feat: Add infractl wait command

.golangci.yml

@@ -92,6 +92,13 @@ linters:
       - linters:
           - revive
         text: avoid meaningless package names
+      - linters:
+          - revive
+        text: avoid package names that conflict with Go standard library package names
+        paths:
+          - cmd/infractl/flavor/list/
+          - pkg/buildinfo/
+          - pkg/service/metrics/
     paths:
       - third_party$
       - builtin$

cmd/infractl/cluster/create/command.go

@@ -4,11 +4,9 @@ package create
 import (
 	"context"
 	"fmt"
-	"os"
 	"strings"
 	"time"
 
-	"github.com/pkg/errors"
 	"github.com/spf13/cobra"
 	"github.com/stackrox/infra/cmd/infractl/cluster/artifacts"
 	"github.com/stackrox/infra/cmd/infractl/cluster/utils"
@@ -65,6 +63,7 @@ func Command() *cobra.Command {
 	cmd.Flags().String("description", "", "description for this cluster")
 	cmd.Flags().Duration("lifespan", 3*time.Hour, "initial lifespan of the cluster")
 	cmd.Flags().Bool("wait", false, "wait for cluster to be ready")
+	common.AddMaxWaitErrorsFlag(cmd)
 	cmd.Flags().Bool("no-slack", false, "skip sending Slack messages for lifecycle events")
 	cmd.Flags().Bool("slack-me", false, "send slack messages directly and not to the #infra_notifications channel")
 	cmd.Flags().StringP("download-dir", "d", "", "wait for readiness and download artifacts to this dir")
@@ -82,6 +81,7 @@ func run(ctx context.Context, conn *grpc.ClientConn, cmd *cobra.Command, args []
 	description, _ := cmd.Flags().GetString("description")
 	lifespan, _ := cmd.Flags().GetDuration("lifespan")
 	wait, _ := cmd.Flags().GetBool("wait")
+	maxWaitErrors := common.GetMaxWaitErrorsFlagValue(cmd)
 	noSlack, _ := cmd.Flags().GetBool("no-slack")
 	slackDM, _ := cmd.Flags().GetBool("slack-me")
 	downloadDir, _ := cmd.Flags().GetString("download-dir")
@@ -129,7 +129,7 @@ func run(ctx context.Context, conn *grpc.ClientConn, cmd *cobra.Command, args []
 	}
 
 	if wait {
-		if err := waitForCluster(client, clusterID); err != nil {
+		if err := common.WaitForCluster(client, clusterID, maxWaitErrors); err != nil {
 			return nil, err
 		}
 		if downloadDir != "" {
@@ -161,36 +161,6 @@ func assignDefaults(cmd *cobra.Command, req *v1.CreateClusterRequest, cwe *curre
 	req.Parameters["main-image"] = registry + "/main:" + getCleaned(cwe.tag)
 }
 
-func waitForCluster(client v1.ClusterServiceClient, clusterID *v1.ResourceByID) error {
-	const timeoutSleep = 30 * time.Second
-	const timeoutAPI = 15 * time.Second
-
-	fmt.Fprintf(os.Stderr, "...creating %s\n", clusterID.Id)
-	for {
-		time.Sleep(timeoutSleep)
-		ctx, cancel := context.WithTimeout(context.Background(), timeoutAPI)
-
-		cluster, err := client.Info(ctx, clusterID)
-		cancel()
-		if err != nil {
-			fmt.Fprintln(os.Stderr, "...error")
-			continue
-		}
-
-		switch cluster.Status {
-		case v1.Status_CREATING:
-			fmt.Fprintln(os.Stderr, "...creating")
-			continue
-		case v1.Status_READY:
-			fmt.Fprintln(os.Stderr, "...ready")
-			return nil
-		default:
-			fmt.Fprintln(os.Stderr, "...failed")
-			return errors.New("failed to provision cluster")
-		}
-	}
-}
-
 func displayUserNotes(cmd *cobra.Command, args []string, req *v1.CreateClusterRequest, cwe *currentWorkingEnvironment) {
 	if wasNameProvided(args) {
 		if isQaDemoFlavor(args[0]) && cwe.isInStackroxRepo() {

cmd/infractl/cluster/wait/command.go

@@ -0,0 +1,49 @@
+// Package wait implements the infractl wait command.
+package wait
+
+import (
+	"context"
+	"errors"
+
+	"github.com/spf13/cobra"
+	"github.com/stackrox/infra/cmd/infractl/cluster/utils"
+	"github.com/stackrox/infra/cmd/infractl/common"
+	v1 "github.com/stackrox/infra/generated/api/v1"
+	"google.golang.org/grpc"
+)
+
+const examples = `Wait for the "example-s3maj" cluster to become ready.
+$ infractl wait example-s3maj`
+
+// Command defines the handler for infractl wait.
+func Command() *cobra.Command {
+	// $ infractl wait
+	cmd := &cobra.Command{
+		Use:     "wait CLUSTER",
+		Short:   "Wait for a specific cluster",
+		Long:    "Wait for the the specific cluster to become ready.",
+		Example: examples,
+		Args:    common.ArgsWithHelp(cobra.ExactArgs(1), args),
+		RunE:    common.WithGRPCHandler(run),
+	}
+
+	common.AddMaxWaitErrorsFlag(cmd)
+
+	return cmd
+}
+
+func args(_ *cobra.Command, args []string) error {
+	if args[0] == "" {
+		return errors.New("no cluster ID given")
+	}
+	return utils.ValidateClusterName(args[0])
+}
+
+func run(_ context.Context, conn *grpc.ClientConn, cmd *cobra.Command, args []string) (common.PrettyPrinter, error) {
+	maxWaitErrors := common.GetMaxWaitErrorsFlagValue(cmd)
+
+	client := v1.NewClusterServiceClient(conn)
+	err := common.WaitForCluster(client, &v1.ResourceByID{Id: args[0]}, maxWaitErrors)
+
+	return prettyNoop{}, err
+}

cmd/infractl/cluster/wait/fancy.go

@@ -0,0 +1,18 @@
+package wait
+
+import (
+	"github.com/spf13/cobra"
+)
+
+// prettyNoop does not output anything because 'wait' command does not need output.
+type prettyNoop struct {
+}
+
+func (p prettyNoop) PrettyPrint(cmd *cobra.Command) {
+	cmd.Printf("\n")
+}
+
+func (p prettyNoop) PrettyJSONPrint(cmd *cobra.Command) error {
+	cmd.Printf("{}\n")
+	return nil
+}

cmd/infractl/common/wait.go

@@ -0,0 +1,66 @@
+package common
+
+import (
+	"errors"
+	"fmt"
+	"os"
+	"time"
+
+	"github.com/spf13/cobra"
+	v1 "github.com/stackrox/infra/generated/api/v1"
+)
+
+const (
+	maxWaitErrorsFlagName           = "wait-max-errors"
+	defaultMaxConsecutiveWaitErrors = 10
+)
+
+// AddMaxWaitErrorsFlag adds a flag definition to cmd.
+func AddMaxWaitErrorsFlag(cmd *cobra.Command) {
+	cmd.Flags().Int(maxWaitErrorsFlagName, defaultMaxConsecutiveWaitErrors, "maximum number of consecutive errors before giving up waiting")
+}
+
+// GetMaxWaitErrorsFlagValue gets effective value of the flag after arguments are parsed.
+func GetMaxWaitErrorsFlagValue(cmd *cobra.Command) int {
+	value, err := cmd.Flags().GetInt(maxWaitErrorsFlagName)
+	if err != nil {
+		panic(err)
+	}
+	return value
+}
+
+// WaitForCluster waits for a created cluster to be in a ready state.
+func WaitForCluster(client v1.ClusterServiceClient, clusterID *v1.ResourceByID, maxWaitErrors int) error {
+	const timeoutSleep = 30 * time.Second
+
+	nErrors := 0
+
+	fmt.Fprintf(os.Stderr, "...waiting for %s\n", clusterID.Id)
+	for {
+		ctx, cancel := ContextWithTimeout()
+		cluster, err := client.Info(ctx, clusterID)
+		cancel()
+
+		if err != nil {
+			fmt.Fprintf(os.Stderr, "...error %s\n", err)
+			nErrors++
+			if nErrors >= maxWaitErrors {
+				return errors.New("too many errors while waiting")
+			}
+		} else {
+			nErrors = 0
+			switch cluster.Status {
+			case v1.Status_CREATING:
+				fmt.Fprintln(os.Stderr, "...creating")
+			case v1.Status_READY:
+				fmt.Fprintln(os.Stderr, "...ready")
+				return nil
+			default:
+				fmt.Fprintln(os.Stderr, "...failed")
+				return errors.New("cluster failed provisioning")
+			}
+		}
+
+		time.Sleep(timeoutSleep)
+	}
+}

cmd/infractl/main.go

@@ -12,6 +12,7 @@ import (
 	"github.com/stackrox/infra/cmd/infractl/cluster/lifespan"
 	"github.com/stackrox/infra/cmd/infractl/cluster/list"
 	"github.com/stackrox/infra/cmd/infractl/cluster/logs"
+	"github.com/stackrox/infra/cmd/infractl/cluster/wait"
 	"github.com/stackrox/infra/cmd/infractl/common"
 	"github.com/stackrox/infra/cmd/infractl/flavor"
 	janitorFind "github.com/stackrox/infra/cmd/infractl/janitor/find"
@@ -98,6 +99,9 @@ func main() {
 		// $ infractl version
 		version.Command(),
 
+		// $ infractl wait
+		wait.Command(),
+
 		// $ infractl whoami
 		whoami.Command(),
 	)

test/e2e/cluster/wait_test.go

@@ -0,0 +1,28 @@
+//go:build e2e
+// +build e2e
+
+package cluster_test
+
+import (
+	"testing"
+
+	"github.com/stackrox/infra/test/utils"
+	"github.com/stackrox/infra/test/utils/mock"
+	"github.com/stretchr/testify/assert"
+)
+
+func TestWait(t *testing.T) {
+	utils.CheckContext()
+	clusterID, err := mock.InfractlCreateCluster(
+		"test-simulate",
+		"--lifespan=60s",
+	)
+	assert.NoError(t, err)
+
+	err = mock.InfractlWait(clusterID)
+	assert.NoError(t, err)
+
+	cluster, err := mock.InfractlGetCluster(clusterID)
+	assert.NoError(t, err)
+	assert.Equal(t, "READY", cluster.Status)
+}

test/utils/mock/infractl.go

@@ -10,6 +10,7 @@ import (
 	infraClusterLifespan "github.com/stackrox/infra/cmd/infractl/cluster/lifespan"
 	infraClusterList "github.com/stackrox/infra/cmd/infractl/cluster/list"
 	infraClusterLogs "github.com/stackrox/infra/cmd/infractl/cluster/logs"
+	infraClusterWait "github.com/stackrox/infra/cmd/infractl/cluster/wait"
 	infraFlavorGet "github.com/stackrox/infra/cmd/infractl/flavor/get"
 	infraFlavorList "github.com/stackrox/infra/cmd/infractl/flavor/list"
 	infraJanitorFind "github.com/stackrox/infra/cmd/infractl/janitor/find"
@@ -107,6 +108,13 @@ func InfractlLogs(clusterID string) (*v1.LogsResponse, error) {
 	return jsonData, nil
 }
 
+// InfractlWait is a warpper for 'infractl wait <clusterID>'.
+func InfractlWait(clusterID string) error {
+	infraWaitCmd := infraClusterWait.Command()
+	PrepareCommand(infraWaitCmd, false, clusterID)
+	return infraWaitCmd.Execute()
+}
+
 // InfractlWhoami is a wrapper for 'infractl whoami'.
 func InfractlWhoami() (string, error) {
 	whoamiCmd := infraWhoami.Command()