porter/cli/cmd/run.go

package cmd

import (
	"context"
	"errors"
	"fmt"
	"io"
	"os"
	"strings"
	"time"

	"github.com/fatih/color"
	api "github.com/porter-dev/porter/api/client"
	"github.com/porter-dev/porter/api/types"
	"github.com/porter-dev/porter/cli/cmd/utils"
	"github.com/spf13/cobra"
	batchv1 "k8s.io/api/batch/v1"
	v1 "k8s.io/api/core/v1"
	rbacv1 "k8s.io/api/rbac/v1"
	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
	"k8s.io/apimachinery/pkg/fields"
	"k8s.io/apimachinery/pkg/watch"
	"k8s.io/kubectl/pkg/util/term"

	"k8s.io/apimachinery/pkg/runtime"
	"k8s.io/apimachinery/pkg/runtime/schema"
	"k8s.io/client-go/kubernetes"
	"k8s.io/client-go/rest"
	"k8s.io/client-go/tools/clientcmd"
	"k8s.io/client-go/tools/remotecommand"
)

var namespace string
var verbose bool

// runCmd represents the "porter run" base command when called
// without any subcommands
var runCmd = &cobra.Command{
	Use:   "run [release] -- COMMAND [args...]",
	Args:  cobra.MinimumNArgs(2),
	Short: "Runs a command inside a connected cluster container.",
	Run: func(cmd *cobra.Command, args []string) {
		err := checkLoginAndRun(args, run)

		if err != nil {
			os.Exit(1)
		}
	},
}

var existingPod bool

func init() {
	rootCmd.AddCommand(runCmd)

	runCmd.PersistentFlags().StringVar(
		&namespace,
		"namespace",
		"default",
		"namespace of release to connect to",
	)

	runCmd.PersistentFlags().BoolVarP(
		&existingPod,
		"existing_pod",
		"e",
		false,
		"whether to connect to an existing pod",
	)

	runCmd.PersistentFlags().BoolVarP(
		&verbose,
		"verbose",
		"v",
		false,
		"whether to print verbose output",
	)
}

func run(_ *types.GetAuthenticatedUserResponse, client *api.Client, args []string) error {
	color.New(color.FgGreen).Println("Running", strings.Join(args[1:], " "), "for release", args[0])

	podsSimple, err := getPods(client, namespace, args[0])

	if err != nil {
		return fmt.Errorf("Could not retrieve list of pods: %s", err.Error())
	}

	// if length of pods is 0, throw error
	var selectedPod podSimple

	if len(podsSimple) == 0 {
		return fmt.Errorf("At least one pod must exist in this deployment.")
	} else if len(podsSimple) == 1 || !existingPod {
		selectedPod = podsSimple[0]
	} else {
		podNames := make([]string, 0)

		for _, podSimple := range podsSimple {
			podNames = append(podNames, podSimple.Name)
		}

		selectedPodName, err := utils.PromptSelect("Select the pod:", podNames)

		if err != nil {
			return err
		}

		// find selected pod
		for _, podSimple := range podsSimple {
			if selectedPodName == podSimple.Name {
				selectedPod = podSimple
			}
		}
	}

	var selectedContainerName string

	// if the selected pod has multiple container, spawn selector
	if len(selectedPod.ContainerNames) == 0 {
		return fmt.Errorf("At least one pod must exist in this deployment.")
	} else if len(selectedPod.ContainerNames) == 1 {
		selectedContainerName = selectedPod.ContainerNames[0]
	} else {
		selectedContainer, err := utils.PromptSelect("Select the container:", selectedPod.ContainerNames)

		if err != nil {
			return err
		}

		selectedContainerName = selectedContainer
	}

	config := &PorterRunSharedConfig{
		Client: client,
	}

	err = config.setSharedConfig()

	if err != nil {
		return fmt.Errorf("Could not retrieve kube credentials: %s", err.Error())
	}

	if existingPod {
		return executeRun(config, namespace, selectedPod.Name, selectedContainerName, args[1:])
	}

	return executeRunEphemeral(config, namespace, selectedPod.Name, selectedContainerName, args[1:])
}

type PorterRunSharedConfig struct {
	Client     *api.Client
	RestConf   *rest.Config
	Clientset  *kubernetes.Clientset
	RestClient *rest.RESTClient
}

func (p *PorterRunSharedConfig) setSharedConfig() error {
	pID := config.Project
	cID := config.Cluster

	kubeResp, err := p.Client.GetKubeconfig(context.TODO(), pID, cID)

	if err != nil {
		return err
	}

	kubeBytes := kubeResp.Kubeconfig

	cmdConf, err := clientcmd.NewClientConfigFromBytes(kubeBytes)

	if err != nil {
		return err
	}

	restConf, err := cmdConf.ClientConfig()

	if err != nil {
		return err
	}

	restConf.GroupVersion = &schema.GroupVersion{
		Group:   "api",
		Version: "v1",
	}

	restConf.NegotiatedSerializer = runtime.NewSimpleNegotiatedSerializer(runtime.SerializerInfo{})

	p.RestConf = restConf

	clientset, err := kubernetes.NewForConfig(restConf)

	if err != nil {
		return err
	}

	p.Clientset = clientset

	restClient, err := rest.RESTClientFor(restConf)

	if err != nil {
		return err
	}

	p.RestClient = restClient

	return nil
}

type podSimple struct {
	Name           string
	ContainerNames []string
}

func getPods(client *api.Client, namespace, releaseName string) ([]podSimple, error) {
	pID := config.Project
	cID := config.Cluster

	resp, err := client.GetK8sAllPods(context.TODO(), pID, cID, namespace, releaseName)

	if err != nil {
		return nil, err
	}

	pods := *resp

	res := make([]podSimple, 0)

	for _, pod := range pods {
		containerNames := make([]string, 0)

		for _, container := range pod.Spec.Containers {
			containerNames = append(containerNames, container.Name)
		}

		res = append(res, podSimple{
			Name:           pod.ObjectMeta.Name,
			ContainerNames: containerNames,
		})
	}

	return res, nil
}

func executeRun(config *PorterRunSharedConfig, namespace, name, container string, args []string) error {
	req := config.RestClient.Post().
		Resource("pods").
		Name(name).
		Namespace(namespace).
		SubResource("exec")

	for _, arg := range args {
		req.Param("command", arg)
	}
	req.Param("stdin", "true")
	req.Param("stdout", "true")
	req.Param("tty", "true")
	req.Param("container", container)

	t := term.TTY{
		In:  os.Stdin,
		Out: os.Stdout,
		Raw: true,
	}
	size := t.GetSize()
	sizeQueue := t.MonitorSize(size)

	return t.Safe(func() error {
		exec, err := remotecommand.NewSPDYExecutor(config.RestConf, "POST", req.URL())

		if err != nil {
			return err
		}

		return exec.Stream(remotecommand.StreamOptions{
			Stdin:  os.Stdin,
			Stdout: os.Stdout,
			Stderr: os.Stderr,
			Tty:    true,

			TerminalSizeQueue: sizeQueue,
		})
	})
}

func executeRunEphemeral(config *PorterRunSharedConfig, namespace, name, container string, args []string) error {
	existing, err := getExistingPod(config, name, namespace)

	if err != nil {
		return err
	}

	newPod, err := createEphemeralPodFromExisting(config, existing, args)
	if err != nil {
		return err
	}
	podName := newPod.ObjectMeta.Name

	// delete the ephemeral pod no matter what
	defer deletePod(config, podName, namespace)

	color.New(color.FgYellow).Printf("Waiting for pod %s to be ready...", podName)
	if err = waitForPod(config, newPod); err != nil {
		color.New(color.FgRed).Println("failed")
		return handlePodAttachError(err, config, namespace, podName, container)
	}

	err = checkForPodDeletionCronJob(config)
	if err != nil {
		return err
	}

	// refresh pod info for latest status
	newPod, err = config.Clientset.CoreV1().
		Pods(newPod.Namespace).
		Get(context.Background(), newPod.Name, metav1.GetOptions{})

	// pod exited while we were waiting.  maybe an error maybe not.
	// we dont know if the user wanted an interactive shell or not.
	// if it was an error the logs hopefully say so.
	if isPodExited(newPod) {
		color.New(color.FgGreen).Println("complete!")
		var writtenBytes int64
		writtenBytes, _ = pipePodLogsToStdout(config, namespace, podName, container, false)

		if verbose || writtenBytes == 0 {
			color.New(color.FgYellow).Println("Could not get logs. Pod events:")
			pipeEventsToStdout(config, namespace, podName, container, false)
		}
		return nil
	}
	color.New(color.FgGreen).Println("ready!")

	color.New(color.FgYellow).Println("Attempting connection to the container. If you don't see a command prompt, try pressing enter.")
	req := config.RestClient.Post().
		Resource("pods").
		Name(podName).
		Namespace(namespace).
		SubResource("attach")

	req.Param("stdin", "true")
	req.Param("stdout", "true")
	req.Param("tty", "true")
	req.Param("container", container)

	t := term.TTY{
		In:  os.Stdin,
		Out: os.Stdout,
		Raw: true,
	}
	size := t.GetSize()
	sizeQueue := t.MonitorSize(size)

	if err = t.Safe(func() error {
		exec, err := remotecommand.NewSPDYExecutor(config.RestConf, "POST", req.URL())
		if err != nil {
			return err
		}
		return exec.Stream(remotecommand.StreamOptions{
			Stdin:  os.Stdin,
			Stdout: os.Stdout,
			Stderr: os.Stderr,
			Tty:    true,

			TerminalSizeQueue: sizeQueue,
		})
	}); err != nil {
		// ugly way to catch no TTY errors, such as when running command "echo \"hello\""
		return handlePodAttachError(err, config, namespace, podName, container)
	}

	if verbose {
		color.New(color.FgYellow).Println("Pod events:")
		pipeEventsToStdout(config, namespace, podName, container, false)
	}

	return err
}

func checkForPodDeletionCronJob(config *PorterRunSharedConfig) error {
	namespaces, err := config.Clientset.CoreV1().Namespaces().List(context.Background(), metav1.ListOptions{})
	if err != nil {
		return err
	}

	for _, namespace := range namespaces.Items {
		cronJobs, err := config.Clientset.BatchV1().CronJobs(namespace.Name).List(
			context.Background(), metav1.ListOptions{},
		)
		if err != nil {
			return err
		}

		for _, cronJob := range cronJobs.Items {
			if cronJob.Name == "porter-ephemeral-pod-deletion-cronjob" {
				return nil
			}
		}
	}

	// try and create the cron job and all of the other required resources as necessary,
	// starting with the service account, then role and then a role binding

	err = checkForServiceAccount(config)
	if err != nil {
		return err
	}

	err = checkForRole(config)
	if err != nil {
		return err
	}

	err = checkForRoleBinding(config)
	if err != nil {
		return err
	}

	// create the cronjob

	cronJob := &batchv1.CronJob{
		ObjectMeta: metav1.ObjectMeta{
			Name: "porter-ephemeral-pod-deletion-cronjob",
		},
		Spec: batchv1.CronJobSpec{
			Schedule: "* 6 * * *",
			JobTemplate: batchv1.JobTemplateSpec{
				Spec: batchv1.JobSpec{
					Template: v1.PodTemplateSpec{
						Spec: v1.PodSpec{
							ServiceAccountName: "porter-ephemeral-pod-deletion-service-account",
							RestartPolicy:      v1.RestartPolicyNever,
							Containers: []v1.Container{
								{
									Name:            "ephemeral-pods-manager",
									Image:           "public.ecr.aws/o1j4x7p4/ephemeral-pods-manager:latest",
									ImagePullPolicy: v1.PullAlways,
									Args:            []string{"delete"},
								},
							},
						},
					},
				},
			},
		},
	}
	_, err = config.Clientset.BatchV1().CronJobs(namespace).Create(
		context.Background(), cronJob, metav1.CreateOptions{},
	)
	if err != nil {
		return err
	}

	return nil
}

func checkForServiceAccount(config *PorterRunSharedConfig) error {
	serviceAccounts, err := config.Clientset.CoreV1().ServiceAccounts(namespace).List(
		context.Background(), metav1.ListOptions{},
	)
	if err != nil {
		return err
	}

	for _, serviceAccount := range serviceAccounts.Items {
		if serviceAccount.Name == "porter-ephemeral-pod-deletion-service-account" {
			return nil
		}
	}

	serviceAccount := &v1.ServiceAccount{
		ObjectMeta: metav1.ObjectMeta{
			Name: "porter-ephemeral-pod-deletion-service-account",
		},
	}
	_, err = config.Clientset.CoreV1().ServiceAccounts(namespace).Create(
		context.Background(), serviceAccount, metav1.CreateOptions{},
	)
	if err != nil {
		return err
	}

	return nil
}

func checkForRole(config *PorterRunSharedConfig) error {
	roles, err := config.Clientset.RbacV1().ClusterRoles().List(
		context.Background(), metav1.ListOptions{},
	)
	if err != nil {
		return err
	}

	for _, role := range roles.Items {
		if role.Name == "porter-ephemeral-pod-deletion-cluster-role" {
			return nil
		}
	}

	role := &rbacv1.ClusterRole{
		ObjectMeta: metav1.ObjectMeta{
			Name: "porter-ephemeral-pod-deletion-cluster-role",
		},
		Rules: []rbacv1.PolicyRule{
			{
				APIGroups: []string{""},
				Resources: []string{"pods"},
				Verbs:     []string{"list", "delete"},
			},
			{
				APIGroups: []string{""},
				Resources: []string{"namespaces"},
				Verbs:     []string{"list"},
			},
		},
	}
	_, err = config.Clientset.RbacV1().ClusterRoles().Create(
		context.Background(), role, metav1.CreateOptions{},
	)
	if err != nil {
		return err
	}

	return nil
}

func checkForRoleBinding(config *PorterRunSharedConfig) error {
	bindings, err := config.Clientset.RbacV1().ClusterRoleBindings().List(
		context.Background(), metav1.ListOptions{},
	)
	if err != nil {
		return err
	}

	for _, binding := range bindings.Items {
		if binding.Name == "porter-ephemeral-pod-deletion-cluster-rolebinding" {
			return nil
		}
	}

	binding := &rbacv1.ClusterRoleBinding{
		ObjectMeta: metav1.ObjectMeta{
			Name: "porter-ephemeral-pod-deletion-cluster-rolebinding",
		},
		RoleRef: rbacv1.RoleRef{
			APIGroup: "rbac.authorization.k8s.io",
			Kind:     "ClusterRole",
			Name:     "porter-ephemeral-pod-deletion-cluster-role",
		},
		Subjects: []rbacv1.Subject{
			{
				APIGroup: "rbac.authorization.k8s.io",
				Kind:     "ServiceAccount",
				Name:     "porter-ephemeral-pod-deletion-service-account",
			},
		},
	}
	_, err = config.Clientset.RbacV1().ClusterRoleBindings().Create(
		context.Background(), binding, metav1.CreateOptions{},
	)
	if err != nil {
		return err
	}

	return nil
}

func waitForPod(config *PorterRunSharedConfig, pod *v1.Pod) error {
	var (
		w   watch.Interface
		err error
		ok  bool
	)
	// immediately after creating a pod, the API may return a 404. heuristically 1
	// second seems to be plenty.
	watchRetries := 3
	for i := 0; i < watchRetries; i++ {
		selector := fields.OneTermEqualSelector("metadata.name", pod.Name).String()
		w, err = config.Clientset.CoreV1().
			Pods(pod.Namespace).
			Watch(context.Background(), metav1.ListOptions{FieldSelector: selector})

		if err == nil {
			break
		}
		time.Sleep(time.Second)
	}
	if err != nil {
		return err
	}
	defer w.Stop()
	for {
		select {
		case <-time.Tick(time.Second):
			// poll every second in case we already missed the ready event while
			// creating the listener.
			pod, err = config.Clientset.CoreV1().
				Pods(pod.Namespace).
				Get(context.Background(), pod.Name, metav1.GetOptions{})
			if isPodReady(pod) || isPodExited(pod) {
				return nil
			}
		case evt := <-w.ResultChan():
			pod, ok = evt.Object.(*v1.Pod)
			if !ok {
				return fmt.Errorf("unexpected object type: %T", evt.Object)
			}
			if isPodReady(pod) || isPodExited(pod) {
				return nil
			}
		case <-time.After(time.Second * 10):
			return errors.New("timed out waiting for pod")
		}
	}
}

func isPodReady(pod *v1.Pod) bool {
	ready := false
	conditions := pod.Status.Conditions
	for i := range conditions {
		if conditions[i].Type == v1.PodReady {
			ready = pod.Status.Conditions[i].Status == v1.ConditionTrue
		}
	}
	return ready
}

func isPodExited(pod *v1.Pod) bool {
	return pod.Status.Phase == v1.PodSucceeded || pod.Status.Phase == v1.PodFailed
}

func handlePodAttachError(err error, config *PorterRunSharedConfig, namespace, podName, container string) error {
	if verbose {
		color.New(color.FgYellow).Printf("Error: %s\n", err)
	}
	color.New(color.FgYellow).Println("Could not open a shell to this container. Container logs:")

	var writtenBytes int64
	writtenBytes, _ = pipePodLogsToStdout(config, namespace, podName, container, false)

	if verbose || writtenBytes == 0 {
		color.New(color.FgYellow).Println("Could not get logs. Pod events:")
		pipeEventsToStdout(config, namespace, podName, container, false)
	}
	return err
}

func pipePodLogsToStdout(config *PorterRunSharedConfig, namespace, name, container string, follow bool) (int64, error) {
	podLogOpts := v1.PodLogOptions{
		Container: container,
		Follow:    follow,
	}

	req := config.Clientset.CoreV1().Pods(namespace).GetLogs(name, &podLogOpts)

	podLogs, err := req.Stream(
		context.Background(),
	)

	if err != nil {
		return 0, err
	}

	defer podLogs.Close()

	return io.Copy(os.Stdout, podLogs)
}

func pipeEventsToStdout(config *PorterRunSharedConfig, namespace, name, container string, follow bool) error {
	// update the config in case the operation has taken longer than token expiry time
	config.setSharedConfig()

	// creates the clientset
	resp, err := config.Clientset.CoreV1().Events(namespace).List(
		context.TODO(),
		metav1.ListOptions{
			FieldSelector: fmt.Sprintf("involvedObject.name=%s,involvedObject.namespace=%s", name, namespace),
		},
	)

	if err != nil {
		return err
	}

	for _, event := range resp.Items {
		color.New(color.FgRed).Println(event.Message)
	}

	return nil
}

func getExistingPod(config *PorterRunSharedConfig, name, namespace string) (*v1.Pod, error) {
	return config.Clientset.CoreV1().Pods(namespace).Get(
		context.Background(),
		name,
		metav1.GetOptions{},
	)
}

func deletePod(config *PorterRunSharedConfig, name, namespace string) error {
	// update the config in case the operation has taken longer than token expiry time
	config.setSharedConfig()

	err := config.Clientset.CoreV1().Pods(namespace).Delete(
		context.Background(),
		name,
		metav1.DeleteOptions{},
	)

	if err != nil {
		color.New(color.FgRed).Printf("Could not delete ephemeral pod: %s\n", err.Error())
		return err
	}

	color.New(color.FgGreen).Println("Sucessfully deleted ephemeral pod")

	return nil
}

func createEphemeralPodFromExisting(config *PorterRunSharedConfig, existing *v1.Pod, args []string) (*v1.Pod, error) {
	newPod := existing.DeepCopy()

	// only copy the pod spec, overwrite metadata
	newPod.ObjectMeta = metav1.ObjectMeta{
		Name:      strings.ToLower(fmt.Sprintf("%s-copy-%s", existing.ObjectMeta.Name, utils.String(4))),
		Namespace: existing.ObjectMeta.Namespace,
	}

	newPod.Status = v1.PodStatus{}

	// only use "primary" container
	newPod.Spec.Containers = newPod.Spec.Containers[0:1]

	// set restart policy to never
	newPod.Spec.RestartPolicy = v1.RestartPolicyNever

	// change the command in the pod to the passed in pod command
	cmdRoot := args[0]
	cmdArgs := make([]string, 0)

	// annotate with the ephemeral pod tag
	newPod.Labels = make(map[string]string)
	newPod.Labels["porter/ephemeral-pod"] = "true"

	if len(args) > 1 {
		cmdArgs = args[1:]
	}

	newPod.Spec.Containers[0].Command = []string{cmdRoot}
	newPod.Spec.Containers[0].Args = cmdArgs
	newPod.Spec.Containers[0].TTY = true
	newPod.Spec.Containers[0].Stdin = true
	newPod.Spec.Containers[0].StdinOnce = true
	newPod.Spec.NodeName = ""

	// remove health checks and probes
	newPod.Spec.Containers[0].LivenessProbe = nil
	newPod.Spec.Containers[0].ReadinessProbe = nil
	newPod.Spec.Containers[0].StartupProbe = nil

	// create the pod and return it
	return config.Clientset.CoreV1().Pods(existing.ObjectMeta.Namespace).Create(
		context.Background(),
		newPod,
		metav1.CreateOptions{},
	)
}