pulumi/pkg/testing/integration/program.go

882 lines
29 KiB
Go
Raw Normal View History

// Copyright 2016-2018, Pulumi Corporation. All rights reserved.
package integration
import (
"context"
"encoding/json"
"fmt"
"io"
"io/ioutil"
"os"
"os/exec"
"path"
"path/filepath"
"strconv"
"strings"
"testing"
"time"
"github.com/pkg/errors"
"github.com/stretchr/testify/assert"
"github.com/pulumi/pulumi/pkg/apitype"
"github.com/pulumi/pulumi/pkg/backend/local"
"github.com/pulumi/pulumi/pkg/engine"
"github.com/pulumi/pulumi/pkg/resource"
"github.com/pulumi/pulumi/pkg/tokens"
"github.com/pulumi/pulumi/pkg/util/contract"
"github.com/pulumi/pulumi/pkg/util/fsutil"
"github.com/pulumi/pulumi/pkg/util/retry"
"github.com/pulumi/pulumi/pkg/workspace"
)
// RuntimeValidationStackInfo contains details related to the stack that runtime validation logic may want to use.
type RuntimeValidationStackInfo struct {
Deployment *apitype.Deployment
RootResource apitype.Resource
Outputs map[string]interface{}
}
// EditDir is an optional edit to apply to the example, as subsequent deployments.
type EditDir struct {
Dir string
ExtraRuntimeValidation func(t *testing.T, stack RuntimeValidationStackInfo)
// Additive is true if Dir should be copied *on top* of the test directory.
// Otherwise Dir *replaces* the test directory, except we keep .pulumi/ and Pulumi.yaml and Pulumi.<stack>.yaml.
Additive bool
// ExpectFailure is true if we expect this test to fail. This is very coarse grained, and will essentially
// tolerate *any* failure in the program (IDEA: in the future, offer a way to narrow this down more).
ExpectFailure bool
// Stdout is the writer to use for all stdout messages.
Stdout io.Writer
// Stderr is the writer to use for all stderr messages.
Stderr io.Writer
// Verbose may be set to true to print messages as they occur, rather than buffering and showing upon failure.
Verbose bool
}
// TestCommandStats is a collection of data related to running a single command during a test.
type TestCommandStats struct {
// StartTime is the time at which the command was started
StartTime string `json:"startTime"`
// EndTime is the time at which the command exited
EndTime string `json:"endTime"`
// ElapsedSeconds is the time at which the command exited
ElapsedSeconds float64 `json:"elapsedSeconds"`
// StackName is the name of the stack
StackName string `json:"stackName"`
// TestId is the unique ID of the test run
TestID string `json:"testId"`
// StepName is the command line which was invoked1
StepName string `json:"stepName"`
// CommandLine is the command line which was invoked1
CommandLine string `json:"commandLine"`
// TestName is the name of the directory in which the test was executed
TestName string `json:"testName"`
// IsError is true if the command failed
IsError bool `json:"isError"`
// The Cloud that the test was run against, or empty for local deployments
CloudURL string `json:"cloudURL"`
// The PPC that the test was run against, or empty for local deployments or for the default PPC
CloudPPC string `json:"cloudPPC"`
}
// TestStatsReporter reports results and metadata from a test run.
type TestStatsReporter interface {
ReportCommand(stats TestCommandStats)
}
// ProgramTestOptions provides options for ProgramTest
type ProgramTestOptions struct {
// Dir is the program directory to test.
Dir string
// Array of NPM packages which must be `yarn linked` (e.g. {"pulumi", "@pulumi/aws"})
Dependencies []string
// Map of config keys and values to set (e.g. {"aws:config:region": "us-east-2"})
Config map[string]string
// Map of secure config keys and values to set on the stack (e.g. {"aws:config:region": "us-east-2"})
Secrets map[string]string
// EditDirs is an optional list of edits to apply to the example, as subsequent deployments.
EditDirs []EditDir
// ExtraRuntimeValidation is an optional callback for additional validation, called before applying edits.
ExtraRuntimeValidation func(t *testing.T, stack RuntimeValidationStackInfo)
// RelativeWorkDir is an optional path relative to `Dir` which should be used as working directory during tests.
RelativeWorkDir string
// ExpectFailure is true if we expect this test to fail. This is very coarse grained, and will essentially
// tolerate *any* failure in the program (IDEA: in the future, offer a way to narrow this down more).
ExpectFailure bool
// Quick can be set to true to run a "quick" test that skips any non-essential steps (e.g., empty updates).
Quick bool
// UpdateCommandlineFlags specifies flags to add to the `pulumi update` command line (e.g. "--color=raw")
UpdateCommandlineFlags []string
// CloudURL is an optional URL to override the default Pulumi Service API (https://api.pulumi-staging.io). The
// PULUMI_ACCESS_TOKEN environment variable must also be set to a valid access token for the target cloud.
CloudURL string
// Owner and Repo are optional values to specify during calls to `pulumi init`. Otherwise the --owner and
// --repo flags will not be set.
Owner string
Repo string
// PPCName is the name of the PPC to use when running a test against the hosted service. If
// not set, the --ppc flag will not be set on `pulumi stack init`.
PPCName string
// StackName allows the stack name to be explicitly provided instead of computed from the
// environment during tests.
StackName string
// ReportStats optionally specifies how to report results from the test for external collection.
ReportStats TestStatsReporter
// Stdout is the writer to use for all stdout messages.
Stdout io.Writer
// Stderr is the writer to use for all stderr messages.
Stderr io.Writer
// Verbose may be set to true to print messages as they occur, rather than buffering and showing upon failure.
Verbose bool
// DebugLogging may be set to anything >0 to enable excessively verbose debug logging from `pulumi`. This is
// equivalent to `--logtostderr -v=N`, where N is the value of DebugLogLevel. This may also be enabled by setting
// the environment variable PULUMI_TEST_DEBUG_LOG_LEVEL.
DebugLogLevel int
// DebugUpdates may be set to true to enable debug logging from `pulumi preview`, `pulumi update`, and
// `pulumi destroy`. This may also be enabled by setting the environment variable PULUMI_TEST_DEBUG_UPDATES.
DebugUpdates bool
// Bin is a location of a `pulumi` executable to be run. Taken from the $PATH if missing.
Bin string
// YarnBin is a location of a `yarn` executable to be run. Taken from the $PATH if missing.
YarnBin string
// Additional environment variaibles to pass for each command we run.
Env []string
}
func (opts *ProgramTestOptions) GetDebugLogLevel() int {
if opts.DebugLogLevel > 0 {
return opts.DebugLogLevel
}
if du := os.Getenv("PULUMI_TEST_DEBUG_LOG_LEVEL"); du != "" {
2018-03-30 00:15:52 +02:00
if n, e := strconv.Atoi(du); e != nil {
panic(e)
} else if n > 0 {
return n
}
}
return 0
}
func (opts *ProgramTestOptions) GetDebugUpdates() bool {
return opts.DebugUpdates || os.Getenv("PULUMI_TEST_DEBUG_UPDATES") != ""
}
// GetStackName returns a stack name to use for this test.
func (opts *ProgramTestOptions) GetStackName() tokens.QName {
if opts.StackName == "" {
// Fetch the host and test dir names, cleaned so to contain just [a-zA-Z0-9-_] chars.
hostname, err := os.Hostname()
contract.AssertNoErrorf(err, "failure to fetch hostname for stack prefix")
var host string
for _, c := range hostname {
if len(host) >= 10 {
break
}
if (c >= 'a' && c <= 'z') || (c >= 'A' && c <= 'Z') ||
(c >= '0' && c <= '9') || c == '-' || c == '_' {
host += string(c)
}
}
var test string
for _, c := range filepath.Base(opts.Dir) {
if len(test) >= 10 {
break
}
if (c >= 'a' && c <= 'z') || (c >= 'A' && c <= 'Z') ||
(c >= '0' && c <= '9') || c == '-' || c == '_' {
test += string(c)
}
}
lowerStackName := strings.ToLower("p-it-" + host + "-" + test)
legalStackName := strings.TrimRight(lowerStackName, "-_")
opts.StackName = legalStackName
}
return tokens.QName(opts.StackName)
}
// With combines a source set of options with a set of overrides.
func (opts ProgramTestOptions) With(overrides ProgramTestOptions) ProgramTestOptions {
if overrides.Dir != "" {
opts.Dir = overrides.Dir
}
if overrides.Dependencies != nil {
opts.Dependencies = overrides.Dependencies
}
for k, v := range overrides.Config {
if opts.Config == nil {
opts.Config = make(map[string]string)
}
opts.Config[k] = v
}
for k, v := range overrides.Secrets {
if opts.Secrets == nil {
opts.Secrets = make(map[string]string)
}
opts.Secrets[k] = v
}
if overrides.CloudURL != "" {
opts.CloudURL = overrides.CloudURL
}
if overrides.Owner != "" {
opts.Owner = overrides.Owner
}
if overrides.Repo != "" {
opts.Repo = overrides.Repo
}
if overrides.PPCName != "" {
opts.PPCName = overrides.PPCName
}
if overrides.EditDirs != nil {
opts.EditDirs = overrides.EditDirs
}
if overrides.ExtraRuntimeValidation != nil {
opts.ExtraRuntimeValidation = overrides.ExtraRuntimeValidation
}
if overrides.RelativeWorkDir != "" {
opts.RelativeWorkDir = overrides.RelativeWorkDir
}
if overrides.ReportStats != nil {
opts.ReportStats = overrides.ReportStats
}
return opts
}
// ProgramTest runs a lifecycle of Pulumi commands in a program working directory, using the `pulumi` and `yarn`
// binaries available on PATH. It essentially executes the following workflow:
//
// yarn install
// yarn link <each opts.Depencies>
// yarn run build
// pulumi init
// (*) pulumi login
// pulumi stack init integrationtesting
// pulumi config set <each opts.Config>
// pulumi config set --secret <each opts.Secrets>
// pulumi preview
// pulumi update
// pulumi preview (expected to be empty)
// pulumi update (expected to be empty)
// pulumi destroy --yes
// pulumi stack rm --yes integrationtesting
//
// (*) Only if PULUMI_ACCESS_TOKEN is set.
//
// All commands must return success return codes for the test to succeed, unless ExpectFailure is true.
func ProgramTest(t *testing.T, opts *ProgramTestOptions) {
// Disable stack backups for tests to avoid filling up ~/.pulumi/backups with unnecessary
// backups of test stacks.
if err := os.Setenv(local.DisableCheckpointBackupsEnvVar, "1"); err != nil {
t.Errorf("error setting env var '%s': %v", local.DisableCheckpointBackupsEnvVar, err)
}
t.Parallel()
// If the test panics, recover and log instead of letting the panic escape the test. Even though *this* test will
// have run deferred functions and cleaned up, if the panic reaches toplevel it will kill the process and prevent
// other tests running in parallel from cleaning up.
defer func() {
if failure := recover(); failure != nil {
t.Errorf("panic testing %v: %v", opts.Dir, failure)
}
}()
pt := newProgramTester(t, opts)
err := pt.testLifeCycleInitAndDestroy()
assert.NoError(t, err)
}
// fprintf works like fmt.FPrintf, except it explicitly drops the return values. This keeps the linters happy, since
// they don't like to see errors dropped on the floor. It is possible that our call to fmt.Fprintf will fail, even
// for "standard" streams like `stdout` and `stderr`, if they have been set to non-blocking by an external process.
// In that case, we just drop the error on the floor and continue. We see this behavior in Travis when we try to write
// a lot of messages quickly (as we do when logging test failures)
func fprintf(w io.Writer, format string, a ...interface{}) {
_, err := fmt.Fprintf(w, format, a...)
contract.IgnoreError(err)
}
// programTester contains state associated with running a single test pass.
type programTester struct {
t *testing.T // the Go tester for this run.
opts *ProgramTestOptions // options that control this test run.
bin string // the `pulumi` binary we are using.
yarnBin string // the `yarn` binary we are using.
2018-01-06 02:40:41 +01:00
}
func newProgramTester(t *testing.T, opts *ProgramTestOptions) *programTester {
return &programTester{t: t, opts: opts}
}
func (pt *programTester) getBin() (string, error) {
return getCmdBin(&pt.bin, "pulumi", pt.opts.Bin)
}
func (pt *programTester) getYarnBin() (string, error) {
return getCmdBin(&pt.yarnBin, "yarn", pt.opts.YarnBin)
}
func (pt *programTester) pulumiCmd(args []string) ([]string, error) {
bin, err := pt.getBin()
if err != nil {
return nil, err
}
cmd := []string{bin}
if du := pt.opts.GetDebugLogLevel(); du > 0 {
cmd = append(cmd, "--logtostderr")
cmd = append(cmd, "-v="+strconv.Itoa(du))
}
return append(cmd, args...), nil
}
func (pt *programTester) yarnCmd(args []string) ([]string, error) {
bin, err := pt.getYarnBin()
if err != nil {
return nil, err
}
result := []string{bin}
result = append(result, args...)
return withOptionalYarnFlags(result), nil
}
func (pt *programTester) runCommand(name string, args []string, wd string) error {
return RunCommand(pt.t, name, args, wd, pt.opts)
}
func (pt *programTester) runPulumiCommand(name string, args []string, wd string) error {
cmd, err := pt.pulumiCmd(args)
if err != nil {
return err
}
return pt.runCommand(name, cmd, wd)
}
func (pt *programTester) runYarnCommand(name string, args []string, wd string) error {
cmd, err := pt.yarnCmd(args)
if err != nil {
return err
}
_, _, err = retry.Until(context.Background(), retry.Acceptor{
Accept: func(try int, nextRetryTime time.Duration) (bool, interface{}, error) {
runerr := pt.runCommand(name, cmd, wd)
if runerr == nil {
return true, nil, nil
} else if _, ok := runerr.(*exec.ExitError); ok {
// yarn failed, let's try again, assuming we haven't failed a few times.
if try > 3 {
return false, nil, errors.Errorf("%v did not complete after %v tries", cmd, try)
}
return false, nil, nil
}
// someother error, fail
return false, nil, runerr
},
})
return err
}
func (pt *programTester) testLifeCycleInitAndDestroy() error {
dir, err := pt.copyTestToTemporaryDirectory()
if err != nil {
return errors.Wrap(err, "copying test to temp dir")
}
testFinished := false
defer func() {
if !testFinished || pt.t.Failed() {
// Test aborted or failed. Maybe copy to "failed tests" directory.
failedTestsDir := os.Getenv("PULUMI_FAILED_TESTS_DIR")
if failedTestsDir != "" {
dest := filepath.Join(failedTestsDir, pt.t.Name()+uniqueSuffix())
contract.IgnoreError(fsutil.CopyFile(dest, dir, nil))
}
} else {
contract.IgnoreError(os.RemoveAll(dir))
}
}()
err = pt.testLifeCycleInitialize(dir)
if err != nil {
return errors.Wrap(err, "initializing test project")
}
// Ensure that before we exit, we attempt to destroy and remove the stack.
defer func() {
if dir != "" {
destroyErr := pt.testLifeCycleDestroy(dir)
assert.NoError(pt.t, destroyErr)
}
}()
if err = pt.testPreviewUpdateAndEdits(dir); err != nil {
return errors.Wrap(err, "running test preview, update, and edits")
}
testFinished = true
return nil
}
func (pt *programTester) testLifeCycleInitialize(dir string) error {
stackName := pt.opts.GetStackName()
// If RelativeWorkDir is specified, apply that relative to the temp folder for use as working directory during tests.
if pt.opts.RelativeWorkDir != "" {
dir = path.Join(dir, pt.opts.RelativeWorkDir)
}
// Set the default target Pulumi API if not overridden in options.
if pt.opts.CloudURL == "" {
pulumiAPI := os.Getenv("PULUMI_API")
if pulumiAPI != "" {
pt.opts.CloudURL = pulumiAPI
}
}
// Set the owner organization from an environment variable if not overridden in options.
if pt.opts.Owner == "" {
pulumiAPIOwnerOrganization := os.Getenv("PULUMI_API_OWNER_ORGANIZATION")
if pulumiAPIOwnerOrganization != "" {
pt.opts.Owner = pulumiAPIOwnerOrganization
} else {
// Default to the `pulumi` organization.
pt.opts.Owner = "pulumi"
}
}
// Set the target PPC from an environment variable if not overridden in options.
if pt.opts.PPCName == "" {
ppcName := os.Getenv("PULUMI_API_PPC_NAME")
if ppcName != "" {
pt.opts.PPCName = ppcName
}
}
// Ensure all links are present, the stack is created, and all configs are applied.
fprintf(pt.opts.Stdout, "Initializing project (dir %s; stack %s)\n", dir, stackName)
initArgs := []string{"init"}
initArgs = addFlagIfNonNil(initArgs, "--owner", pt.opts.Owner)
initArgs = addFlagIfNonNil(initArgs, "--name", pt.opts.Repo)
if err := pt.runPulumiCommand("pulumi-init", initArgs, dir); err != nil {
return err
}
// Login as needed.
if os.Getenv("PULUMI_ACCESS_TOKEN") == "" && pt.opts.CloudURL == "" {
fmt.Printf("Using existing logged in user for tests. Set PULUMI_ACCESS_TOKEN and/or PULUMI_API to override.\n")
} else {
// Set PulumiCredentialsPathEnvVar to our CWD, so we use credentials specific to just this
// test.
pt.opts.Env = append(pt.opts.Env, fmt.Sprintf("%s=%s", workspace.PulumiCredentialsPathEnvVar, dir))
loginArgs := []string{"login"}
loginArgs = addFlagIfNonNil(loginArgs, "--cloud-url", pt.opts.CloudURL)
if err := pt.runPulumiCommand("pulumi-login", loginArgs, dir); err != nil {
return err
}
}
// Stack init
stackInitArgs := []string{"stack", "init", string(stackName)}
stackInitArgs = addFlagIfNonNil(stackInitArgs, "--ppc", pt.opts.PPCName)
if err := pt.runPulumiCommand("pulumi-stack-init", stackInitArgs, dir); err != nil {
return err
}
for key, value := range pt.opts.Config {
if err := pt.runPulumiCommand("pulumi-config",
[]string{"config", "set", key, value}, dir); err != nil {
return err
}
}
for key, value := range pt.opts.Secrets {
if err := pt.runPulumiCommand("pulumi-config",
[]string{"config", "set", "--secret", key, value}, dir); err != nil {
return err
}
}
return nil
}
func (pt *programTester) testLifeCycleDestroy(dir string) error {
stackName := pt.opts.GetStackName()
// Destroy and remove the stack.
fprintf(pt.opts.Stdout, "Destroying stack\n")
destroy := []string{"destroy", "--yes"}
if pt.opts.GetDebugUpdates() {
destroy = append(destroy, "-d")
}
if err := pt.runPulumiCommand("pulumi-destroy", destroy, dir); err != nil {
return err
}
err := pt.runPulumiCommand("pulumi-stack-rm", []string{"stack", "rm", "--yes", string(stackName)}, dir)
return err
}
func (pt *programTester) testPreviewUpdateAndEdits(dir string) error {
// Now preview and update the real changes.
fprintf(pt.opts.Stdout, "Performing primary preview and update\n")
initErr := pt.previewAndUpdate(dir, "initial", pt.opts.ExpectFailure)
// If the initial preview/update failed, just exit without trying the rest (but make sure to destroy).
if initErr != nil {
return initErr
}
// Perform an empty preview and update; nothing is expected to happen here.
if !pt.opts.Quick {
fprintf(pt.opts.Stdout, "Performing empty preview and update (no changes expected)\n")
if err := pt.previewAndUpdate(dir, "empty", false); err != nil {
return err
}
}
// Run additional validation provided by the test options, passing in the checkpoint info.
if err := pt.performExtraRuntimeValidation(pt.opts.ExtraRuntimeValidation, dir); err != nil {
return err
}
// If there are any edits, apply them and run a preview and update for each one.
return pt.testEdits(dir)
}
func (pt *programTester) previewAndUpdate(dir string, name string, shouldFail bool) error {
preview := []string{"preview"}
update := []string{"update"}
if pt.opts.GetDebugUpdates() {
preview = append(preview, "-d")
update = append(update, "-d")
}
if pt.opts.UpdateCommandlineFlags != nil {
update = append(update, pt.opts.UpdateCommandlineFlags...)
}
if !pt.opts.Quick {
if err := pt.runPulumiCommand("pulumi-preview-"+name, preview, dir); err != nil {
if shouldFail {
fprintf(pt.opts.Stdout, "Permitting failure (ExpectFailure=true for this preview)\n")
return nil
}
return err
}
}
if err := pt.runPulumiCommand("pulumi-update-"+name, update, dir); err != nil {
if shouldFail {
fprintf(pt.opts.Stdout, "Permitting failure (ExpectFailure=true for this update)\n")
return nil
}
return err
}
// If we expected a failure, but none occurred, return an error.
if shouldFail {
return errors.New("expected this step to fail, but it succeeded")
}
return nil
}
func (pt *programTester) testEdits(dir string) error {
for i, edit := range pt.opts.EditDirs {
var err error
if err = pt.testEdit(dir, i, edit); err != nil {
return err
}
}
return nil
}
func (pt *programTester) testEdit(dir string, i int, edit EditDir) error {
fprintf(pt.opts.Stdout, "Applying edit '%v' and rerunning preview and update\n", edit.Dir)
if edit.Additive {
// Just copy new files into dir
if err := fsutil.CopyFile(dir, edit.Dir, nil); err != nil {
return errors.Wrapf(err, "Couldn't copy %v into %v", edit.Dir, dir)
}
} else {
// Create a new temporary directory
newDir, err := ioutil.TempDir("", pt.opts.StackName+"-")
if err != nil {
return errors.Wrapf(err, "Couldn't create new temporary directory")
}
// Delete whichever copy of the test is unused when we return
dirToDelete := newDir
defer func() {
contract.IgnoreError(os.RemoveAll(dirToDelete))
}()
// Copy everything except Pulumi.yaml, Pulumi.<stack-name>.yaml, and .pulumi from source into new directory
exclusions := make(map[string]bool)
projectYaml := workspace.ProjectFile + ".yaml"
configYaml := workspace.ProjectFile + "." + pt.opts.StackName + ".yaml"
exclusions[workspace.BookkeepingDir] = true
exclusions[projectYaml] = true
exclusions[configYaml] = true
if err := fsutil.CopyFile(newDir, edit.Dir, exclusions); err != nil {
return errors.Wrapf(err, "Couldn't copy %v into %v", edit.Dir, newDir)
}
// Copy Pulumi.yaml, Pulumi.<stack-name>.yaml, and .pulumi from old directory to new directory
oldProjectYaml := filepath.Join(dir, projectYaml)
newProjectYaml := filepath.Join(newDir, projectYaml)
oldConfigYaml := filepath.Join(dir, configYaml)
newConfigYaml := filepath.Join(newDir, configYaml)
oldProjectDir := filepath.Join(dir, workspace.BookkeepingDir)
newProjectDir := filepath.Join(newDir, workspace.BookkeepingDir)
if err := fsutil.CopyFile(newProjectYaml, oldProjectYaml, nil); err != nil {
return errors.Wrap(err, "Couldn't copy Pulumi.yaml")
}
if err := fsutil.CopyFile(newConfigYaml, oldConfigYaml, nil); err != nil {
return errors.Wrapf(err, "Couldn't copy Pulumi.%s.yaml", pt.opts.StackName)
}
if err := fsutil.CopyFile(newProjectDir, oldProjectDir, nil); err != nil {
return errors.Wrap(err, "Couldn't copy .pulumi")
}
// Finally, replace our current temp directory with the new one.
dirOld := dir + ".old"
if err := os.Rename(dir, dirOld); err != nil {
return errors.Wrapf(err, "Couldn't rename %v to %v", dir, dirOld)
}
// There's a brief window here where the old temp dir name could be taken from us.
if err := os.Rename(newDir, dir); err != nil {
return errors.Wrapf(err, "Couldn't rename %v to %v", newDir, dir)
}
// Keep dir, delete oldDir
dirToDelete = dirOld
}
err := pt.prepareProject(dir)
if err != nil {
return errors.Wrapf(err, "Couldn't prepare project in %v", dir)
}
oldStdOut := pt.opts.Stdout
oldStderr := pt.opts.Stderr
oldVerbose := pt.opts.Verbose
if edit.Stdout != nil {
pt.opts.Stdout = edit.Stdout
}
if edit.Stderr != nil {
pt.opts.Stderr = edit.Stderr
}
if edit.Verbose {
pt.opts.Verbose = true
}
defer func() {
pt.opts.Stdout = oldStdOut
pt.opts.Stderr = oldStderr
pt.opts.Verbose = oldVerbose
}()
if err = pt.previewAndUpdate(dir, fmt.Sprintf("edit-%d", i), edit.ExpectFailure); err != nil {
return err
}
return pt.performExtraRuntimeValidation(edit.ExtraRuntimeValidation, dir)
}
func (pt *programTester) performExtraRuntimeValidation(
extraRuntimeValidation func(t *testing.T, stack RuntimeValidationStackInfo), dir string) error {
if extraRuntimeValidation == nil {
return nil
}
stackName := pt.opts.GetStackName()
// Create a temporary file name for the stack export
tempDir, err := ioutil.TempDir("", string(stackName))
if err != nil {
return err
}
fileName := path.Join(tempDir, "stack.json")
// Invoke `pulumi stack export`
if err = pt.runPulumiCommand("pulumi-export", []string{"stack", "export", "--file", fileName}, dir); err != nil {
return errors.Wrapf(err, "expected to export stack to file: %s", fileName)
}
// Open the exported JSON file
f, err := os.Open(fileName)
if err != nil {
return errors.Wrapf(err, "expected to be able to open file with stack exports: %s", fileName)
}
defer func() {
contract.IgnoreClose(f)
contract.IgnoreError(os.RemoveAll(tempDir))
}()
// Unmarshal the Deployment
var untypedDeployment apitype.UntypedDeployment
if err = json.NewDecoder(f).Decode(&untypedDeployment); err != nil {
return err
}
var deployment apitype.Deployment
if err = json.Unmarshal(untypedDeployment.Deployment, &deployment); err != nil {
return err
}
// Get the root resource and outputs from the deployment
var rootResource apitype.Resource
var outputs map[string]interface{}
for _, res := range deployment.Resources {
if res.Type == resource.RootStackType {
rootResource = res
outputs = res.Outputs
}
}
// Populate stack info object with all of this data to pass to the validation function
stackInfo := RuntimeValidationStackInfo{
Deployment: &deployment,
RootResource: rootResource,
Outputs: outputs,
}
extraRuntimeValidation(pt.t, stackInfo)
return nil
}
// copyTestToTemporaryDirectory creates a temporary directory to run the test in and copies the test to it.
func (pt *programTester) copyTestToTemporaryDirectory() (dir string, err error) {
// Set up a prefix so that all output has the test directory name in it. This is important for debugging
// because we run tests in parallel, and so all output will be interleaved and difficult to follow otherwise.
sourceDir := pt.opts.Dir
var prefix string
if len(sourceDir) <= 30 {
prefix = fmt.Sprintf("[ %30.30s ] ", sourceDir)
} else {
prefix = fmt.Sprintf("[ %30.30s ] ", sourceDir[len(sourceDir)-30:])
}
stdout := pt.opts.Stdout
if stdout == nil {
stdout = newPrefixer(os.Stdout, prefix)
pt.opts.Stdout = stdout
}
stderr := pt.opts.Stderr
if stderr == nil {
stderr = newPrefixer(os.Stderr, prefix)
pt.opts.Stderr = stderr
}
fprintf(pt.opts.Stdout, "sample: %v\n", sourceDir)
bin, err := pt.getBin()
if err != nil {
return "", err
}
fprintf(pt.opts.Stdout, "pulumi: %v\n", bin)
stackName := string(pt.opts.GetStackName())
targetDir, err := ioutil.TempDir("", stackName+"-")
if err != nil {
return "", errors.Wrap(err, "Couldn't create temporary directory")
}
// Copy the source project
if err = fsutil.CopyFile(targetDir, sourceDir, nil); err != nil {
return "", err
}
err = pt.prepareProject(targetDir)
if err != nil {
return "", errors.Wrapf(err, "Failed to prepare %v", targetDir)
}
fprintf(stdout, "projdir: %v\n", targetDir)
return targetDir, nil
}
// prepareProject runs setup necessary to get the project ready for `pulumi` commands.
func (pt *programTester) prepareProject(projectDir string) error {
Get the empty Python program working This change gets enough of the Python SDK up and running that the empty Python program will work. Mostly just scaffolding, but the basic structure is now in place. The primary remaining work is to wire up resource creation to the gRPC interfaces. In summary: * The basic structure is as follows: - Everything goes into sdk/python/. - sdk/python/cmd/pulumi-langhost-python is a Go language host that simply knows how to spawn Python processes to run out entrypoint in response to requests by the engine. - sdk/python/cmd/pulumi-langhost-python-exec is a little Python shim that is invoked by the language host to run Python programs, and is responsible for setting up the minimal goo before we can do so (RPC connections and the like). - sdk/python/lib/ contains a Python Pip package suitable for PyPi. - In there, we have two packages: the root pulumi package that contains all of the basic Pulumi programming model abstractions, and pulumi.runtime, which contains the implementation of resource registration, RPC interfacing with the engine, and so on. * Add logic in our test framework to conditionalize on the language type and react accordingly. This will allow us to skip Yarn for Python projects and eventually run Pip if there's a requirements.txt. * Created the basic project structure, including all of the usual Make targets for installing into the proper places. * Building also runs Pylint and we are clean. There are a few other minor things in here: * Add an "empty" test for both Node.js and Python. These pass. * Fix an existing bug in plugin shutdown logic. At some point, we started waiting for stderr/stdout to flush before shutting down the plugin; but if certain failures happen "early" during the plugin launch process, these channels will never get initialized and so waiting for them deadlocks. * Recently we seem to have added logic to delete test temp directories if a failure happened during initialization of said temp directories. This is unfortunate, because you often need to look at the temp directory to see what failed. We already clean them up elsewhere after the full test completes successfully, so I don't think we need to be doing this, and I've removed it. Still many loose ends (config, resources, etc), but it's a start!
2018-01-13 19:29:34 +01:00
// Load up the package so we know things like what language the project is.
projfile := filepath.Join(projectDir, workspace.ProjectFile+".yaml")
proj, err := workspace.LoadProject(projfile)
if err != nil {
return err
}
projinfo := &engine.Projinfo{Proj: proj, Root: projectDir}
// Based on the language, invoke the right routine to prepare the target directory.
switch proj.Runtime {
case "nodejs":
return pt.prepareNodeJSProject(projinfo)
case "python":
return pt.preparePythonProject(projinfo)
default:
return errors.Errorf("unrecognized project runtime: %s", proj.Runtime)
}
}
// prepareNodeJSProject runs setup necessary to get a Node.js project ready for `pulumi` commands.
func (pt *programTester) prepareNodeJSProject(projinfo *engine.Projinfo) error {
// Write a .yarnrc file to pass --mutex network to all yarn invocations, since tests
// may run concurrently and yarn may fail if invoked concurrently
// https://github.com/yarnpkg/yarn/issues/683
// Also add --network-concurrency 1 since we've been seeing
// https://github.com/yarnpkg/yarn/issues/4563 as well
Get the empty Python program working This change gets enough of the Python SDK up and running that the empty Python program will work. Mostly just scaffolding, but the basic structure is now in place. The primary remaining work is to wire up resource creation to the gRPC interfaces. In summary: * The basic structure is as follows: - Everything goes into sdk/python/. - sdk/python/cmd/pulumi-langhost-python is a Go language host that simply knows how to spawn Python processes to run out entrypoint in response to requests by the engine. - sdk/python/cmd/pulumi-langhost-python-exec is a little Python shim that is invoked by the language host to run Python programs, and is responsible for setting up the minimal goo before we can do so (RPC connections and the like). - sdk/python/lib/ contains a Python Pip package suitable for PyPi. - In there, we have two packages: the root pulumi package that contains all of the basic Pulumi programming model abstractions, and pulumi.runtime, which contains the implementation of resource registration, RPC interfacing with the engine, and so on. * Add logic in our test framework to conditionalize on the language type and react accordingly. This will allow us to skip Yarn for Python projects and eventually run Pip if there's a requirements.txt. * Created the basic project structure, including all of the usual Make targets for installing into the proper places. * Building also runs Pylint and we are clean. There are a few other minor things in here: * Add an "empty" test for both Node.js and Python. These pass. * Fix an existing bug in plugin shutdown logic. At some point, we started waiting for stderr/stdout to flush before shutting down the plugin; but if certain failures happen "early" during the plugin launch process, these channels will never get initialized and so waiting for them deadlocks. * Recently we seem to have added logic to delete test temp directories if a failure happened during initialization of said temp directories. This is unfortunate, because you often need to look at the temp directory to see what failed. We already clean them up elsewhere after the full test completes successfully, so I don't think we need to be doing this, and I've removed it. Still many loose ends (config, resources, etc), but it's a start!
2018-01-13 19:29:34 +01:00
if err := ioutil.WriteFile(
filepath.Join(projinfo.Root, ".yarnrc"),
[]byte("--mutex network\n--network-concurrency 1\n"), 0644); err != nil {
return err
}
Get the empty Python program working This change gets enough of the Python SDK up and running that the empty Python program will work. Mostly just scaffolding, but the basic structure is now in place. The primary remaining work is to wire up resource creation to the gRPC interfaces. In summary: * The basic structure is as follows: - Everything goes into sdk/python/. - sdk/python/cmd/pulumi-langhost-python is a Go language host that simply knows how to spawn Python processes to run out entrypoint in response to requests by the engine. - sdk/python/cmd/pulumi-langhost-python-exec is a little Python shim that is invoked by the language host to run Python programs, and is responsible for setting up the minimal goo before we can do so (RPC connections and the like). - sdk/python/lib/ contains a Python Pip package suitable for PyPi. - In there, we have two packages: the root pulumi package that contains all of the basic Pulumi programming model abstractions, and pulumi.runtime, which contains the implementation of resource registration, RPC interfacing with the engine, and so on. * Add logic in our test framework to conditionalize on the language type and react accordingly. This will allow us to skip Yarn for Python projects and eventually run Pip if there's a requirements.txt. * Created the basic project structure, including all of the usual Make targets for installing into the proper places. * Building also runs Pylint and we are clean. There are a few other minor things in here: * Add an "empty" test for both Node.js and Python. These pass. * Fix an existing bug in plugin shutdown logic. At some point, we started waiting for stderr/stdout to flush before shutting down the plugin; but if certain failures happen "early" during the plugin launch process, these channels will never get initialized and so waiting for them deadlocks. * Recently we seem to have added logic to delete test temp directories if a failure happened during initialization of said temp directories. This is unfortunate, because you often need to look at the temp directory to see what failed. We already clean them up elsewhere after the full test completes successfully, so I don't think we need to be doing this, and I've removed it. Still many loose ends (config, resources, etc), but it's a start!
2018-01-13 19:29:34 +01:00
// Get the correct pwd to run Yarn in.
cwd, _, err := projinfo.GetPwdMain()
if err != nil {
return err
}
// Now ensure dependencies are present.
Get the empty Python program working This change gets enough of the Python SDK up and running that the empty Python program will work. Mostly just scaffolding, but the basic structure is now in place. The primary remaining work is to wire up resource creation to the gRPC interfaces. In summary: * The basic structure is as follows: - Everything goes into sdk/python/. - sdk/python/cmd/pulumi-langhost-python is a Go language host that simply knows how to spawn Python processes to run out entrypoint in response to requests by the engine. - sdk/python/cmd/pulumi-langhost-python-exec is a little Python shim that is invoked by the language host to run Python programs, and is responsible for setting up the minimal goo before we can do so (RPC connections and the like). - sdk/python/lib/ contains a Python Pip package suitable for PyPi. - In there, we have two packages: the root pulumi package that contains all of the basic Pulumi programming model abstractions, and pulumi.runtime, which contains the implementation of resource registration, RPC interfacing with the engine, and so on. * Add logic in our test framework to conditionalize on the language type and react accordingly. This will allow us to skip Yarn for Python projects and eventually run Pip if there's a requirements.txt. * Created the basic project structure, including all of the usual Make targets for installing into the proper places. * Building also runs Pylint and we are clean. There are a few other minor things in here: * Add an "empty" test for both Node.js and Python. These pass. * Fix an existing bug in plugin shutdown logic. At some point, we started waiting for stderr/stdout to flush before shutting down the plugin; but if certain failures happen "early" during the plugin launch process, these channels will never get initialized and so waiting for them deadlocks. * Recently we seem to have added logic to delete test temp directories if a failure happened during initialization of said temp directories. This is unfortunate, because you often need to look at the temp directory to see what failed. We already clean them up elsewhere after the full test completes successfully, so I don't think we need to be doing this, and I've removed it. Still many loose ends (config, resources, etc), but it's a start!
2018-01-13 19:29:34 +01:00
if err = pt.runYarnCommand("yarn-install", []string{"install", "--verbose"}, cwd); err != nil {
return err
}
for _, dependency := range pt.opts.Dependencies {
Get the empty Python program working This change gets enough of the Python SDK up and running that the empty Python program will work. Mostly just scaffolding, but the basic structure is now in place. The primary remaining work is to wire up resource creation to the gRPC interfaces. In summary: * The basic structure is as follows: - Everything goes into sdk/python/. - sdk/python/cmd/pulumi-langhost-python is a Go language host that simply knows how to spawn Python processes to run out entrypoint in response to requests by the engine. - sdk/python/cmd/pulumi-langhost-python-exec is a little Python shim that is invoked by the language host to run Python programs, and is responsible for setting up the minimal goo before we can do so (RPC connections and the like). - sdk/python/lib/ contains a Python Pip package suitable for PyPi. - In there, we have two packages: the root pulumi package that contains all of the basic Pulumi programming model abstractions, and pulumi.runtime, which contains the implementation of resource registration, RPC interfacing with the engine, and so on. * Add logic in our test framework to conditionalize on the language type and react accordingly. This will allow us to skip Yarn for Python projects and eventually run Pip if there's a requirements.txt. * Created the basic project structure, including all of the usual Make targets for installing into the proper places. * Building also runs Pylint and we are clean. There are a few other minor things in here: * Add an "empty" test for both Node.js and Python. These pass. * Fix an existing bug in plugin shutdown logic. At some point, we started waiting for stderr/stdout to flush before shutting down the plugin; but if certain failures happen "early" during the plugin launch process, these channels will never get initialized and so waiting for them deadlocks. * Recently we seem to have added logic to delete test temp directories if a failure happened during initialization of said temp directories. This is unfortunate, because you often need to look at the temp directory to see what failed. We already clean them up elsewhere after the full test completes successfully, so I don't think we need to be doing this, and I've removed it. Still many loose ends (config, resources, etc), but it's a start!
2018-01-13 19:29:34 +01:00
if err = pt.runYarnCommand("yarn-link", []string{"link", dependency}, cwd); err != nil {
return err
}
}
// And finally compile it using whatever build steps are in the package.json file.
return pt.runYarnCommand("yarn-build", []string{"run", "build"}, cwd)
}
Get the empty Python program working This change gets enough of the Python SDK up and running that the empty Python program will work. Mostly just scaffolding, but the basic structure is now in place. The primary remaining work is to wire up resource creation to the gRPC interfaces. In summary: * The basic structure is as follows: - Everything goes into sdk/python/. - sdk/python/cmd/pulumi-langhost-python is a Go language host that simply knows how to spawn Python processes to run out entrypoint in response to requests by the engine. - sdk/python/cmd/pulumi-langhost-python-exec is a little Python shim that is invoked by the language host to run Python programs, and is responsible for setting up the minimal goo before we can do so (RPC connections and the like). - sdk/python/lib/ contains a Python Pip package suitable for PyPi. - In there, we have two packages: the root pulumi package that contains all of the basic Pulumi programming model abstractions, and pulumi.runtime, which contains the implementation of resource registration, RPC interfacing with the engine, and so on. * Add logic in our test framework to conditionalize on the language type and react accordingly. This will allow us to skip Yarn for Python projects and eventually run Pip if there's a requirements.txt. * Created the basic project structure, including all of the usual Make targets for installing into the proper places. * Building also runs Pylint and we are clean. There are a few other minor things in here: * Add an "empty" test for both Node.js and Python. These pass. * Fix an existing bug in plugin shutdown logic. At some point, we started waiting for stderr/stdout to flush before shutting down the plugin; but if certain failures happen "early" during the plugin launch process, these channels will never get initialized and so waiting for them deadlocks. * Recently we seem to have added logic to delete test temp directories if a failure happened during initialization of said temp directories. This is unfortunate, because you often need to look at the temp directory to see what failed. We already clean them up elsewhere after the full test completes successfully, so I don't think we need to be doing this, and I've removed it. Still many loose ends (config, resources, etc), but it's a start!
2018-01-13 19:29:34 +01:00
// preparePythonProject runs setup necessary to get a Python project ready for `pulumi` commands.
func (pt *programTester) preparePythonProject(projinfo *engine.Projinfo) error {
return nil
}