Allow GKE to resume create after interruption

Signed-off-by: Modular Magician <[email protected]>

google/bootstrap_utils_test.go

@@ -1,6 +1,7 @@
 package google
 
 import (
+	"context"
 	"fmt"
 	"log"
 	"os"
@@ -73,7 +74,7 @@ func BootstrapKMSKeyWithPurposeInLocation(t *testing.T, purpose, locationID stri
 
 	ConfigureBasePaths(config)
 
-	if err := config.LoadAndValidate(); err != nil {
+	if err := config.LoadAndValidate(context.Background()); err != nil {
 		t.Errorf("Unable to bootstrap KMS key: %s", err)
 	}
 
@@ -213,7 +214,7 @@ func BootstrapServiceAccount(t *testing.T, project, testRunner string) string {
 
 	ConfigureBasePaths(config)
 
-	if err := config.LoadAndValidate(); err != nil {
+	if err := config.LoadAndValidate(context.Background()); err != nil {
 		t.Fatalf("Bootstrapping failed. Unable to load test config: %s", err)
 	}
 

google/config_test.go

@@ -24,7 +24,7 @@ func TestConfigLoadAndValidate_accountFilePath(t *testing.T) {
 
 	ConfigureBasePaths(config)
 
-	err := config.LoadAndValidate()
+	err := config.LoadAndValidate(context.Background())
 	if err != nil {
 		t.Fatalf("error: %v", err)
 	}
@@ -43,7 +43,7 @@ func TestConfigLoadAndValidate_accountFileJSON(t *testing.T) {
 
 	ConfigureBasePaths(config)
 
-	err = config.LoadAndValidate()
+	err = config.LoadAndValidate(context.Background())
 	if err != nil {
 		t.Fatalf("error: %v", err)
 	}
@@ -58,7 +58,7 @@ func TestConfigLoadAndValidate_accountFileJSONInvalid(t *testing.T) {
 
 	ConfigureBasePaths(config)
 
-	if config.LoadAndValidate() == nil {
+	if config.LoadAndValidate(context.Background()) == nil {
 		t.Fatalf("expected error, but got nil")
 	}
 }
@@ -80,7 +80,7 @@ func TestAccConfigLoadValidate_credentials(t *testing.T) {
 
 	ConfigureBasePaths(config)
 
-	err := config.LoadAndValidate()
+	err := config.LoadAndValidate(context.Background())
 	if err != nil {
 		t.Fatalf("error: %v", err)
 	}
@@ -118,7 +118,7 @@ func TestAccConfigLoadValidate_accessToken(t *testing.T) {
 
 	ConfigureBasePaths(config)
 
-	err = config.LoadAndValidate()
+	err = config.LoadAndValidate(context.Background())
 	if err != nil {
 		t.Fatalf("error: %v", err)
 	}
@@ -139,7 +139,7 @@ func TestConfigLoadAndValidate_customScopes(t *testing.T) {
 
 	ConfigureBasePaths(config)
 
-	err := config.LoadAndValidate()
+	err := config.LoadAndValidate(context.Background())
 	if err != nil {
 		t.Fatalf("unexpected error: %v", err)
 	}
@@ -165,7 +165,7 @@ func TestConfigLoadAndValidate_defaultBatchingConfig(t *testing.T) {
 		BatchingConfig: batchCfg,
 	}
 
-	err = config.LoadAndValidate()
+	err = config.LoadAndValidate(context.Background())
 	if err != nil {
 		t.Fatalf("unexpected error: %v", err)
 	}
@@ -202,7 +202,7 @@ func TestConfigLoadAndValidate_customBatchingConfig(t *testing.T) {
 		BatchingConfig: batchCfg,
 	}
 
-	err = config.LoadAndValidate()
+	err = config.LoadAndValidate(context.Background())
 	if err != nil {
 		t.Fatalf("unexpected error: %v", err)
 	}

google/container_operation.go

@@ -1,13 +1,17 @@
 package google
 
 import (
+	"context"
+	"errors"
 	"fmt"
+	"log"
 
 	container "google.golang.org/api/container/v1beta1"
 )
 
 type ContainerOperationWaiter struct {
 	Service  *container.Service
+	Context  context.Context
 	Op       *container.Operation
 	Project  string
 	Location string
@@ -62,6 +66,13 @@ func (w *ContainerOperationWaiter) QueryOp() (interface{}, error) {
 		w.Project, w.Location, w.Op.Name)
 
 	var op *container.Operation
+	select {
+	case <-w.Context.Done():
+		log.Println("[WARN] request has been cancelled early")
+		return op, errors.New("unable to finish polling, context has been cancelled")
+	default:
+		// default must be here to keep the previous case from blocking
+	}
 	err := retryTimeDuration(func() (opErr error) {
 		op, opErr = w.Service.Projects.Locations.Operations.Get(name).Do()
 		return opErr
@@ -88,6 +99,7 @@ func (w *ContainerOperationWaiter) TargetStates() []string {
 func containerOperationWait(config *Config, op *container.Operation, project, location, activity string, timeoutMinutes int) error {
 	w := &ContainerOperationWaiter{
 		Service:  config.clientContainerBeta,
+		Context:  config.context,
 		Op:       op,
 		Project:  project,
 		Location: location,

google/provider.go

@@ -462,7 +462,7 @@ func Provider() terraform.ResourceProvider {
 			// We can therefore assume that if it's missing it's 0.10 or 0.11
 			terraformVersion = "0.11+compatible"
 		}
-		return providerConfigure(d, terraformVersion)
+		return providerConfigure(d, provider, terraformVersion)
 	}
 
 	return provider
@@ -721,7 +721,7 @@ func ResourceMapWithErrors() (map[string]*schema.Resource, error) {
 	)
 }
 
-func providerConfigure(d *schema.ResourceData, terraformVersion string) (interface{}, error) {
+func providerConfigure(d *schema.ResourceData, p *schema.Provider, terraformVersion string) (interface{}, error) {
 	config := Config{
 		Project:             d.Get("project").(string),
 		Region:              d.Get("region").(string),
@@ -814,7 +814,7 @@ func providerConfigure(d *schema.ResourceData, terraformVersion string) (interfa
 	config.StorageTransferBasePath = d.Get(StorageTransferCustomEndpointEntryKey).(string)
 	config.BigtableAdminBasePath = d.Get(BigtableAdminCustomEndpointEntryKey).(string)
 
-	if err := config.LoadAndValidate(); err != nil {
+	if err := config.LoadAndValidate(p.StopContext()); err != nil {
 		return nil, err
 	}
 

google/resource_composer_environment_test.go

@@ -1,6 +1,7 @@
 package google
 
 import (
+	"context"
 	"fmt"
 	"testing"
 
@@ -551,7 +552,7 @@ func testSweepComposerResources(region string) error {
 		return fmt.Errorf("error getting shared config for region: %s", err)
 	}
 
-	err = config.LoadAndValidate()
+	err = config.LoadAndValidate(context.Background())
 	if err != nil {
 		log.Fatalf("error loading: %s", err)
 	}

google/resource_compute_instance_migrate_test.go

@@ -1,6 +1,7 @@
 package google
 
 import (
+	"context"
 	"fmt"
 	"log"
 	"os"
@@ -941,7 +942,7 @@ func getInitializedConfig(t *testing.T) *Config {
 
 	ConfigureBasePaths(config)
 
-	err := config.LoadAndValidate()
+	err := config.LoadAndValidate(context.Background())
 	if err != nil {
 		t.Fatal(err)
 	}

google/resource_container_cluster.go

@@ -109,6 +109,11 @@ func resourceContainerCluster() *schema.Resource {
 				},
 			},
 
+			"operation": {
+				Type:     schema.TypeString,
+				Computed: true,
+			},
+
 			"location": {
 				Type:     schema.TypeString,
 				Optional: true,
@@ -916,6 +921,18 @@ func resourceContainerClusterCreate(d *schema.ResourceData, meta interface{}) er
 	timeoutInMinutes := int(d.Timeout(schema.TimeoutCreate).Minutes())
 	waitErr := containerOperationWait(config, op, project, location, "creating GKE cluster", timeoutInMinutes)
 	if waitErr != nil {
+		// Check if the create operation failed because Terraform was prematurely terminated. If it was we can persist the
+		// operation id to state so that a subsequent refresh of this resource will wait until the operation has terminated
+		// before attempting to Read the state of the cluster. This allows a graceful resumption of a Create that was killed
+		// by the upstream Terraform process exiting early such as a sigterm.
+		select {
+		case <-config.context.Done():
+			log.Printf("[DEBUG] Persisting %s so this operation can be resumed \n", op.Name)
+			d.Set("operation", op.Name)
+			return nil
+		default:
+			// leaving default case to ensure this is non blocking
+		}
 		// Try a GET on the cluster so we can see the state in debug logs. This will help classify error states.
 		_, getErr := config.clientContainerBeta.Projects.Locations.Clusters.Get(containerClusterFullName(project, location, clusterName)).Do()
 		if getErr != nil {
@@ -974,6 +991,19 @@ func resourceContainerClusterRead(d *schema.ResourceData, meta interface{}) erro
 		return err
 	}
 
+	operation := d.Get("operation").(string)
+	if operation != "" {
+		log.Printf("[DEBUG] in progress operation detected at %v, attempting to resume", operation)
+		op := &containerBeta.Operation{
+			Name: operation,
+		}
+		d.Set("operation", "")
+		waitErr := containerOperationWait(config, op, project, location, "resuming GKE cluster", int(d.Timeout(schema.TimeoutCreate).Minutes()))
+		if waitErr != nil {
+			return waitErr
+		}
+	}
+
 	clusterName := d.Get("name").(string)
 	name := containerClusterFullName(project, location, clusterName)
 	cluster, err := config.clientContainerBeta.Projects.Locations.Clusters.Get(name).Do()

google/resource_container_cluster_test.go

@@ -2,6 +2,7 @@ package google
 
 import (
 	"bytes"
+	"context"
 	"fmt"
 	"log"
 	"regexp"
@@ -27,7 +28,7 @@ func testSweepContainerClusters(region string) error {
 		log.Fatalf("error getting shared config for region: %s", err)
 	}
 
-	err = config.LoadAndValidate()
+	err = config.LoadAndValidate(context.Background())
 	if err != nil {
 		log.Fatalf("error loading: %s", err)
 	}

google/resource_monitoring_group_test.go

@@ -1,6 +1,7 @@
 package google
 
 import (
+	"context"
 	"fmt"
 	"log"
 	"strings"
@@ -23,7 +24,7 @@ func testSweepMonitoringGroups(region string) error {
 		log.Fatalf("error getting shared config for region: %s", err)
 	}
 
-	err = config.LoadAndValidate()
+	err = config.LoadAndValidate(context.Background())
 	if err != nil {
 		log.Fatalf("error loading: %s", err)
 	}

google/resource_sql_database_instance_test.go

@@ -1,6 +1,7 @@
 package google
 
 import (
+	"context"
 	"fmt"
 	"log"
 	"strings"
@@ -40,7 +41,7 @@ func testSweepDatabases(region string) error {
 		return fmt.Errorf("error getting shared config for region: %s", err)
 	}
 
-	err = config.LoadAndValidate()
+	err = config.LoadAndValidate(context.Background())
 	if err != nil {
 		log.Fatalf("error loading: %s", err)
 	}