)
}

PreChecks

Acceptance test cases have a PreCheck. The PreCheck ensures that the testing environment meets certain preconditions. If the environment does not meet the preconditions, Go skips the test. Skipping a test avoids reporting a failure and wasting resources where the test cannot succeed.

Here is an example of the default PreCheck:

func TestAccAwsExampleThing_basic(t *testing.T) {
  rName := acctest.RandomWithPrefix("tf-acc-test")
  resourceName := "aws_example_thing.test"

  resource.ParallelTest(t, resource.TestCase{
    PreCheck:     func() { testAccPreCheck(t) },
    // ... additional checks follow ...
  })
}

Extend the default PreCheck by adding calls to functions in the anonymous PreCheck function. The functions can be existing functions in the provider or custom functions you add for new capabilities.

Standard Provider PreChecks

If you add a new test that has preconditions which are checked by an existing provider function, use that standard PreCheck instead of creating a new one. Some existing tests are missing standard PreChecks and you can help by adding them where appropriate.

These are some of the standard provider PreChecks:

  • testAccPartitionHasServicePreCheck(serviceId string, t *testing.T) checks whether the current partition lists the service as part of its offerings. Note: AWS may not add new or public preview services to the service list immediately. This function will return a false positive in that case.
  • testAccOrganizationsAccountPreCheck(t *testing.T) checks whether the current account can perform AWS Organizations tests.
  • testAccAlternateAccountPreCheck(t *testing.T) checks whether the environment is set up for tests across accounts.
  • testAccMultipleRegionPreCheck(t *testing.T, regions int) checks whether the environment is set up for tests across regions.

This is an example of using a standard PreCheck function. For an established service, such as WAF or FSx, use testAccPartitionHasServicePreCheck() and the service endpoint ID to check that a partition supports the service.

func TestAccAwsExampleThing_basic(t *testing.T) {
  rName := acctest.RandomWithPrefix("tf-acc-test")
  resourceName := "aws_example_thing.test"

  resource.ParallelTest(t, resource.TestCase{
    PreCheck:     func() { testAccPreCheck(t); testAccPartitionHasServicePreCheck(waf.EndpointsID, t) },
    // ... additional checks follow ...
  })
}
Custom PreChecks

In situations where standard PreChecks do not test for the required preconditions, create a custom PreCheck.

Below is an example of adding a custom PreCheck function. For a new or preview service that AWS does not include in the partition service list yet, you can verify the existence of the service with a simple read-only request (e.g., list all X service things). (For acceptance tests of established services, use testAccPartitionHasServicePreCheck() instead.)

func TestAccAwsExampleThing_basic(t *testing.T) {
  rName := acctest.RandomWithPrefix("tf-acc-test")
  resourceName := "aws_example_thing.test"

  resource.ParallelTest(t, resource.TestCase{
    PreCheck:     func() { testAccPreCheck(t), testAccPreCheckAwsExample(t) },
    // ... additional checks follow ...
  })
}

func testAccPreCheckAwsExample(t *testing.T) {
	conn := testAccProvider.Meta().(*AWSClient).exampleconn
	input := &example.ListThingsInput{}
	_, err := conn.ListThings(input)
	if testAccPreCheckSkipError(err) {
		t.Skipf("skipping acceptance testing: %s", err)
	}
	if err != nil {
		t.Fatalf("unexpected PreCheck error: %s", err)
	}
}

Disappears Acceptance Tests

This test is generally implemented second. It is straightforward to setup once the basic test is passing since it can reuse that test configuration. It prevents a common bug report with Terraform resources that error when they can not be found (e.g. deleted outside Terraform).

These are typically named TestAccAws{SERVICE}{THING}_disappears, e.g. TestAccAwsCloudWatchDashboard_disappears

For example:

func TestAccAwsExampleThing_disappears(t *testing.T) {
  rName := acctest.RandomWithPrefix("tf-acc-test")
  resourceName := "aws_example_thing.test"

  resource.ParallelTest(t, resource.TestCase{
    PreCheck:     func() { testAccPreCheck(t) },
    Providers:    testAccProviders,
    CheckDestroy: testAccCheckAwsExampleThingDestroy,
    Steps: []resource.TestStep{
      {
        Config: testAccAwsExampleThingConfigName(rName),
        Check: resource.ComposeTestCheckFunc(
          testAccCheckAwsExampleThingExists(resourceName, &job),
          testAccCheckResourceDisappears(testAccProvider, resourceAwsExampleThing(), resourceName),
        ),
        ExpectNonEmptyPlan: true,
      },
    },
  })
}

If this test does fail, the fix for this is generally adding error handling immediately after the Read API call that catches the error and tells Terraform to remove the resource before returning the error:

output, err := conn.GetThing(input)

if isAWSErr(err, example.ErrCodeResourceNotFound, "") {
  log.Printf("[WARN] Example Thing (%s) not found, removing from state", d.Id())
  d.SetId("")
  return nil
}

if err != nil {
  return fmt.Errorf("error reading Example Thing (%s): %w", d.Id(), err)
}

For children resources that are encapsulated by a parent resource, it is also preferable to verify that removing the parent resource will not generate an error either. These are typically named TestAccAws{SERVICE}{THING}_disappears_{PARENT}, e.g. TestAccAwsRoute53ZoneAssociation_disappears_Vpc

func TestAccAwsExampleChildThing_disappears_ParentThing(t *testing.T) {
  rName := acctest.RandomWithPrefix("tf-acc-test")
  parentResourceName := "aws_example_parent_thing.test"
  resourceName := "aws_example_child_thing.test"

  resource.ParallelTest(t, resource.TestCase{
    PreCheck:     func() { testAccPreCheck(t) },
    Providers:    testAccProviders,
    CheckDestroy: testAccCheckAwsExampleChildThingDestroy,
    Steps: []resource.TestStep{
      {
        Config: testAccAwsExampleThingConfigName(rName),
        Check: resource.ComposeTestCheckFunc(
          testAccCheckAwsExampleThingExists(resourceName),
          testAccCheckResourceDisappears(testAccProvider, resourceAwsExampleParentThing(), parentResourceName),
        ),
        ExpectNonEmptyPlan: true,
      },
    },
  })
}

Per Attribute Acceptance Tests

These are typically named TestAccAws{SERVICE}{THING}_{ATTRIBUTE}, e.g. TestAccAwsCloudWatchDashboard_Name

For example:

func TestAccAwsExampleThing_Description(t *testing.T) {
  rName := acctest.RandomWithPrefix("tf-acc-test")
  resourceName := "aws_example_thing.test"

  resource.ParallelTest(t, resource.TestCase{
    PreCheck:     func() { testAccPreCheck(t) },
    Providers:    testAccProviders,
    CheckDestroy: testAccCheckAwsExampleThingDestroy,
    Steps: []resource.TestStep{
      {
        Config: testAccAwsExampleThingConfigDescription(rName, "description1"),
        Check: resource.ComposeTestCheckFunc(
          testAccCheckAwsExampleThingExists(resourceName),
          resource.TestCheckResourceAttr(resourceName, "description", "description1"),
        ),
      },
      {
        ResourceName:      resourceName,
        ImportState:       true,
        ImportStateVerify: true,
      },
      {
        Config: testAccAwsExampleThingConfigDescription(rName, "description2"),
        Check: resource.ComposeTestCheckFunc(
          testAccCheckAwsExampleThingExists(resourceName),
          resource.TestCheckResourceAttr(resourceName, "description", "description2"),
        ),
      },
    },
  })
}

// below all TestAcc functions

func testAccAwsExampleThingConfigDescription(rName string, description string) string {
  return fmt.Sprintf(`
resource "aws_example_thing" "test" {
  description = %[2]q
  name        = %[1]q
}
`, rName, description)
}

Cross-Account Acceptance Tests

When testing requires AWS infrastructure in a second AWS account, the below changes to the normal setup will allow the management or reference of resources and data sources across accounts:

  • In the PreCheck function, include testAccAlternateAccountPreCheck(t) to ensure a standardized set of information is required for cross-account testing credentials
  • Declare a providers variable at the top of the test function: var providers []*schema.Provider
  • Switch usage of Providers: testAccProviders to ProviderFactories: testAccProviderFactoriesAlternate(&providers)
  • Add testAccAlternateAccountProviderConfig() to the test configuration and use provider = awsalternate for cross-account resources. The resource that is the focus of the acceptance test should not use the alternate provider identification to simplify the testing setup.
  • For any TestStep that includes ImportState: true, add the Config that matches the previous TestStep Config

An example acceptance test implementation can be seen below:

func TestAccAwsExample_basic(t *testing.T) {
  var providers []*schema.Provider
  resourceName := "aws_example.test"

  resource.ParallelTest(t, resource.TestCase{
    PreCheck: func() {
      testAccPreCheck(t)
      testAccAlternateAccountPreCheck(t)
    },
    ProviderFactories: testAccProviderFactoriesAlternate(&providers),
    CheckDestroy:      testAccCheckAwsExampleDestroy,
    Steps: []resource.TestStep{
      {
        Config: testAccAwsExampleConfig(),
        Check: resource.ComposeTestCheckFunc(
          testAccCheckAwsExampleExists(resourceName),
          // ... additional checks ...
        ),
      },
      {
        Config:            testAccAwsExampleConfig(),
        ResourceName:      resourceName,
        ImportState:       true,
        ImportStateVerify: true,
      },
    },
  })
}

func testAccAwsExampleConfig() string {
  return testAccAlternateAccountProviderConfig() + fmt.Sprintf(`
# Cross account resources should be handled by the cross account provider.
# The standardized provider block to use is awsalternate as seen below.
resource "aws_cross_account_example" "test" {
  provider = awsalternate

  # ... configuration ...
}

# The resource that is the focus of the testing should be handled by the default provider,
# which is automatically done by not specifying the provider configuration in the resource.
resource "aws_example" "test" {
  # ... configuration ...
}
`)
}

Searching for usage of testAccAlternateAccountPreCheck in the codebase will yield real world examples of this setup in action.

Cross-Region Acceptance Tests

When testing requires AWS infrastructure in a second or third AWS region, the below changes to the normal setup will allow the management or reference of resources and data sources across regions:

  • In the PreCheck function, include testAccMultipleRegionPreCheck(t, ###) to ensure a standardized set of information is required for cross-region testing configuration. If the infrastructure in the second AWS region is also in a second AWS account also include testAccAlternateAccountPreCheck(t)
  • Declare a providers variable at the top of the test function: var providers []*schema.Provider
  • Switch usage of Providers: testAccProviders to ProviderFactories: testAccProviderFactoriesMultipleRegion(&providers, 2) (where the last parameter is number of regions)
  • Add testAccMultipleRegionProviderConfig(###) to the test configuration and use provider = awsalternate (and potentially provider = awsthird) for cross-region resources. The resource that is the focus of the acceptance test should not use the alternative providers to simplify the testing setup. If the infrastructure in the second AWS region is also in a second AWS account use testAccAlternateAccountAlternateRegionProviderConfig() instead
  • For any TestStep that includes ImportState: true, add the Config that matches the previous TestStep Config

An example acceptance test implementation can be seen below:

func TestAccAwsExample_basic(t *testing.T) {
  var providers []*schema.Provider
  resourceName := "aws_example.test"

  resource.ParallelTest(t, resource.TestCase{
    PreCheck: func() {
      testAccPreCheck(t)
      testAccMultipleRegionPreCheck(t, 2)
    },
    ProviderFactories: testAccProviderFactoriesMultipleRegion(&providers, 2),
    CheckDestroy:      testAccCheckAwsExampleDestroy,
    Steps: []resource.TestStep{
      {
        Config: testAccAwsExampleConfig(),
        Check: resource.ComposeTestCheckFunc(
          testAccCheckAwsExampleExists(resourceName),
          // ... additional checks ...
        ),
      },
      {
        Config:            testAccAwsExampleConfig(),
        ResourceName:      resourceName,
        ImportState:       true,
        ImportStateVerify: true,
      },
    },
  })
}

func testAccAwsExampleConfig() string {
  return testAccMultipleRegionProviderConfig(2) + fmt.Sprintf(`
# Cross region resources should be handled by the cross region provider.
# The standardized provider is awsalternate as seen below.
resource "aws_cross_region_example" "test" {
  provider = awsalternate

  # ... configuration ...
}

# The resource that is the focus of the testing should be handled by the default provider,
# which is automatically done by not specifying the provider configuration in the resource.
resource "aws_example" "test" {
  # ... configuration ...
}
`)
}

Searching for usage of testAccMultipleRegionPreCheck in the codebase will yield real world examples of this setup in action.

Service-Specific Region Acceptance Testing

Certain AWS service APIs are only available in specific AWS regions. For example as of this writing, the pricing service is available in ap-south-1 and us-east-1, but no other regions or partitions. When encountering these types of services, the acceptance testing can be setup to automatically detect the correct region(s), while skipping the testing in unsupported partitions.

To prepare the shared service functionality, create a file named aws/{SERVICE}_test.go. A starting example with the Pricing service (aws/pricing_test.go):

package aws

import (
  "context"
  "sync"
  "testing"

  "github.com/aws/aws-sdk-go/aws/endpoints"
  "github.com/aws/aws-sdk-go/service/pricing"
  "github.com/hashicorp/terraform-plugin-sdk/v2/diag"
  "github.com/hashicorp/terraform-plugin-sdk/v2/helper/schema"
  "github.com/hashicorp/terraform-plugin-sdk/v2/terraform"
)

// testAccPricingRegion is the chosen Pricing testing region
//
// Cached to prevent issues should multiple regions become available.
var testAccPricingRegion string

// testAccProviderPricing is the Pricing provider instance
//
// This Provider can be used in testing code for API calls without requiring
// the use of saving and referencing specific ProviderFactories instances.
//
// testAccPreCheckPricing(t) must be called before using this provider instance.
var testAccProviderPricing *schema.Provider

// testAccProviderPricingConfigure ensures the provider is only configured once
var testAccProviderPricingConfigure sync.Once

// testAccPreCheckPricing verifies AWS credentials and that Pricing is supported
func testAccPreCheckPricing(t *testing.T) {
  testAccPartitionHasServicePreCheck(pricing.EndpointsID, t)

  // Since we are outside the scope of the Terraform configuration we must
  // call Configure() to properly initialize the provider configuration.
  testAccProviderPricingConfigure.Do(func() {
    testAccProviderPricing = Provider()

    config := map[string]interface{}{
      "region": testAccGetPricingRegion(),
    }

    diags := testAccProviderPricing.Configure(context.Background(), terraform.NewResourceConfigRaw(config))

    if diags != nil && diags.HasError() {
      for _, d := range diags {
        if d.Severity == diag.Error {
          t.Fatalf("error configuring Pricing provider: %s", d.Summary)
        }
      }
    }
  })
}

// testAccPricingRegionProviderConfig is the Terraform provider configuration for Pricing region testing
//
// Testing Pricing assumes no other provider configurations
// are necessary and overwrites the "aws" provider configuration.
func testAccPricingRegionProviderConfig() string {
  return testAccRegionalProviderConfig(testAccGetPricingRegion())
}

// testAccGetPricingRegion returns the Pricing region for testing
func testAccGetPricingRegion() string {
  if testAccPricingRegion != "" {
    return testAccPricingRegion
  }

  if rs, ok := endpoints.RegionsForService(endpoints.DefaultPartitions(), testAccGetPartition(), pricing.ServiceName); ok {
    // return available region (random if multiple)
    for regionID := range rs {
      testAccPricingRegion = regionID
      return testAccPricingRegion
    }
  }

  testAccPricingRegion = testAccGetRegion()

  return testAccPricingRegion
}

For the resource or data source acceptance tests, the key items to adjust are:

  • Ensure TestCase uses ProviderFactories: testAccProviderFactories instead of Providers: testAccProviders
  • Add the call for the new PreCheck function (keeping testAccPreCheck(t)), e.g. PreCheck: func() { testAccPreCheck(t); testAccPreCheckPricing(t) },
  • If the testing is for a managed resource with a CheckDestroy function, ensure it uses the new provider instance, e.g. testAccProviderPricing, instead of testAccProvider.
  • If the testing is for a managed resource with a Check...Exists function, ensure it uses the new provider instance, e.g. testAccProviderPricing, instead of testAccProvider.
  • In each TestStep configuration, ensure the new provider configuration function is called, e.g.
func testAccDataSourceAwsPricingProductConfigRedshift() string {
  return composeConfig(
    testAccPricingRegionProviderConfig(),
    `
# ... test configuration ...
`)
}

If the testing configurations require more than one region, reach out to the maintainers for further assistance.

Data Source Acceptance Testing

Writing acceptance testing for data sources is similar to resources, with the biggest changes being:

  • Adding DataSource to the test and configuration naming, such as TestAccAwsExampleThingDataSource_Filter
  • The basic test may be named after the easiest lookup attribute instead, e.g. TestAccAwsExampleThingDataSource_Name
  • No disappears testing
  • Almost all checks should be done with resource.TestCheckResourceAttrPair() to compare the data source attributes to the resource attributes
  • The usage of an additional dataSourceName variable to store a data source reference, e.g. data.aws_example_thing.test

Data sources testing should still utilize the CheckDestroy function of the resource, just to continue verifying that there are no dangling AWS resources after a test is ran.

Please note that we do not recommend re-using test configurations between resources and their associated data source as it is harder to discover testing regressions. Authors are encouraged to potentially implement similar "base" configurations though.

For example:

func TestAccAwsExampleThingDataSource_Name(t *testing.T) {
  rName := acctest.RandomWithPrefix("tf-acc-test")
  dataSourceName := "data.aws_example_thing.test"
  resourceName := "aws_example_thing.test"

  resource.ParallelTest(t, resource.TestCase{
    PreCheck:     func() { testAccPreCheck(t) },
    Providers:    testAccProviders,
    CheckDestroy: testAccCheckAwsExampleThingDestroy,
    Steps: []resource.TestStep{
      {
        Config: testAccAwsExampleThingDataSourceConfigName(rName),
        Check: resource.ComposeTestCheckFunc(
          testAccCheckAwsExampleThingExists(resourceName),
          resource.TestCheckResourceAttrPair(resourceName, "arn", dataSourceName, "arn"),
          resource.TestCheckResourceAttrPair(resourceName, "description", dataSourceName, "description"),
          resource.TestCheckResourceAttrPair(resourceName, "name", dataSourceName, "name"),
        ),
      },
    },
  })
}

// below all TestAcc functions

func testAccAwsExampleThingDataSourceConfigName(rName string) string {
  return fmt.Sprintf(`
resource "aws_example_thing" "test" {
  name = %[1]q
}

data "aws_example_thing" "test" {
  name = aws_example_thing.test.name
}
`, rName)
}

Acceptance Test Sweepers

When running the acceptance tests, especially when developing or troubleshooting Terraform resources, its possible for code bugs or other issues to prevent the proper destruction of AWS infrastructure. To prevent lingering resources from consuming quota or causing unexpected billing, the Terraform Plugin SDK supports the test sweeper framework to clear out an AWS region of all resources. This section is meant to augment the Extending Terraform documentation on test sweepers with Terraform AWS Provider specific details.

Running Test Sweepers

WARNING: Test Sweepers will destroy AWS infrastructure and backups in the target AWS account and region! These are designed to override any API deletion protection. Never run these outside a development AWS account that should be completely empty of resources.

To run the sweepers for all resources in us-west-2 and us-east-1 (default testing regions):

$ make sweep

To run a specific resource sweeper:

$ SWEEPARGS=-sweep-run=aws_example_thing make sweep

Writing Test Sweepers

The first step is to initialize the resource into the test sweeper framework:

func init() {
  resource.AddTestSweepers("aws_example_thing", &resource.Sweeper{
    Name: "aws_example_thing",
    F:    testSweepExampleThings,
    // Optionally
    Dependencies: []string{
      "aws_other_thing",
    },
  })
}

Then add the actual implementation. Preferably, if a paginated SDK call is available:

func testSweepExampleThings(region string) error {
  client, err := sharedClientForRegion(region)

  if err != nil {
    return fmt.Errorf("error getting client: %w", err)
  }

  conn := client.(*AWSClient).exampleconn
  input := &example.ListThingsInput{}
  var sweeperErrs *multierror.Error

  err = conn.ListThingsPages(input, func(page *example.ListThingsOutput, isLast bool) bool {
    if page == nil {
      return !isLast
    }

    for _, thing := range page.Things {
      id := aws.StringValue(thing.Id)
      input := &example.DeleteThingInput{
        Id: thing.Id,
      }

      log.Printf("[INFO] Deleting Example Thing: %s", id)
      _, err := conn.DeleteThing(input)

      if err != nil {
        sweeperErr := fmt.Errorf("error deleting Example Thing (%s): %w", id, err)
        log.Printf("[ERROR] %s", sweeperErr)
        sweeperErrs = multierror.Append(sweeperErrs, sweeperErr)
        continue
      }
    }

    return !isLast
  })

  if testSweepSkipSweepError(err) {
    log.Printf("[WARN] Skipping Example Thing sweep for %s: %s", region, err)
    return sweeperErrs.ErrorOrNil()
  }

  if err != nil {
    sweeperErrs = multierror.Append(sweeperErrs, fmt.Errorf("error retrieving Example Things: %w", err))
  }

  return sweeperErrs.ErrorOrNil()
}

Otherwise, if no paginated SDK call is available:

func testSweepExampleThings(region string) error {
  client, err := sharedClientForRegion(region)

  if err != nil {
    return fmt.Errorf("error getting client: %w", err)
  }

  conn := client.(*AWSClient).exampleconn
  input := &example.ListThingsInput{}
  var sweeperErrs *multierror.Error

  for {
    output, err := conn.ListThings(input)

    if testSweepSkipSweepError(err) {
      log.Printf("[WARN] Skipping Example Thing sweep for %s: %s", region, err)
      return sweeperErrs.ErrorOrNil()
    }

    if err != nil {
      sweeperErrs = multierror.Append(sweeperErrs, fmt.Errorf("error retrieving Example Thing: %w", err))
      return sweeperErrs
    }

    for _, thing := range output.Things {
      id := aws.StringValue(thing.Id)
      input := &example.DeleteThingInput{
        Id: thing.Id,
      }

      log.Printf("[INFO] Deleting Example Thing: %s", id)
      _, err := conn.DeleteThing(input)

      if err != nil {
        sweeperErr := fmt.Errorf("error deleting Example Thing (%s): %w", id, err)
        log.Printf("[ERROR] %s", sweeperErr)
        sweeperErrs = multierror.Append(sweeperErrs, sweeperErr)
        continue
      }
    }

    if aws.StringValue(output.NextToken) == "" {
      break
    }

    input.NextToken = output.NextToken
  }

  return sweeperErrs.ErrorOrNil()
}

Acceptance Test Checklists

There are several aspects to writing good acceptance tests. These checklists will help ensure effective testing from the design stage through to implementation details.

Basic Acceptance Test Design

These are basic principles to help guide the creation of acceptance tests.

  • Covers Changes: Every line of resource or data source code added or changed should be covered by one or more tests. For example, if a resource has two ways of functioning, tests should cover both possible paths. Nearly every codebase change needs test coverage to ensure functionality and prevent future regressions. If a bug or other problem prompted a fix, a test should be added that previously would have failed, especially if the report included a configuration.
  • Follows the Single Responsibility Principle: Every test should have a single responsibility and effectively test that responsibility. This may include individual tests for verifying basic functionality of the resource (Create, Read, Delete), separately verifying using and updating a single attribute in a resource, or separately changing between two attributes to verify two "modes"/"types" possible with a resource configuration. In following this principle, test configurations should be as simple as possible. For example, not including extra configuration unless it is necessary for the specific test.

Test Implementation

The below are required items that will be noted during submission review and prevent immediate merging:

  • Implements CheckDestroy: Resource testing should include a CheckDestroy function (typically named testAccCheckAws{SERVICE}{RESOURCE}Destroy) that calls the API to verify that the Terraform resource has been deleted or disassociated as appropriate. More information about CheckDestroy functions can be found in the Extending Terraform TestCase documentation.
  • Implements Exists Check Function: Resource testing should include a TestCheckFunc function (typically named testAccCheckAws{SERVICE}{RESOURCE}Exists) that calls the API to verify that the Terraform resource has been created or associated as appropriate. Preferably, this function will also accept a pointer to an API object representing the Terraform resource from the API response that can be set for potential usage in later TestCheckFunc. More information about these functions can be found in the Extending Terraform Custom Check Functions documentation.
  • Excludes Provider Declarations: Test configurations should not include provider "aws" {...} declarations. If necessary, only the provider declarations in provider_test.go should be used for multiple account/region or otherwise specialized testing.
  • Passes in us-west-2 Region: Tests default to running in us-west-2 and at a minimum should pass in that region or include necessary PreCheck functions to skip the test when ran outside an expected environment.
  • Uses resource.ParallelTest: Tests should utilize resource.ParallelTest() instead of resource.Test() except where serialized testing is absolutely required.
  • Uses fmt.Sprintf(): Test configurations preferably should to be separated into their own functions (typically named testAccAws{SERVICE}{RESOURCE}Config{PURPOSE}) that call fmt.Sprintf() for variable injection or a string const for completely static configurations. Test configurations should avoid var or other variable injection functionality such as text/template.
  • Uses Randomized Infrastructure Naming: Test configurations that utilize resources where a unique name is required should generate a random name. Typically this is created via rName := acctest.RandomWithPrefix("tf-acc-test") in the acceptance test function before generating the configuration.

For resources that support import, the additional item below is required that will be noted during submission review and prevent immediate merging:

  • Implements ImportState Testing: Tests should include an additional TestStep configuration that verifies resource import via ImportState: true and ImportStateVerify: true. This TestStep should be added to all possible tests for the resource to ensure that all infrastructure configurations are properly imported into Terraform.

The below are style-based items that may be noted during review and are recommended for simplicity, consistency, and quality assurance:

  • Uses Builtin Check Functions: Tests should utilize already available check functions, e.g. resource.TestCheckResourceAttr(), to verify values in the Terraform state over creating custom TestCheckFunc. More information about these functions can be found in the Extending Terraform Builtin Check Functions documentation.
  • Uses TestCheckResoureAttrPair() for Data Sources: Tests should utilize resource.TestCheckResourceAttrPair() to verify values in the Terraform state for data sources attributes to compare them with their expected resource attributes.
  • Excludes Timeouts Configurations: Test configurations should not include timeouts {...} configuration blocks except for explicit testing of customizable timeouts (typically very short timeouts with ExpectError).
  • Implements Default and Zero Value Validation: The basic test for a resource (typically named TestAccAws{SERVICE}{RESOURCE}_basic) should utilize available check functions, e.g. resource.TestCheckResourceAttr(), to verify default and zero values in the Terraform state for all attributes. Empty/missing configuration blocks can be verified with resource.TestCheckResourceAttr(resourceName, "{ATTRIBUTE}.#", "0") and empty maps with resource.TestCheckResourceAttr(resourceName, "{ATTRIBUTE}.%", "0")

Avoid Hard Coding

Avoid hard coding values in acceptance test checks and configurations for consistency and testing flexibility. Resource testing is expected to pass across multiple AWS environments supported by the Terraform AWS Provider (e.g. AWS Standard and AWS GovCloud (US)). Contributors are not expected or required to perform testing outside of AWS Standard, e.g. running only in the us-west-2 region is perfectly acceptable. However, contributors are expected to avoid hard coding with these guidelines.

Hardcoded Account IDs

  • Uses Account Data Sources: Any hardcoded account numbers in configuration, e.g., 137112412989, should be replaced with a data source. Depending on the situation, there are several data sources for account IDs including:
  • Uses Account Test Checks: Any check required to verify an AWS Account ID of the current testing account or another account should use one of the following available helper functions over the usage of resource.TestCheckResourceAttrSet() and resource.TestMatchResourceAttr():
    • testAccCheckResourceAttrAccountID(): Validates the state value equals the AWS Account ID of the current account running the test. This is the most common implementation.
    • testAccMatchResourceAttrAccountID(): Validates the state value matches any AWS Account ID (e.g. a 12 digit number). This is typically only used in data source testing of AWS managed components.

Here's an example of using aws_caller_identity:

data "aws_caller_identity" "current" {}

resource "aws_backup_selection" "test" {
  plan_id      = aws_backup_plan.test.id
  name         = "tf_acc_test_backup_selection_%[1]d"
  iam_role_arn = "arn:${data.aws_partition.current.partition}:iam::${data.aws_caller_identity.current.account_id}:role/service-role/AWSBackupDefaultServiceRole"
}

Hardcoded AMI IDs

  • Uses aws_ami Data Source: Any hardcoded AMI ID configuration, e.g. ami-12345678, should be replaced with the aws_ami data source pointing to an Amazon Linux image. The codebase includes test configuration helper functions to simplify these lookups:
    • testAccLatestAmazonLinuxHvmEbsAmiConfig(): The recommended AMI for most situations, using Amazon Linux, HVM virtualization, and EBS storage. To reference the AMI ID in the test configuration: data.aws_ami.amzn-ami-minimal-hvm-ebs.id.
    • testAccLatestAmazonLinuxHvmInstanceStoreAmiConfig(): AMI lookup using Amazon Linux, HVM virtualization, and Instance Store storage. Should only be used in testing that requires Instance Store storage rather than EBS. To reference the AMI ID in the test configuration: data.aws_ami.amzn-ami-minimal-hvm-instance-store.id.
    • testAccLatestAmazonLinuxPvEbsAmiConfig(): AMI lookup using Amazon Linux, Paravirtual virtualization, and EBS storage. Should only be used in testing that requires Paravirtual over Hardware Virtual Machine (HVM) virtualization. To reference the AMI ID in the test configuration: data.aws_ami.amzn-ami-minimal-pv-ebs.id.
    • testAccLatestAmazonLinuxPvInstanceStoreAmiConfig: AMI lookup using Amazon Linux, Paravirtual virtualization, and Instance Store storage. Should only be used in testing that requires Paravirtual virtualization over HVM and Instance Store storage over EBS. To reference the AMI ID in the test configuration: data.aws_ami.amzn-ami-minimal-pv-instance-store.id.
    • testAccLatestWindowsServer2016CoreAmiConfig(): AMI lookup using Windows Server 2016 Core, HVM virtualization, and EBS storage. Should only be used in testing that requires Windows. To reference the AMI ID in the test configuration: data.aws_ami.win2016core-ami.id.

Here's an example of using testAccLatestAmazonLinuxHvmEbsAmiConfig() and data.aws_ami.amzn-ami-minimal-hvm-ebs.id:

func testAccLaunchConfigurationDataSourceConfig_basic(rName string) string {
	return composeConfig(testAccLatestAmazonLinuxHvmEbsAmiConfig(), fmt.Sprintf(`
resource "aws_launch_configuration" "test" {
  name          = %[1]q
  image_id      = data.aws_ami.amzn-ami-minimal-hvm-ebs.id
  instance_type = "m1.small"
}
`, rName))
}

Hardcoded Availability Zones

  • Uses aws_availability_zones Data Source: Any hardcoded AWS Availability Zone configuration, e.g. us-west-2a, should be replaced with the aws_availability_zones data source. Use the convenience function called testAccAvailableAZsNoOptInConfig() (defined in resource_aws_instance_test.go) to declare data "aws_availability_zones" "available" {...}. You can then reference the data source via data.aws_availability_zones.available.names[0] or data.aws_availability_zones.available.names[count.index] in resources utilizing count.

Here's an example of using testAccAvailableAZsNoOptInConfig() and data.aws_availability_zones.available.names[0]:

func testAccAwsInstanceVpcConfigBasic(rName string) string {
	return composeConfig(testAccAvailableAZsNoOptInConfig(), fmt.Sprintf(`
resource "aws_subnet" "test" {
  availability_zone = data.aws_availability_zones.available.names[0]
  cidr_block        = "10.0.0.0/24"
  vpc_id            = aws_vpc.test.id
}
`, rName))
}

Hardcoded Versions

  • Uses Version Data Sources: Hardcoded versions, e.g. RDS MySQL Engine Version 5.7.42, should be removed (which means the AWS-defined default version will be used) or replaced with a list of preferred versions using a data source. Because versions change over times and version offerings vary from region to region and partition to partition, using the default version or providing a list of preferences ensures a version will be available. Depending on the situation, there are several data sources for versions, including:

Here's an example of using aws_rds_engine_version and data.aws_rds_engine_version.default.version:

data "aws_rds_engine_version" "default" {
  engine = "mysql"
}

data "aws_rds_orderable_db_instance" "test" {
  engine                     = data.aws_rds_engine_version.default.engine
  engine_version             = data.aws_rds_engine_version.default.version
  preferred_instance_classes = ["db.t3.small", "db.t2.small", "db.t2.medium"]
}

resource "aws_db_instance" "bar" {
  engine               = data.aws_rds_engine_version.default.engine
  engine_version       = data.aws_rds_engine_version.default.version
  instance_class       = data.aws_rds_orderable_db_instance.test.instance_class
  skip_final_snapshot  = true
  parameter_group_name = "default.${data.aws_rds_engine_version.default.parameter_group_family}"
}

Hardcoded Instance Types

  • Uses Instance Type Data Source: Singular hardcoded instance types and classes, e.g., t2.micro and db.t2.micro, should be replaced with a list of preferences using a data source. Because offerings vary from region to region and partition to partition, providing a list of preferences dramatically improves the likelihood that one of the options will be available. Depending on the situation, there are several data sources for instance types and classes, including:

Here's an example of using testAccAvailableEc2InstanceTypeForRegion() and data.aws_ec2_instance_type_offering.available.instance_type:

func testAccAWSSpotInstanceRequestConfig(rInt int) string {
	return composeConfig(testAccAvailableEc2InstanceTypeForRegion("t3.micro", "t2.micro"),
		fmt.Sprintf(`
resource "aws_spot_instance_request" "test" {
  instance_type        = data.aws_ec2_instance_type_offering.available.instance_type
  spot_price           = "0.05"
  wait_for_fulfillment = true
}
`, rInt))
}

Here's an example of using aws_rds_orderable_db_instance and data.aws_rds_orderable_db_instance.test.instance_class:

data "aws_rds_orderable_db_instance" "test" {
  engine                     = "mysql"
  engine_version             = "5.7.31"
  preferred_instance_classes = ["db.t3.micro", "db.t2.micro", "db.t3.small"]
}

resource "aws_db_instance" "test" {
  engine              = data.aws_rds_orderable_db_instance.test.engine
  engine_version      = data.aws_rds_orderable_db_instance.test.engine_version
  instance_class      = data.aws_rds_orderable_db_instance.test.instance_class
  skip_final_snapshot = true
  username            = "test"
}

Hardcoded Partition DNS Suffix

  • Uses aws_partition Data Source: Any hardcoded DNS suffix configuration, e.g. the amazonaws.com in a ec2.amazonaws.com service principal, should be replaced with the aws_partition data source. A common pattern is declaring data "aws_partition" "current" {} and referencing it via data.aws_partition.current.dns_suffix.

Here's an example of using aws_partition and data.aws_partition.current.dns_suffix:

data "aws_partition" "current" {}

resource "aws_iam_role" "test" {
  assume_role_policy = <<POLICY
{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Sid": "",
      "Effect": "Allow",
      "Principal": {
        "Service": "cloudtrail.${data.aws_partition.current.dns_suffix}"
      },
      "Action": "sts:AssumeRole"
    }
  ]
}
POLICY
}

Hardcoded Partition in ARN

  • Uses aws_partition Data Source: Any hardcoded AWS Partition configuration, e.g. the aws in a arn:aws:SERVICE:REGION:ACCOUNT:RESOURCE ARN, should be replaced with the aws_partition data source. A common pattern is declaring data "aws_partition" "current" {} and referencing it via data.aws_partition.current.partition.

  • Uses Builtin ARN Check Functions: Tests should utilize available ARN check functions to validate ARN attribute values in the Terraform state over resource.TestCheckResourceAttrSet() and resource.TestMatchResourceAttr():

    • testAccCheckResourceAttrRegionalARN() verifies an ARN matches the account ID and region of the test execution and an exact resource value
    • testAccMatchResourceAttrRegionalARN() verifies an ARN matches the account ID and region of the test execution and a regular expression of the resource value
    • testAccCheckResourceAttrGlobalARN() verifies an ARN matches the account ID of the test execution and an exact resource value
    • testAccMatchResourceAttrGlobalARN() verifies an ARN matches the account ID of the test execution and a regular expression of the resource value

Here's an example of using aws_partition and data.aws_partition.current.partition:

data "aws_partition" "current" {}

resource "aws_iam_role_policy_attachment" "test" {
  policy_arn = "arn:${data.aws_partition.current.partition}:iam::aws:policy/service-role/AmazonRDSEnhancedMonitoringRole"
  role       = aws_iam_role.test.name
}

Hardcoded Region

  • Uses aws_region Data Source: Any hardcoded AWS Region configuration, e.g. us-west-2, should be replaced with the aws_region data source. A common pattern is declaring data "aws_region" "current" {} and referencing it via data.aws_region.current.name

Here's an example of using aws_region and data.aws_region.current.name:

data "aws_region" "current" {}

resource "aws_route53_zone" "test" {
  vpc {
    vpc_id     = aws_vpc.test.id
    vpc_region = data.aws_region.current.name
  }
}

Hardcoded Spot Price

  • Uses aws_ec2_spot_price Data Source: Any hardcoded spot prices, e.g. 0.05, should be replaced with the aws_ec2_spot_price data source. A common pattern is declaring data "aws_ec2_spot_price" "current" {} and referencing it via data.aws_ec2_spot_price.current.spot_price.

Here's an example of using aws_ec2_spot_price and data.aws_ec2_spot_price.current.spot_price:

data "aws_ec2_spot_price" "current" {
  instance_type = "t3.medium"

  filter {
    name   = "product-description"
    values = ["Linux/UNIX"]
  }
}

resource "aws_spot_fleet_request" "test" {
  spot_price      = data.aws_ec2_spot_price.current.spot_price
  target_capacity = 2
}