opentofu/internal/command/test.go

815 lines
25 KiB
Go

package command
import (
"context"
"fmt"
"path"
"sort"
"strings"
"time"
"github.com/hashicorp/terraform/internal/addrs"
"github.com/hashicorp/terraform/internal/backend"
"github.com/hashicorp/terraform/internal/command/arguments"
"github.com/hashicorp/terraform/internal/command/views"
"github.com/hashicorp/terraform/internal/configs"
"github.com/hashicorp/terraform/internal/logging"
"github.com/hashicorp/terraform/internal/moduletest"
"github.com/hashicorp/terraform/internal/plans"
"github.com/hashicorp/terraform/internal/states"
"github.com/hashicorp/terraform/internal/terraform"
"github.com/hashicorp/terraform/internal/tfdiags"
)
type TestCommand struct {
Meta
}
func (c *TestCommand) Help() string {
helpText := `
Usage: terraform [global options] test [options]
Executes automated integration tests against the current Terraform
configuration.
Terraform will search for .tftest.hcl files within the current configuration
and testing directories. Terraform will then execute the testing run blocks
within any testing files in order, and verify conditional checks and
assertions against the created infrastructure.
This command creates real infrastructure and will attempt to clean up the
testing infrastructure on completion. Monitor the output carefully to ensure
this cleanup process is successful.
Options:
-filter=testfile If specified, Terraform will only execute the test files
specified by this flag. You can use this option multiple
times to execute more than one test file.
-json If specified, machine readable output will be printed in
JSON format
-test-directory=path Set the Terraform test directory, defaults to "tests".
-var 'foo=bar' Set a value for one of the input variables in the root
module of the configuration. Use this option more than
once to set more than one variable.
-var-file=filename Load variable values from the given file, in addition
to the default files terraform.tfvars and *.auto.tfvars.
Use this option more than once to include more than one
variables file.
-verbose Print the plan or state for each test run block as it
executes.
`
return strings.TrimSpace(helpText)
}
func (c *TestCommand) Synopsis() string {
return "Execute integration tests for Terraform modules"
}
func (c *TestCommand) Run(rawArgs []string) int {
var diags tfdiags.Diagnostics
common, rawArgs := arguments.ParseView(rawArgs)
c.View.Configure(common)
args, diags := arguments.ParseTest(rawArgs)
if diags.HasErrors() {
c.View.Diagnostics(diags)
c.View.HelpPrompt("test")
return 1
}
view := views.NewTest(args.ViewType, c.View)
config, configDiags := c.loadConfigWithTests(".", args.TestDirectory)
diags = diags.Append(configDiags)
if configDiags.HasErrors() {
view.Diagnostics(nil, nil, diags)
return 1
}
var fileDiags tfdiags.Diagnostics
suite := moduletest.Suite{
Files: func() map[string]*moduletest.File {
files := make(map[string]*moduletest.File)
if len(args.Filter) > 0 {
for _, name := range args.Filter {
file, ok := config.Module.Tests[name]
if !ok {
// If the filter is invalid, we'll simply skip this
// entry and print a warning. But we could still execute
// any other tests within the filter.
fileDiags.Append(tfdiags.Sourceless(
tfdiags.Warning,
"Unknown test file",
fmt.Sprintf("The specified test file, %s, could not be found.", name)))
continue
}
var runs []*moduletest.Run
for ix, run := range file.Runs {
runs = append(runs, &moduletest.Run{
Config: run,
Index: ix,
Name: run.Name,
})
}
files[name] = &moduletest.File{
Config: file,
Name: name,
Runs: runs,
}
}
return files
}
// Otherwise, we'll just do all the tests in the directory!
for name, file := range config.Module.Tests {
var runs []*moduletest.Run
for ix, run := range file.Runs {
runs = append(runs, &moduletest.Run{
Config: run,
Index: ix,
Name: run.Name,
})
}
files[name] = &moduletest.File{
Config: file,
Name: name,
Runs: runs,
}
}
return files
}(),
}
diags = diags.Append(fileDiags)
if fileDiags.HasErrors() {
view.Diagnostics(nil, nil, diags)
return 1
}
// Users can also specify variables via the command line, so we'll parse
// all that here.
var items []rawFlag
for _, variable := range args.Vars.All() {
items = append(items, rawFlag{
Name: variable.Name,
Value: variable.Value,
})
}
c.variableArgs = rawFlags{items: &items}
variables, variableDiags := c.collectVariableValues()
diags = diags.Append(variableDiags)
if variableDiags.HasErrors() {
view.Diagnostics(nil, nil, diags)
return 1
}
// We have two levels of interrupt here. A 'stop' and a 'cancel'. A 'stop'
// is a soft request to stop. We'll finish the current test, do the tidy up,
// but then skip all remaining tests and run blocks. A 'cancel' is a hard
// request to stop now. We'll cancel the current operation immediately
// even if it's a delete operation, and we won't clean up any infrastructure
// if we're halfway through a test. We'll print details explaining what was
// stopped so the user can do their best to recover from it.
runningCtx, done := context.WithCancel(context.Background())
stopCtx, stop := context.WithCancel(runningCtx)
cancelCtx, cancel := context.WithCancel(context.Background())
runner := &TestRunner{
command: c,
Suite: &suite,
Config: config,
View: view,
CancelledCtx: cancelCtx,
StoppedCtx: stopCtx,
// Just to be explicit, we'll set the following fields even though they
// default to these values.
Cancelled: false,
Stopped: false,
Verbose: args.Verbose,
}
view.Abstract(&suite)
go func() {
defer logging.PanicHandler()
defer done()
defer stop()
defer cancel()
runner.Start(variables)
}()
// Wait for the operation to complete, or for an interrupt to occur.
select {
case <-c.ShutdownCh:
// Nice request to be cancelled.
view.Interrupted()
runner.Stopped = true
stop()
select {
case <-c.ShutdownCh:
// The user pressed it again, now we have to get it to stop as
// fast as possible.
view.FatalInterrupt()
runner.Cancelled = true
cancel()
// We'll wait 5 seconds for this operation to finish now, regardless
// of whether it finishes successfully or not.
select {
case <-runningCtx.Done():
case <-time.After(5 * time.Second):
}
case <-runningCtx.Done():
// The application finished nicely after the request was stopped.
}
case <-runningCtx.Done():
// tests finished normally with no interrupts.
}
if runner.Cancelled {
// Don't print out the conclusion if the test was cancelled.
return 1
}
view.Conclusion(&suite)
if suite.Status != moduletest.Pass {
return 1
}
return 0
}
// test runner
type TestRunner struct {
command *TestCommand
Suite *moduletest.Suite
Config *configs.Config
View views.Test
// Stopped and Cancelled track whether the user requested the testing
// process to be interrupted. Stopped is a nice graceful exit, we'll still
// tidy up any state that was created and mark the tests with relevant
// `skipped` status updates. Cancelled is a hard stop right now exit, we
// won't attempt to clean up any state left hanging, and tests will just
// be left showing `pending` as the status. We will still print out the
// destroy summary diagnostics that tell the user what state has been left
// behind and needs manual clean up.
Stopped bool
Cancelled bool
// StoppedCtx and CancelledCtx allow in progress Terraform operations to
// respond to external calls from the test command.
StoppedCtx context.Context
CancelledCtx context.Context
// Verbose tells the runner to print out plan files during each test run.
Verbose bool
}
func (runner *TestRunner) Start(globals map[string]backend.UnparsedVariableValue) {
var files []string
for name := range runner.Suite.Files {
files = append(files, name)
}
sort.Strings(files) // execute the files in alphabetical order
runner.Suite.Status = moduletest.Pass
for _, name := range files {
if runner.Cancelled {
return
}
file := runner.Suite.Files[name]
runner.ExecuteTestFile(file, globals)
runner.Suite.Status = runner.Suite.Status.Merge(file.Status)
}
}
func (runner *TestRunner) ExecuteTestFile(file *moduletest.File, globals map[string]backend.UnparsedVariableValue) {
mgr := new(TestStateManager)
mgr.runner = runner
mgr.State = states.NewState()
defer mgr.cleanupStates(file, globals)
file.Status = file.Status.Merge(moduletest.Pass)
for _, run := range file.Runs {
if runner.Cancelled {
// This means a hard stop has been requested, in this case we don't
// even stop to mark future tests as having been skipped. They'll
// just show up as pending in the printed summary.
return
}
if runner.Stopped {
// Then the test was requested to be stopped, so we just mark each
// following test as skipped and move on.
run.Status = moduletest.Skip
continue
}
if file.Status == moduletest.Error {
// If the overall test file has errored, we don't keep trying to
// execute tests. Instead, we mark all remaining run blocks as
// skipped.
run.Status = moduletest.Skip
continue
}
if run.Config.ConfigUnderTest != nil {
// Then we want to execute a different module under a kind of
// sandbox.
state := runner.ExecuteTestRun(mgr, run, file, states.NewState(), run.Config.ConfigUnderTest, globals)
mgr.States = append(mgr.States, &TestModuleState{
State: state,
Run: run,
})
} else {
mgr.State = runner.ExecuteTestRun(mgr, run, file, mgr.State, runner.Config, globals)
}
file.Status = file.Status.Merge(run.Status)
}
runner.View.File(file)
for _, run := range file.Runs {
runner.View.Run(run, file)
}
}
func (runner *TestRunner) ExecuteTestRun(mgr *TestStateManager, run *moduletest.Run, file *moduletest.File, state *states.State, config *configs.Config, globals map[string]backend.UnparsedVariableValue) *states.State {
if runner.Cancelled {
// Don't do anything, just give up and return immediately.
// The surrounding functions should stop this even being called, but in
// case of race conditions or something we can still verify this.
return state
}
if runner.Stopped {
// Basically the same as above, except we'll be a bit nicer.
run.Status = moduletest.Skip
return state
}
targets, diags := run.GetTargets()
run.Diagnostics = run.Diagnostics.Append(diags)
replaces, diags := run.GetReplaces()
run.Diagnostics = run.Diagnostics.Append(diags)
references, diags := run.GetReferences()
run.Diagnostics = run.Diagnostics.Append(diags)
if run.Diagnostics.HasErrors() {
run.Status = moduletest.Error
return state
}
ctx, plan, state, diags := runner.execute(mgr, run, file, config, state, &terraform.PlanOpts{
Mode: func() plans.Mode {
switch run.Config.Options.Mode {
case configs.RefreshOnlyTestMode:
return plans.RefreshOnlyMode
default:
return plans.NormalMode
}
}(),
Targets: targets,
ForceReplace: replaces,
SkipRefresh: !run.Config.Options.Refresh,
ExternalReferences: references,
}, run.Config.Command, globals)
diags = run.ValidateExpectedFailures(diags)
run.Diagnostics = run.Diagnostics.Append(diags)
if runner.Cancelled {
// Print out the diagnostics from the run now, since it was cancelled
// the normal set of diagnostics will not be printed otherwise.
runner.View.Diagnostics(run, file, run.Diagnostics)
run.Status = moduletest.Error
return state
}
if diags.HasErrors() {
run.Status = moduletest.Error
return state
}
if runner.Stopped {
run.Status = moduletest.Skip
return state
}
// If the user wants to render the plans as part of the test output, we
// track that here.
if runner.Verbose {
schemas, diags := ctx.Schemas(config, state)
// If we're going to fail to render the plan, let's not fail the overall
// test. It can still have succeeded. So we'll add the diagnostics, but
// still report the test status as a success.
if diags.HasErrors() {
// This is very unlikely.
diags = diags.Append(tfdiags.Sourceless(
tfdiags.Warning,
"Failed to print verbose output",
fmt.Sprintf("Terraform failed to print the verbose output for %s, other diagnostics will contain more details as to why.", path.Join(file.Name, run.Name))))
} else {
run.Verbose = &moduletest.Verbose{
Plan: plan,
State: state,
Config: config,
Providers: schemas.Providers,
Provisioners: schemas.Provisioners,
}
}
run.Diagnostics = run.Diagnostics.Append(diags)
}
variables, diags := buildInputVariablesForAssertions(run, file, config, globals)
run.Diagnostics = run.Diagnostics.Append(diags)
if diags.HasErrors() {
run.Status = moduletest.Error
return state
}
if run.Config.Command == configs.ApplyTestCommand {
ctx.TestContext(config, state, plan, variables).EvaluateAgainstState(run)
return state
}
ctx.TestContext(config, plan.PlannedState, plan, variables).EvaluateAgainstPlan(run)
return state
}
// execute executes Terraform plan and apply operations for the given arguments.
//
// The command argument decides whether it executes only a plan or also applies
// the plan it creates during the planning.
func (runner *TestRunner) execute(mgr *TestStateManager, run *moduletest.Run, file *moduletest.File, config *configs.Config, state *states.State, opts *terraform.PlanOpts, command configs.TestCommand, globals map[string]backend.UnparsedVariableValue) (*terraform.Context, *plans.Plan, *states.State, tfdiags.Diagnostics) {
if opts.Mode == plans.DestroyMode && state.Empty() {
// Nothing to do!
return nil, nil, state, nil
}
// First, transform the config for the given test run and test file.
var diags tfdiags.Diagnostics
if run == nil {
reset, cfgDiags := config.TransformForTest(nil, file.Config)
defer reset()
diags = diags.Append(cfgDiags)
} else {
reset, cfgDiags := config.TransformForTest(run.Config, file.Config)
defer reset()
diags = diags.Append(cfgDiags)
}
if diags.HasErrors() {
return nil, nil, state, diags
}
// Second, gather any variables and give them to the plan options.
variables, variableDiags := buildInputVariablesForTest(run, file, config, globals)
diags = diags.Append(variableDiags)
if variableDiags.HasErrors() {
return nil, nil, state, diags
}
opts.SetVariables = variables
// Third, execute planning stage.
tfCtxOpts, err := runner.command.contextOpts()
diags = diags.Append(err)
if err != nil {
return nil, nil, state, diags
}
tfCtx, ctxDiags := terraform.NewContext(tfCtxOpts)
diags = diags.Append(ctxDiags)
if ctxDiags.HasErrors() {
return nil, nil, state, diags
}
runningCtx, done := context.WithCancel(context.Background())
var plan *plans.Plan
var planDiags tfdiags.Diagnostics
go func() {
defer logging.PanicHandler()
defer done()
plan, planDiags = tfCtx.Plan(config, state, opts)
}()
waitDiags, cancelled := runner.wait(tfCtx, runningCtx, mgr, run, file, nil)
planDiags = planDiags.Append(waitDiags)
diags = diags.Append(planDiags)
if planDiags.HasErrors() || command == configs.PlanTestCommand {
// Either the plan errored, or we only wanted to see the plan. Either
// way, just return what we have: The plan and diagnostics from making
// it and the unchanged state.
return tfCtx, plan, state, diags
}
if cancelled {
// If the execution was cancelled during the plan, we'll exit here to
// stop the plan being applied and using more time.
return tfCtx, plan, state, diags
}
// We're also going to strip out any warnings from check blocks, as we do
// for normal executions. Since we're going to go ahead and execute the
// plan immediately, any warnings from the check block are just not relevant
// any more.
var filteredDiags tfdiags.Diagnostics
for _, diag := range diags {
if rule, ok := addrs.DiagnosticOriginatesFromCheckRule(diag); ok && rule.Container.CheckableKind() == addrs.CheckableCheck {
continue
}
filteredDiags = filteredDiags.Append(diag)
}
diags = filteredDiags
// Fourth, execute apply stage.
tfCtx, ctxDiags = terraform.NewContext(tfCtxOpts)
diags = diags.Append(ctxDiags)
if ctxDiags.HasErrors() {
return nil, nil, state, diags
}
runningCtx, done = context.WithCancel(context.Background())
// If things get cancelled while we are executing the apply operation below
// we want to print out all the objects that we were creating so the user
// can verify we managed to tidy everything up possibly.
//
// Unfortunately, this creates a race condition as the apply operation can
// edit the plan (by removing changes once they are applied) while at the
// same time our cancellation process will try to read the plan.
//
// We take a quick copy of the changes we care about here, which will then
// be used in place of the plan when we print out the objects to be created
// as part of the cancellation process.
var created []*plans.ResourceInstanceChangeSrc
for _, change := range plan.Changes.Resources {
if change.Action != plans.Create {
continue
}
created = append(created, change)
}
var updated *states.State
var applyDiags tfdiags.Diagnostics
go func() {
defer logging.PanicHandler()
defer done()
updated, applyDiags = tfCtx.Apply(plan, config)
}()
waitDiags, _ = runner.wait(tfCtx, runningCtx, mgr, run, file, created)
applyDiags = applyDiags.Append(waitDiags)
diags = diags.Append(applyDiags)
return tfCtx, plan, updated, diags
}
func (runner *TestRunner) wait(ctx *terraform.Context, runningCtx context.Context, mgr *TestStateManager, run *moduletest.Run, file *moduletest.File, created []*plans.ResourceInstanceChangeSrc) (diags tfdiags.Diagnostics, cancelled bool) {
// This function handles what happens when the user presses the second
// interrupt. This is a "hard cancel", we are going to stop doing whatever
// it is we're doing. This means even if we're halfway through creating or
// destroying infrastructure we just give up.
handleCancelled := func() {
states := make(map[*moduletest.Run]*states.State)
states[nil] = mgr.State
for _, module := range mgr.States {
states[module.Run] = module.State
}
runner.View.FatalInterruptSummary(run, file, states, created)
cancelled = true
go ctx.Stop()
// Just wait for things to finish now, the overall test execution will
// exit early if this takes too long.
<-runningCtx.Done()
}
// This function handles what happens when the user presses the first
// interrupt. This is essentially a "soft cancel", we're not going to do
// anything but just wait for things to finish safely. But, we do listen
// for the crucial second interrupt which will prompt a hard stop / cancel.
handleStopped := func() {
select {
case <-runner.CancelledCtx.Done():
// We've been asked again. This time we stop whatever we're doing
// and abandon all attempts to do anything reasonable.
handleCancelled()
case <-runningCtx.Done():
// Do nothing, we finished safely and skipping the remaining tests
// will be handled elsewhere.
}
}
select {
case <-runner.StoppedCtx.Done():
handleStopped()
case <-runner.CancelledCtx.Done():
handleCancelled()
case <-runningCtx.Done():
// The operation exited normally.
}
return diags, cancelled
}
// state management
// TestStateManager is a helper struct to maintain the various state objects
// that a test file has to keep track of.
type TestStateManager struct {
runner *TestRunner
// State is the main state of the module under test during a single test
// file execution. This state will be updated by every run block without
// a modifier module block within the test file. At the end of the test
// file's execution everything in this state should be executed.
State *states.State
// States contains the states of every run block within a test file that
// executed using an alternative module. Any resources created by these
// run blocks also need to be tidied up, but only after the main state file
// has been handled.
States []*TestModuleState
}
// TestModuleState holds the config and the state for a given run block that
// executed with a custom module.
type TestModuleState struct {
// State is the state after the module executed.
State *states.State
// Run is the config for the given run block, that contains the config
// under test and the variable values.
Run *moduletest.Run
}
func (manager *TestStateManager) cleanupStates(file *moduletest.File, globals map[string]backend.UnparsedVariableValue) {
if manager.runner.Cancelled {
// Don't try and clean anything up if the execution has been cancelled.
return
}
// First, we'll clean up the main state.
_, _, state, diags := manager.runner.execute(manager, nil, file, manager.runner.Config, manager.State, &terraform.PlanOpts{
Mode: plans.DestroyMode,
}, configs.ApplyTestCommand, globals)
manager.runner.View.DestroySummary(diags, nil, file, state)
if manager.runner.Cancelled {
// In case things were cancelled during the last execution.
return
}
// Then we'll clean up the additional states for custom modules in reverse
// order.
for ix := len(manager.States); ix > 0; ix-- {
module := manager.States[ix-1]
if manager.runner.Cancelled {
// In case the cancellation came while a previous state was being
// destroyed.
return
}
_, _, state, diags := manager.runner.execute(manager, module.Run, file, module.Run.Config.ConfigUnderTest, module.State, &terraform.PlanOpts{
Mode: plans.DestroyMode,
}, configs.ApplyTestCommand, globals)
manager.runner.View.DestroySummary(diags, module.Run, file, state)
}
}
// helper functions
// buildInputVariablesForTest creates a terraform.InputValues mapping for
// variable values that are relevant to the config being tested.
//
// Crucially, it differs from buildInputVariablesForAssertions in that it only
// includes variables that are reference by the config and not everything that
// is defined within the test run block and test file.
func buildInputVariablesForTest(run *moduletest.Run, file *moduletest.File, config *configs.Config, globals map[string]backend.UnparsedVariableValue) (terraform.InputValues, tfdiags.Diagnostics) {
variables := make(map[string]backend.UnparsedVariableValue)
for name := range config.Module.Variables {
if run != nil {
if expr, exists := run.Config.Variables[name]; exists {
// Local variables take precedence.
variables[name] = unparsedVariableValueExpression{
expr: expr,
sourceType: terraform.ValueFromConfig,
}
continue
}
}
if file != nil {
if expr, exists := file.Config.Variables[name]; exists {
// If it's not set locally, it maybe set for the entire file.
variables[name] = unparsedVariableValueExpression{
expr: expr,
sourceType: terraform.ValueFromConfig,
}
continue
}
}
if globals != nil {
// If it's not set locally or at the file level, maybe it was
// defined globally.
if variable, exists := globals[name]; exists {
variables[name] = variable
}
}
// If it's not set at all that might be okay if the variable is optional
// so we'll just not add anything to the map.
}
return backend.ParseVariableValues(variables, config.Module.Variables)
}
// buildInputVariablesForAssertions creates a terraform.InputValues mapping that
// contains all the variables defined for a given run and file, alongside any
// unset variables that have defaults within the provided config.
//
// Crucially, it differs from buildInputVariablesForTest in that the returned
// input values include all variables available even if they are not defined
// within the config.
//
// This does mean the returned diags might contain warnings about variables not
// defined within the config. We might want to remove these warnings in the
// future, since it is actually okay for test files to have variables defined
// outside the configuration.
func buildInputVariablesForAssertions(run *moduletest.Run, file *moduletest.File, config *configs.Config, globals map[string]backend.UnparsedVariableValue) (terraform.InputValues, tfdiags.Diagnostics) {
variables := make(map[string]backend.UnparsedVariableValue)
if run != nil {
for name, expr := range run.Config.Variables {
variables[name] = unparsedVariableValueExpression{
expr: expr,
sourceType: terraform.ValueFromConfig,
}
}
}
if file != nil {
for name, expr := range file.Config.Variables {
if _, exists := variables[name]; exists {
// Then this variable was defined at the run level and we want
// that value to take precedence.
continue
}
variables[name] = unparsedVariableValueExpression{
expr: expr,
sourceType: terraform.ValueFromConfig,
}
}
}
for name, variable := range globals {
if _, exists := variables[name]; exists {
// Then this value was already defined at either the run level
// or the file level, and we want those values to take
// precedence.
continue
}
variables[name] = variable
}
return backend.ParseVariableValues(variables, config.Module.Variables)
}