arouterd/arouterd.go

/*
 * Copyright 2018-present Open Networking Foundation

 * 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 main

import (
	"time"
	"regexp"
	"errors"
	"strconv"

	"k8s.io/client-go/rest"
	"google.golang.org/grpc"
	"golang.org/x/net/context"
	"k8s.io/client-go/kubernetes"
	"github.com/golang/protobuf/ptypes"
	"github.com/opencord/voltha-go/common/log"
	kafka "github.com/opencord/voltha-go/kafka"
	metav1 "k8s.io/apimachinery/pkg/apis/meta/v1"
	empty "github.com/golang/protobuf/ptypes/empty"
	vpb "github.com/opencord/voltha-go/protos/voltha"
	cmn "github.com/opencord/voltha-go/protos/common"
	pb "github.com/opencord/voltha-go/protos/afrouter"
	ic "github.com/opencord/voltha-go/protos/inter_container"
)

type configConn struct {
	Server string `json:"Server"`
	Cluster string `json:"Cluster"`
	Backend string `json:"Backend"`
	connections map[string]connection
}

type connection struct {
	Name string `json:"Connection"`
	Addr string `json:"Addr"`
	Port uint64 `json:"Port"`
}

type rwPod struct {
	name string
	ipAddr string
	node string
	devIds map[string]struct{}
	backend string
	connection string
}

type podTrack struct {
		pod *rwPod
		dn bool
}

var nPods int = 6

// Topic is affinityRouter
// port: 9092

func newKafkaClient(clientType string, host string, port int, instanceID string) (kafka.Client, error) {

	log.Infow("kafka-client-type", log.Fields{"client": clientType})
	switch clientType {
	case "sarama":
		return kafka.NewSaramaClient(
			kafka.Host(host),
			kafka.Port(port),
			kafka.ConsumerType(kafka.GroupCustomer),
			kafka.ProducerReturnOnErrors(true),
			kafka.ProducerReturnOnSuccess(true),
			kafka.ProducerMaxRetries(6),
			kafka.NumPartitions(3),
			kafka.ConsumerGroupName(instanceID),
			kafka.ConsumerGroupPrefix(instanceID),
			kafka.AutoCreateTopic(false),
			kafka.ProducerFlushFrequency(5),
			kafka.ProducerRetryBackoff(time.Millisecond*30)), nil
	}
	return nil, errors.New("unsupported-client-type")
}


func k8sClientSet() *kubernetes.Clientset {
	// creates the in-cluster config
	config, err := rest.InClusterConfig()
	if err != nil {
		panic(err.Error())
	}
	// creates the clientset
	clientset, err := kubernetes.NewForConfig(config)
	if err != nil {
		panic(err.Error())
	}

	return clientset
}


func connect(addr string) (*grpc.ClientConn, error) {
	for ctr :=0 ; ctr < 100; ctr++ {
		log.Debugf("Trying to connect to %s", addr)
		conn, err := grpc.Dial(addr, grpc.WithInsecure())
		if err != nil {
			log.Debugf("Attempt to connect failed, retrying %v:", err)
		} else {
			log.Debugf("Connection succeeded")
			return conn,err
		}
		time.Sleep(10 * time.Second)
	}
	log.Debugf("Too many connection attempts, giving up!")
	return nil,errors.New("Timeout attempting to conect")
}

func getRwPods(cs *kubernetes.Clientset, coreFilter * regexp.Regexp) []*rwPod {
	var rtrn []*rwPod

	pods, err := cs.CoreV1().Pods("").List(metav1.ListOptions{})
	if err != nil {
		panic(err.Error())
	}
	//log.Debugf("There are a total of %d pods in the cluster\n", len(pods.Items))

	for _,v := range pods.Items {
		if v.Namespace == "voltha" && coreFilter.MatchString(v.Name) {
			log.Debugf("Namespace: %s, PodName: %s, PodIP: %s, Host: %s\n", v.Namespace, v.Name,
																	v.Status.PodIP, v.Spec.NodeName)
			// Only add the pod if it has an IP address. If it doesn't then it likely crashed and
			// and is still in the process of getting re-started.
			if v.Status.PodIP != "" {
				rtrn = append(rtrn, &rwPod{name:v.Name,ipAddr:v.Status.PodIP,node:v.Spec.NodeName,
							devIds:make(map[string]struct{}), backend:"", connection:""})
			}
		}
	}
	return rtrn
}

func reconcilePodDeviceIds(pod * rwPod, ids map[string]struct{}) bool {
	var idList cmn.IDs
	for k,_ := range ids  {
		 idList.Items = append(idList.Items, &cmn.ID{Id:k})
	}
	conn,err := connect(pod.ipAddr+":50057")
	defer conn.Close()
	if err != nil {
		log.Debugf("Could not query devices from %s, could not connect", pod.name)
		return false
	}
	client := vpb.NewVolthaServiceClient(conn)
	_,err = client.ReconcileDevices(context.Background(), &idList)
	if err != nil {
		log.Error(err)
		return false
	}

	return true
}

func queryPodDeviceIds(pod * rwPod) map[string]struct{} {
	var rtrn map[string]struct{} = make(map[string]struct{})
	// Open a connection to the pod
	// port 50057
	conn, err := connect(pod.ipAddr+":50057")
	if err != nil {
		log.Debugf("Could not query devices from %s, could not connect", pod.name)
		return rtrn
	}
	defer conn.Close()
	client := vpb.NewVolthaServiceClient(conn)
	devs,err := client.ListDeviceIds(context.Background(), &empty.Empty{})
	if err != nil {
		log.Error(err)
		return rtrn
	}
	for _,dv := range devs.Items {
		rtrn[dv.Id]=struct{}{}
	}

	return rtrn
}

func queryDeviceIds(pods []*rwPod) {
	for pk,_ := range pods {
		// Keep the old Id list if a new list is not returned
		if idList := queryPodDeviceIds(pods[pk]); len(idList) != 0 {
			pods[pk].devIds = idList
		}
	}
}

func allEmpty(pods []*rwPod) bool {
	for k,_ := range pods {
		if len(pods[k].devIds) != 0 {
			return false
		}
	}
	return true
}

func rmPod(pods []*rwPod, idx int) []*rwPod {
	return append(pods[:idx],pods[idx+1:]...)
}

func groupIntersectingPods1(pods []*rwPod, podCt int) ([][]*rwPod,[]*rwPod) {
	var rtrn [][]*rwPod
	var out []*rwPod

	for {
		if len(pods) == 0 {
			break
		}
		if len(pods[0].devIds) == 0 { // Ignore pods with no devices
			////log.Debugf("%s empty pod", pd[k].pod.name)
			out = append(out, pods[0])
			pods = rmPod(pods, 0)
			continue
		}
		// Start a pod group with this pod
		var grp []*rwPod
		grp = append(grp, pods[0])
		pods = rmPod(pods,0)
		//log.Debugf("Creating new group %s", pd[k].pod.name)
		// Find the peer pod based on device overlap
		// It's ok if one isn't found, an empty one will be used instead
		for k,_ := range pods {
			if len(pods[k].devIds) == 0 { // Skip pods with no devices
				//log.Debugf("%s empty pod", pd[k1].pod.name)
				continue
			}
			if intersect(grp[0].devIds, pods[k].devIds) == true {
				//log.Debugf("intersection found %s:%s", pd[k].pod.name, pd[k1].pod.name)
				if grp[0].node == pods[k].node {
					// This should never happen
					log.Errorf("pods %s and %s intersect and are on the same server!! Not pairing",
								grp[0].name, pods[k].name)
					continue
				}
				grp = append(grp, pods[k])
				pods = rmPod(pods, k)
				break

			}
		}
		rtrn = append(rtrn, grp)
		//log.Debugf("Added group %s", grp[0].name)
		// Check if the number of groups = half the pods, if so all groups are started.
		if len(rtrn) == podCt >> 1 {
			// Append any remaining pods to out
			out = append(out,pods[0:]...)
			break
		}
	}
	return rtrn,out
}

func unallocPodCount(pd []*podTrack) int {
	var rtrn int = 0
	for _,v := range pd {
		if v.dn == false {
			rtrn++
		}
	}
	return rtrn
}


func sameNode(pod *rwPod, grps [][]*rwPod) bool {
	for _,v := range grps {
		if v[0].node == pod.node {
			return true
		}
		if len(v) == 2 && v[1].node == pod.node {
			return true
		}
	}
	return false
}

func startRemainingGroups1(grps [][]*rwPod, pods []*rwPod, podCt int) ([][]*rwPod, []*rwPod) {
	var grp []*rwPod

	for k,_ := range pods {
		if sameNode(pods[k], grps) {
			continue
		}
		grp = []*rwPod{}
		grp = append(grp, pods[k])
		pods = rmPod(pods, k)
		grps = append(grps, grp)
		if len(grps) == podCt >> 1 {
			break
		}
	}
	return grps, pods
}

func hasSingleSecondNode(grp []*rwPod) bool {
	var srvrs map[string]struct{} = make(map[string]struct{})
	for k,_ := range grp {
		if k == 0 {
			continue // Ignore the first item
		}
		srvrs[grp[k].node] = struct{}{}
	}
	if len(srvrs) == 1 {
		return true
	}
	return false
}

func addNode(grps [][]*rwPod, idx *rwPod, item *rwPod) [][]*rwPod {
	for k,_ := range grps {
		if grps[k][0].name == idx.name {
			grps[k] = append(grps[k], item)
			return grps
		}
	}
	// TODO: Error checking required here.
	return grps
}

func removeNode(grps [][]*rwPod, item *rwPod) [][]*rwPod {
	for k,_ := range grps {
		for k1,_ := range grps[k] {
			if grps[k][k1].name == item.name {
				grps[k] = append(grps[k][:k1],grps[k][k1+1:]...)
				break
			}
		}
	}
	return grps
}

func groupRemainingPods1(grps [][]*rwPod, pods []*rwPod) [][]*rwPod {
	var lgrps [][]*rwPod
	// All groups must be started when this function is called.
	// Copy incomplete groups
	for k,_ := range grps {
		if len(grps[k]) != 2 {
			lgrps = append(lgrps, grps[k])
		}
	}

	// Add all pairing candidates to each started group.
	for k,_ := range pods {
		for k2,_ := range lgrps {
			if lgrps[k2][0].node != pods[k].node {
				lgrps[k2] = append(lgrps[k2], pods[k])
			}
		}
	}

	//TODO: If any member of lgrps doesn't have at least 2
	// nodes something is wrong. Check for that here

	for {
		for { // Address groups with only a single server choice
			var ssn bool = false

			for k,_ := range lgrps {
				// Now if any of the groups only have a single
				// node as the choice for the second member
				// address that one first.
				if hasSingleSecondNode(lgrps[k]) == true {
					ssn =  true
					// Add this pairing to the groups
					grps = addNode(grps, lgrps[k][0], lgrps[k][1])
					// Since this node is now used, remove it from all
					// remaining tenative groups
					lgrps = removeNode(lgrps, lgrps[k][1])
					// Now remove this group completely since
					// it's been addressed
					lgrps  = append(lgrps[:k],lgrps[k+1:]...)
					break
				}
			}
			if ssn == false {
				break
			}
		}
		// Now adress one of the remaining groups
		if len(lgrps) == 0 {
			break // Nothing left to do, exit the loop
		}
		grps = addNode(grps, lgrps[0][0], lgrps[0][1])
		lgrps = removeNode(lgrps, lgrps[0][1])
		lgrps  = append(lgrps[:0],lgrps[1:]...)
	}
	return grps
}

func groupPods1(pods []*rwPod) [][]*rwPod {
	var rtrn [][]*rwPod
	var podCt int = len(pods)

	rtrn,pods = groupIntersectingPods1(pods, podCt)
	// There are several outcomes here 
	// 1) All pods have been paired and we're done
	// 2) Some un-allocated pods remain
	// 2.a) All groups have been started
	// 2.b) Not all groups have been started
	if len(pods) == 0 {
		return rtrn
	} else if len(rtrn) == podCt >> 1 { // All groupings started
		// Allocate the remaining (presumably empty) pods to the started groups
		return groupRemainingPods1(rtrn, pods)
	} else { // Some groupings started
		// Start empty groups with remaining pods
		// each grouping is on a different server then
		// allocate remaining pods.
		rtrn, pods = startRemainingGroups1(rtrn, pods, podCt)
		return groupRemainingPods1(rtrn, pods)
	}
}

func intersect(d1 map[string]struct{}, d2 map[string]struct{}) bool {
	for k,_ := range d1 {
		if _,ok := d2[k]; ok == true {
			return true
		}
	}
	return false
}

func setConnection(client pb.ConfigurationClient, backend string, connection string, addr string, port uint64) {
	log.Debugf("Configuring backend %s : connection %s\n\n", backend, connection)
	cnf := &pb.Conn{Server:"grpc_command",Cluster:"vcore",Backend:backend,
					Connection:connection,Addr:addr,
					Port:port}
	if res, err := client.SetConnection(context.Background(), cnf); err != nil {
		log.Debugf("failed SetConnection RPC call: %s", err)
	} else {
		log.Debugf("Result: %v", res)
	}
}

func setAffinity(client pb.ConfigurationClient, ids map[string]struct{}, backend string) {
	log.Debugf("Configuring backend %s : affinities \n", backend)
	aff := &pb.Affinity{Router:"vcore",Route:"dev_manager",Cluster:"vcore",Backend:backend}
	for k,_ := range ids {
		log.Debugf("Setting affinity for id %s", k)
		aff.Id = k
		if res, err := client.SetAffinity(context.Background(), aff); err != nil {
			log.Debugf("failed affinity RPC call: %s", err)
		} else {
			log.Debugf("Result: %v", res)
		}
	}
}

func getBackendForCore(coreId string, coreGroups [][]*rwPod) string {
	for _,v := range coreGroups {
		for _,v2 := range v {
			if v2.name == coreId {
				return v2.backend
			}
		}
	}
	log.Errorf("No backend found for core %s\n", coreId)
	return ""
}

func monitorDiscovery(client pb.ConfigurationClient,
						ch <-chan *ic.InterContainerMessage,
						coreGroups [][]*rwPod) {
	var id map[string]struct{} = make(map[string]struct{})

	select {
	case msg := <-ch:
		log.Debugf("Received a device discovery notification")
		device := &ic.DeviceDiscovered{}
		if err := ptypes.UnmarshalAny(msg.Body, device); err != nil {
			log.Errorf("Could not unmarshal received notification %v", msg)
		} else {
			// Set the affinity of the discovered device.
			if be := getBackendForCore(device.Id, coreGroups); be != "" {
				id[device.Id]=struct{}{}
				setAffinity(client, id, be)
			} else {
				log.Error("Cant use an empty string as a backend name")
			}
		}
		break
	}
}

func startDiscoveryMonitor(client pb.ConfigurationClient,
							coreGroups [][]*rwPod) error {
	var ch <-chan *ic.InterContainerMessage
	// Connect to kafka for discovery events
	topic := &kafka.Topic{Name: "AffinityRouter"}
	kc,err := newKafkaClient("sarama", "kafka", 9092, "arouterd")
	kc.Start()

	if ch, err = kc.Subscribe(topic); err != nil {
		log.Error("Could not subscribe to the 'AffinityRouter' channel, discovery disabled")
		return err
	}
	go monitorDiscovery(client, ch, coreGroups)
	return nil
}

// Determines which items in core groups
// have changed based on the list provided
// and returns a coreGroup with only the changed
// items and a pod list with the new items
func getAddrDiffs(coreGroups [][]*rwPod, rwPods []*rwPod) ([][]*rwPod, []*rwPod) {
	var nList []*rwPod
	var rtrn [][]*rwPod = make([][]*rwPod, nPods>>1)
	var ipAddrs map[string]struct{} = make(map[string]struct{})

	log.Debug("Get addr diffs")

	// Start with an empty array
	for k,_ := range rtrn {
		rtrn[k] = make([]*rwPod, 2)
	}

	// Build a list with only the new items
	for _,v := range rwPods {
		if hasIpAddr(coreGroups, v.ipAddr) == false {
			nList = append(nList, v)
		}
		ipAddrs[v.ipAddr] = struct{}{} // for the search below
	}

	// Now build the coreGroups with only the changed items
	for k1,v1 := range coreGroups {
		for k2,v2 := range v1 {
			if _,ok := ipAddrs[v2.ipAddr]; ok == false {
				rtrn[k1][k2] = v2
			}
		}
	}
	return rtrn, nList
}

// Figure out where best to put the new pods
// in the coreGroup array based on the old
// pods being replaced. The criteria is that
// the new pod be on the same server as the
// old pod was.
func reconcileAddrDiffs(coreGroupDiffs [][]*rwPod, rwPodDiffs []*rwPod) ([][]*rwPod) {
	var srvrs map[string][]*rwPod = make(map[string][]*rwPod)

	log.Debug("Reconciling diffs")
	log.Debug("Building server list")
	for _,v := range rwPodDiffs {
		log.Debugf("Adding %v to the server list", *v)
		srvrs[v.node] = append(srvrs[v.node], v)
	}

	for k1,v1 := range coreGroupDiffs {
		log.Debugf("k1:%v, v1:%v", k1,v1)
		for k2,v2 :=  range v1 {
			log.Debugf("k2:%v, v2:%v", k2,v2)
			if v2 == nil { // Nothing to do here
				continue
			}
			if _,ok := srvrs[v2.node]; ok == true {
				coreGroupDiffs[k1][k2] = srvrs[v2.node][0]
				if len(srvrs[v2.node]) > 1 { // remove one entry from the list
					srvrs[v2.node] = append(srvrs[v2.node][:0], srvrs[v2.node][1:]...)
				} else { // Delete the endtry from the map
					delete(srvrs, v2.node)
				}
			} else {
				log.Error("This should never happen, node appears to have changed names")
				// attempt to limp along by keeping this old entry
			}
		}
	}

	return coreGroupDiffs
}

func applyAddrDiffs(client pb.ConfigurationClient, coreGroups [][]*rwPod, rwPods []*rwPod) {
	var newEntries [][]*rwPod

	log.Debug("Applying diffs")
	newEntries = reconcileAddrDiffs(getAddrDiffs(coreGroups, rwPods))

	// Now replace the information in coreGropus with the new
	// entries and then reconcile the device ids on the core
	// that's in the new entry with the device ids of it's
	// active-active peer.
	for k1,v1 := range coreGroups {
		for k2,v2 := range v1 {
			if newEntries[k1][k2] != nil {
				// TODO: Missing is the case where bothe the primary
				// and the secondary core crash and come back.
				// Pull the device ids from the active-active peer
				ids := queryPodDeviceIds(coreGroups[k1][k2^1])
				if len(ids) != 0 {
					if reconcilePodDeviceIds(newEntries[k1][k2], ids) == false {
						log.Errorf("Attempt to reconcile ids on pod %v failed",newEntries[k1][k2])
					}
				}
				// Send the affininty router new connection information
				setConnection(client, v2.backend, v2.connection, newEntries[k1][k2].ipAddr, 50057)
				// Copy the new entry information over
				coreGroups[k1][k2].ipAddr = newEntries[k1][k2].ipAddr
				coreGroups[k1][k2].name = newEntries[k1][k2].name
				coreGroups[k1][k2].devIds = ids
			}
		}
	}
}

func updateDeviceIds(coreGroups [][]*rwPod, rwPods []*rwPod) {
	var byName map[string]*rwPod = make(map[string]*rwPod)

	// Convinience
	for _,v := range rwPods {
		byName[v.name] = v
	}

	for k1,v1 := range coreGroups {
		for k2,_ := range v1 {
			coreGroups[k1][k2].devIds = byName[v1[k2].name].devIds
		}
	}
}

func startCoreMonitor(client pb.ConfigurationClient,
					clientset *kubernetes.Clientset,
					coreFltr *regexp.Regexp,
					coreGroups [][]*rwPod) error {
	// Now that initial allocation has been completed, monitor the pods
	// for IP changes
	// The main loop needs to do the following:
	// 1) Periodically query the pods and filter out
	//    the vcore ones
	// 2) Validate that the pods running are the same
	//    as the previous check
	// 3) Validate that the IP addresses are the same
	//    as the last check.
	// If the pod name(s) ha(s/ve) changed then remove
	// the unused pod names and add in the new pod names
	// maintaining the cluster/backend information.
	// If an IP address has changed (which shouldn't
	// happen unless a pod is re-started) it should get
	// caught by the pod name change.
	for {
		time.Sleep(10 * time.Second) // Wait a while
		// Get the rw core list from k8s
		rwPods := getRwPods(clientset, coreFltr)
		queryDeviceIds(rwPods)
		updateDeviceIds(coreGroups, rwPods)
		// If we didn't get 2n+1 pods then wait since
		// something is down and will hopefully come
		// back up at some point.
		// TODO: remove the 6 pod hardcoding
		if len(rwPods) != 6 {
			continue
		}
		// We have all pods, check if any IP addresses
		// have changed.
		for _,v := range rwPods {
			if hasIpAddr(coreGroups, v.ipAddr) == false {
				log.Debug("Address has changed...")
				applyAddrDiffs(client, coreGroups, rwPods)

			}
		}
	}
}

func hasIpAddr(coreGroups [][]*rwPod, ipAddr string) bool {
	for _,v1 := range coreGroups {
		for _,v2 := range v1 {
			if v2.ipAddr == ipAddr {
				return true
			}
		}
	}
	return false
}


func main() {
	// This is currently hard coded to a cluster with 3 servers
	//var connections map[string]configConn = make(map[string]configConn)
	//var rwCorePodsPrev map[string]rwPod = make(map[string]rwPod)
	var err error
	var conn *grpc.ClientConn


	// Set up the regular expression to identify the voltha cores
	coreFltr := regexp.MustCompile(`rw-core[0-9]-`)

	// Set up logging
	if _, err := log.SetDefaultLogger(log.JSON, 0, nil); err != nil {
		log.With(log.Fields{"error": err}).Fatal("Cannot setup logging")
	}

	// Set up kubernetes api
	clientset := k8sClientSet()

	// Connect to the affinity router and set up the client
	conn, err = connect("localhost:55554") // This is a sidecar container so communicating over localhost
	defer conn.Close()
	if err != nil {
		panic(err.Error())
	}
	client := pb.NewConfigurationClient(conn)

	// Get the voltha rw-core podes
	rwPods := getRwPods(clientset, coreFltr)

	// Fetch the devices held by each running core
	queryDeviceIds(rwPods)

	// For debugging... comment out l8r
	for _,v := range rwPods {
		log.Debugf("Pod list %v", *v)
	}

	coreGroups := groupPods1(rwPods)

	// Assign the groupings to the the backends and connections
	for k,_ := range coreGroups {
		for k1,_ := range coreGroups[k] {
			coreGroups[k][k1].backend = "vcore"+strconv.Itoa(k+1)
			coreGroups[k][k1].connection = "vcore"+strconv.Itoa(k+1)+strconv.Itoa(k1+1)
		}
	}
	log.Info("Core gouping completed")

	// TODO: Debugging code, comment out for production
	for k,v := range coreGroups {
		for k2,v2 := range v {
			log.Debugf("Core group %d,%d: %v", k, k2, v2)
		}
	}
	log.Info("Setting affinities")
	// Now set the affinities for exising devices in the cores
	for _,v := range coreGroups {
		setAffinity(client, v[0].devIds, v[0].backend)
		setAffinity(client, v[1].devIds, v[1].backend)
	}
	log.Info("Setting connections")
	// Configure the backeds based on the calculated core groups
	for _,v := range coreGroups {
		setConnection(client, v[0].backend, v[0].connection, v[0].ipAddr, 50057)
		setConnection(client, v[1].backend, v[1].connection, v[1].ipAddr, 50057)
	}

	log.Info("Starting discovery monitoring")
	startDiscoveryMonitor(client, coreGroups)

	log.Info("Starting core monitoring")
	startCoreMonitor(client, clientset, coreFltr, coreGroups) // Never returns
	return
}