fission-src/poolmgr/gp.go

/*
Copyright 2016 The Fission Authors.

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
*/

package poolmgr

import (
	"bytes"
	"errors"
	"fmt"
	"log"
	"math/rand"
	"net"
	"net/http"
	"net/url"
	"strings"
	"time"

	"github.com/dchest/uniuri"
	"k8s.io/client-go/1.5/kubernetes"
	"k8s.io/client-go/1.5/pkg/api"
	"k8s.io/client-go/1.5/pkg/api/v1"
	"k8s.io/client-go/1.5/pkg/apis/extensions/v1beta1"
	"k8s.io/client-go/1.5/pkg/labels"
	"k8s.io/client-go/1.5/pkg/util/intstr"

	"github.com/fission/fission"
)

const POOLMGR_INSTANCEID_LABEL string = "poolmgrInstanceId"

type (
	GenericPool struct {
		env              *fission.Environment
		replicas         int32               // num containers
		deployment       *v1beta1.Deployment // kubernetes deployment
		namespace        string              // namespace to keep our resources
		podReadyTimeout  time.Duration       // timeout for generic pods to become ready
		controllerUrl    string
		idlePodReapTime  time.Duration         // pods unused for idlePodReapTime are deleted
		fsCache          *functionServiceCache // cache funcSvc's by function, address and podname
		useSvc           bool                  // create service
		poolInstanceId   string                // small random string to uniquify pod names
		kubernetesClient *kubernetes.Clientset
		instanceId       string

		requestChannel chan *choosePodRequest
	}

	// serialize the choosing of pods so that choices don't conflict
	choosePodRequest struct {
		newLabels       map[string]string
		responseChannel chan *choosePodResponse
	}
	choosePodResponse struct {
		pod *v1.Pod
		error
	}
)

func MakeGenericPool(
	controllerUrl string,
	kubernetesClient *kubernetes.Clientset,
	env *fission.Environment,
	initialReplicas int32,
	namespace string,
	fsCache *functionServiceCache,
	instanceId string) (*GenericPool, error) {

	log.Printf("Creating pool for environment %v", env.Metadata)
	// TODO: in general we need to provide the user a way to configure pools.  Initial
	// replicas, autoscaling params, various timeouts, etc.
	gp := &GenericPool{
		env:              env,
		replicas:         initialReplicas, // TODO make this an env param instead?
		requestChannel:   make(chan *choosePodRequest),
		kubernetesClient: kubernetesClient,
		namespace:        namespace,
		podReadyTimeout:  5 * time.Minute, // TODO make this an env param?
		controllerUrl:    controllerUrl,
		idlePodReapTime:  3 * time.Minute, // TODO make this configurable
		fsCache:          fsCache,
		poolInstanceId:   uniuri.NewLen(8),
		instanceId:       instanceId,

		useSvc: false,
	}

	// create the pool
	err := gp.createPool()
	if err != nil {
		return nil, err
	}

	// wait for at least one pod to be ready
	log.Printf("[%v] Deployment created, waiting for a ready pod", env.Metadata)
	err = gp.waitForReadyPod()
	if err != nil {
		return nil, err
	}

	go gp.choosePodService()

	go gp.idlePodReaper()

	return gp, nil
}

// choosePodService serializes the choosing of pods
func (gp *GenericPool) choosePodService() {
	for {
		select {
		case req := <-gp.requestChannel:
			pod, err := gp._choosePod(req.newLabels)
			if err != nil {
				req.responseChannel <- &choosePodResponse{error: err}
				continue
			}
			req.responseChannel <- &choosePodResponse{pod: pod}
		}
	}
}

// choosePod picks a ready pod from the pool and relabels it, waiting if necessary.
// returns the pod API object.
func (gp *GenericPool) choosePod(newLabels map[string]string) (*v1.Pod, error) {
	req := &choosePodRequest{
		newLabels:       newLabels,
		responseChannel: make(chan *choosePodResponse),
	}
	gp.requestChannel <- req
	resp := <-req.responseChannel
	return resp.pod, resp.error
}

// _choosePod is called serially by choosePodService
func (gp *GenericPool) _choosePod(newLabels map[string]string) (*v1.Pod, error) {
	startTime := time.Now()
	for {
		// Retries took too long, error out.
		if time.Now().Sub(startTime) > gp.podReadyTimeout {
			log.Printf("[%v] Erroring out, timed out", newLabels)
			return nil, errors.New("timeout: waited too long to get a ready pod")
		}

		// Get pods; filter the ones that are ready
		podList, err := gp.kubernetesClient.Core().Pods(gp.namespace).List(
			api.ListOptions{
				LabelSelector: labels.Set(
					gp.deployment.Spec.Selector.MatchLabels).AsSelector(),
			})
		if err != nil {
			return nil, err
		}
		readyPods := make([]*v1.Pod, 0, len(podList.Items))
		for _, pod := range podList.Items {
			podReady := true
			for _, cs := range pod.Status.ContainerStatuses {
				podReady = podReady && cs.Ready
			}
			if podReady {
				readyPods = append(readyPods, &pod)
			}
		}
		log.Printf("[%v] found %v ready pods of %v total",
			newLabels, len(readyPods), len(podList.Items))

		// If there are no ready pods, wait and retry.
		if len(readyPods) == 0 {
			err = gp.waitForReadyPod()
			if err != nil {
				return nil, err
			}
			continue
		}

		// Pick a ready pod.  For now just choose randomly;
		// ideally we'd care about which node it's running on,
		// and make a good scheduling decision.
		chosenPod := readyPods[rand.Intn(len(readyPods))]

		// Relabel.  If the pod already got picked and
		// modified, this should fail; in that case just
		// retry.
		chosenPod.ObjectMeta.Labels = newLabels
		log.Printf("relabeling pod: [%v]", chosenPod.ObjectMeta.Name)
		_, err = gp.kubernetesClient.Core().Pods(gp.namespace).Update(chosenPod)
		if err != nil {
			log.Printf("failed to relabel pod [%v]: %v", chosenPod.ObjectMeta.Name, err)
			continue
		}
		log.Printf("Chosen pod: %v (in %v)", chosenPod.ObjectMeta.Name, time.Now().Sub(startTime))
		return chosenPod, nil
	}
}

func (gp *GenericPool) labelsForMetadata(metadata *fission.Metadata) map[string]string {
	return map[string]string{
		"functionName":           metadata.Name,
		"functionUid":            metadata.Uid,
		"unmanaged":              "true", // this allows us to easily find pods not managed by the deployment
		POOLMGR_INSTANCEID_LABEL: gp.instanceId,
	}
}

func (gp *GenericPool) scheduleDeletePod(name string) {
	go func() {
		// The sleep allows debugging or collecting logs from the pod before it's
		// cleaned up.  (We need a better solutions for both those things; log
		// aggregation and storage will help.)
		time.Sleep(5 * time.Minute)
		gp.kubernetesClient.Core().Pods(gp.namespace).Delete(name, nil)
	}()
}

// specializePod chooses a pod, copies the required user-defined function to that pod
// (via fetcher), and calls the function-run container to load it, resulting in a
// specialized pod.
func (gp *GenericPool) specializePod(pod *v1.Pod, metadata *fission.Metadata) error {
	// for fetcher we don't need to create a service, just talk to the pod directly
	podIP := pod.Status.PodIP
	if len(podIP) == 0 {
		return errors.New("Pod has no IP")
	}

	// tell fetcher to get the function.
	fetcherUrl := fmt.Sprintf("http://%v:8000/", podIP)
	functionUrl := fmt.Sprintf("%v/v1/functions/%v?uid=%v&raw=1",
		gp.controllerUrl, metadata.Name, metadata.Uid)
	fetcherRequest := fmt.Sprintf("{\"url\": \"%v\", \"filename\": \"user\"}", functionUrl)

	log.Printf("[%v] calling fetcher to copy function", metadata)
	resp, err := http.Post(fetcherUrl, "application/json", bytes.NewReader([]byte(fetcherRequest)))
	if err != nil {
		// TODO we should retry this call in case fetcher hasn't come up yet
		return err
	}
	defer resp.Body.Close()
	if resp.StatusCode != 200 {
		return errors.New(fmt.Sprintf("Error from fetcher: %v", resp.Status))
	}

	// get function run container to specialize
	log.Printf("[%v] specializing pod", metadata)
	specializeUrl := fmt.Sprintf("http://%v:8888/specialize", podIP)

	// retry the specialize call a few times in case the env server hasn't come up yet
	maxRetries := 20
	for i := 0; i < maxRetries; i++ {
		resp2, err := http.Post(specializeUrl, "text/plain", bytes.NewReader([]byte{}))
		if err == nil && resp2.StatusCode < 300 {
			// Success
			resp2.Body.Close()
			return nil
		}

		// Only retry for the specific case of a connection error.
		if urlErr, ok := err.(*url.Error); ok {
			if netErr, ok := urlErr.Err.(*net.OpError); ok {
				if netErr.Op == "dial" {
					if i < maxRetries-1 {
						time.Sleep(500 * time.Duration(2*i) * time.Millisecond)
						log.Printf("Error connecting to pod (%v), retrying", netErr)
						continue
					}
				}
			}
		}

		if err == nil {
			err = fission.MakeErrorFromHTTP(resp2)
		}
		log.Printf("Failed to specialize pod: %v", err)
		return err
	}

	return nil
}

// A pool is a deployment of generic containers for an env.  This
// creates the pool but doesn't wait for any pods to be ready.
func (gp *GenericPool) createPool() error {
	poolDeploymentName := fmt.Sprintf("%v-%v-%v",
		gp.env.Metadata.Name, gp.env.Metadata.Uid, strings.ToLower(gp.poolInstanceId))

	podLabels := map[string]string{
		"pool": poolDeploymentName,
		POOLMGR_INSTANCEID_LABEL: gp.instanceId,
	}

	sharedMountPath := "/userfunc"
	deployment := &v1beta1.Deployment{
		ObjectMeta: v1.ObjectMeta{
			Name: poolDeploymentName,
			Labels: map[string]string{
				"environmentName":        gp.env.Metadata.Name,
				"environmentUid":         gp.env.Metadata.Uid,
				POOLMGR_INSTANCEID_LABEL: gp.instanceId,
			},
		},
		Spec: v1beta1.DeploymentSpec{
			Replicas: &gp.replicas,
			Selector: &v1beta1.LabelSelector{
				MatchLabels: podLabels,
			},
			Template: v1.PodTemplateSpec{
				ObjectMeta: v1.ObjectMeta{
					Labels: podLabels,
				},
				Spec: v1.PodSpec{
					Volumes: []v1.Volume{
						{
							Name: "userfunc",
							VolumeSource: v1.VolumeSource{
								EmptyDir: &v1.EmptyDirVolumeSource{},
							},
						},
					},
					Containers: []v1.Container{
						{
							Name:                   gp.env.Metadata.Name,
							Image:                  gp.env.RunContainerImageUrl,
							ImagePullPolicy:        v1.PullIfNotPresent,
							TerminationMessagePath: "/dev/termination-log",
							VolumeMounts: []v1.VolumeMount{
								{
									Name:      "userfunc",
									MountPath: sharedMountPath,
								},
							},
						},
						{
							Name:                   "fetcher",
							Image:                  "fission/fetcher",
							ImagePullPolicy:        v1.PullIfNotPresent,
							TerminationMessagePath: "/dev/termination-log",
							VolumeMounts: []v1.VolumeMount{
								{
									Name:      "userfunc",
									MountPath: sharedMountPath,
								},
							},
							Command: []string{"/fetcher", sharedMountPath},
						},
					},
				},
			},
		},
	}
	depl, err := gp.kubernetesClient.Extensions().Deployments(gp.namespace).Create(deployment)
	if err != nil {
		return err
	}
	gp.deployment = depl
	return nil
}

func (gp *GenericPool) waitForReadyPod() error {
	startTime := time.Now()
	for {
		// TODO: for now we just poll; use a watch instead
		depl, err := gp.kubernetesClient.Extensions().Deployments(gp.namespace).Get(gp.deployment.ObjectMeta.Name)
		if err != nil {
			log.Printf("err: %v", err)
			return err
		}
		gp.deployment = depl
		if gp.deployment.Status.AvailableReplicas > 0 {
			return nil
		}

		if time.Now().Sub(startTime) > gp.podReadyTimeout {
			return errors.New("timeout: waited too long for pod to be ready")
		}
		time.Sleep(1000 * time.Millisecond)
	}
}

func (gp *GenericPool) createSvc(name string, labels map[string]string) (*v1.Service, error) {
	service := v1.Service{
		ObjectMeta: v1.ObjectMeta{
			Name: name,
		},
		Spec: v1.ServiceSpec{
			Type: v1.ServiceTypeClusterIP,
			Ports: []v1.ServicePort{
				{
					Protocol:   v1.ProtocolTCP,
					Port:       80,
					TargetPort: intstr.FromInt(8888),
				},
			},
			Selector: labels,
		},
	}
	svc, err := gp.kubernetesClient.Core().Services(gp.namespace).Create(&service)
	return svc, err
}

func (gp *GenericPool) GetFuncSvc(m *fission.Metadata) (*funcSvc, error) {

	log.Printf("[%v] Choosing pod from pool", m)
	newLabels := gp.labelsForMetadata(m)
	pod, err := gp.choosePod(newLabels)
	if err != nil {
		return nil, err
	}

	err = gp.specializePod(pod, m)
	if err != nil {
		gp.scheduleDeletePod(pod.ObjectMeta.Name)
		return nil, err
	}
	log.Printf("Specialized pod: %v", pod.ObjectMeta.Name)

	var svcHost string
	if gp.useSvc {
		svcName := fmt.Sprintf("svc-%v", m.Name)
		if len(m.Uid) > 0 {
			svcName += ("-" + m.Uid)
		}

		labels := gp.labelsForMetadata(m)
		svc, err := gp.createSvc(svcName, labels)
		if err != nil {
			gp.scheduleDeletePod(pod.ObjectMeta.Name)
			return nil, err
		}
		if svc.ObjectMeta.Name != svcName {
			gp.scheduleDeletePod(pod.ObjectMeta.Name)
			return nil, errors.New(fmt.Sprintf("sanity check failed for svc %v", svc.ObjectMeta.Name))
		}

		// the fission router isn't in the same namespace, so return a
		// namespace-qualified hostname
		svcHost = fmt.Sprintf("%v.%v", svcName, gp.namespace)
	} else {
		log.Printf("Using pod IP for specialized pod")
		svcHost = fmt.Sprintf("%v:8888", pod.Status.PodIP)
	}

	fsvc := &funcSvc{
		function:    m,
		environment: gp.env,
		address:     svcHost,
		podName:     pod.ObjectMeta.Name,
		ctime:       time.Now(),
		atime:       time.Now(),
	}

	err, existingFsvc := gp.fsCache.Add(*fsvc)
	if err != nil {
		// Some other thread beat us to it -- return the other thread's fsvc and clean up
		// our own.  TODO: this is grossly inefficient, improve it with some sort of state
		// machine
		log.Printf("func svc already exists: %v", existingFsvc.podName)
		go func() {
			gp.kubernetesClient.Core().Pods(gp.namespace).Delete(fsvc.podName, nil)
		}()
		return existingFsvc, nil
	}
	return fsvc, nil
}

func (gp *GenericPool) CleanupFunctionService(podName string) error {
	// remove ourselves from fsCache (only if we're still old)
	deleted, err := gp.fsCache.DeleteByPod(podName, gp.idlePodReapTime)
	if err != nil {
		return err
	}

	if !deleted {
		log.Printf("Not deleting %v, in use", podName)
		return nil
	}

	// delete pod
	err = gp.kubernetesClient.Core().Pods(gp.namespace).Delete(podName, nil)
	if err != nil {
		return err
	}

	return nil
}

func (gp *GenericPool) idlePodReaper() {
	for {
		time.Sleep(time.Minute)
		podNames, err := gp.fsCache.ListOld(gp.idlePodReapTime)
		if err != nil {
			log.Printf("Error reaping idle pods: %v", err)
			continue
		}
		for _, podName := range podNames {
			log.Printf("Reaping idle pod '%v'", podName)
			err := gp.CleanupFunctionService(podName)
			if err != nil {
				log.Printf("Error deleting idle pod '%v': %v", podName, err)
			}
		}
	}
}