etcd-io__etcd/etcd/participant.go

/*
Copyright 2014 CoreOS Inc.

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 etcd

import (
	"encoding/json"
	"fmt"
	"log"
	"math/rand"
	"net/http"
	"os"
	"path"
	"time"

	"github.com/coreos/etcd/conf"
	etcdErr "github.com/coreos/etcd/error"
	"github.com/coreos/etcd/raft"
	"github.com/coreos/etcd/snap"
	"github.com/coreos/etcd/store"
	"github.com/coreos/etcd/wal"
)

const (
	defaultHeartbeat = 1
	defaultElection  = 5
	defaultCompact   = 10000

	maxBufferedProposal = 128

	defaultTickDuration = time.Millisecond * 100

	v2machineKVPrefix = "/_etcd/machines"
	v2configKVPrefix  = "/_etcd/config"

	v2Prefix              = "/v2/keys"
	v2machinePrefix       = "/v2/machines"
	v2peersPrefix         = "/v2/peers"
	v2LeaderPrefix        = "/v2/leader"
	v2SelfStatsPrefix     = "/v2/stats/self"
	v2LeaderStatsPrefix   = "/v2/stats/leader"
	v2StoreStatsPrefix    = "/v2/stats/store"
	v2adminConfigPrefix   = "/v2/admin/config"
	v2adminMachinesPrefix = "/v2/admin/machines/"
)

var (
	tmpErr      = fmt.Errorf("try again")
	stopErr     = fmt.Errorf("server is stopped")
	raftStopErr = fmt.Errorf("raft is stopped")
)

type participant struct {
	id           int64
	clusterId    int64
	cfg          *conf.Config
	pubAddr      string
	raftPubAddr  string
	tickDuration time.Duration

	client  *v2client
	peerHub *peerHub

	proposal    chan v2Proposal
	addNodeC    chan raft.Config
	removeNodeC chan raft.Config
	node        *v2Raft
	store.Store
	rh          *raftHandler
	w           *wal.WAL
	snapshotter *snap.Snapshotter
	serverStats *raftServerStats

	stopNotifyc chan struct{}

	*http.ServeMux
}

func newParticipant(c *conf.Config, client *v2client, peerHub *peerHub, tickDuration time.Duration) (*participant, error) {
	p := &participant{
		clusterId:    -1,
		cfg:          c,
		tickDuration: tickDuration,

		client:  client,
		peerHub: peerHub,

		proposal:    make(chan v2Proposal, maxBufferedProposal),
		addNodeC:    make(chan raft.Config, 1),
		removeNodeC: make(chan raft.Config, 1),
		node: &v2Raft{
			result: make(map[wait]chan interface{}),
		},
		Store:       store.New(),
		serverStats: NewRaftServerStats(c.Name),

		stopNotifyc: make(chan struct{}),

		ServeMux: http.NewServeMux(),
	}
	p.rh = newRaftHandler(peerHub, p.Store.Version(), p.serverStats)
	p.peerHub.setServerStats(p.serverStats)

	snapDir := path.Join(p.cfg.DataDir, "snap")
	if err := os.Mkdir(snapDir, 0700); err != nil {
		if !os.IsExist(err) {
			log.Printf("id=%x participant.new.mkdir err=%v", p.id, err)
			return nil, err
		}
	}
	p.snapshotter = snap.New(snapDir)

	walDir := path.Join(p.cfg.DataDir, "wal")
	if err := os.Mkdir(walDir, 0700); err != nil {
		if !os.IsExist(err) {
			log.Printf("id=%x participant.new.mkdir err=%v", p.id, err)
			return nil, err
		}
	}

	var w *wal.WAL
	var err error
	if !wal.Exist(walDir) {
		p.id = genId()
		p.pubAddr = c.Addr
		p.raftPubAddr = c.Peer.Addr
		if w, err = wal.Create(walDir); err != nil {
			return nil, err
		}
		p.node.Node = raft.New(p.id, defaultHeartbeat, defaultElection)
		info := p.node.Info()
		if err = w.SaveInfo(&info); err != nil {
			return nil, err
		}
		log.Printf("id=%x participant.new path=%s\n", p.id, walDir)
	} else {
		s, err := p.snapshotter.Load()
		if err != nil && err != snap.ErrNoSnapshot {
			log.Printf("id=%x participant.snapload err=%s\n", p.id, err)
			return nil, err
		}
		var snapIndex int64
		if s != nil {
			if err := p.Recovery(s.Data); err != nil {
				panic(err)
			}
			log.Printf("id=%x participant.store.recovered index=%d\n", p.id, s.Index)
			snapIndex = s.Index
		}
		n, err := wal.Read(walDir, snapIndex)
		if err != nil {
			return nil, err
		}
		p.id = n.Id
		p.node.Node = raft.Recover(n.Id, s, n.Ents, n.State, defaultHeartbeat, defaultElection)
		p.apply(p.node.Next())
		log.Printf("id=%x participant.load path=%s snapIndex=%d state=\"%+v\" len(ents)=%d", p.id, p.cfg.DataDir, snapIndex, n.State, len(n.Ents))
		if w, err = wal.Open(walDir); err != nil {
			return nil, err
		}
	}
	p.w = w

	p.Handle(v2Prefix+"/", handlerErr(p.serveValue))
	p.Handle(v2machinePrefix, handlerErr(p.serveMachines))
	p.Handle(v2peersPrefix, handlerErr(p.serveMachines))
	p.Handle(v2LeaderPrefix, handlerErr(p.serveLeader))
	p.Handle(v2SelfStatsPrefix, handlerErr(p.serveSelfStats))
	p.Handle(v2LeaderStatsPrefix, handlerErr(p.serveLeaderStats))
	p.Handle(v2StoreStatsPrefix, handlerErr(p.serveStoreStats))
	p.rh.Handle(v2adminConfigPrefix, handlerErr(p.serveAdminConfig))
	p.rh.Handle(v2adminMachinesPrefix, handlerErr(p.serveAdminMachines))

	// TODO: remind to set application/json for /v2/stats endpoint

	return p, nil
}

func (p *participant) run(stop chan struct{}) {
	defer p.cleanup()

	if p.node.IsEmpty() {
		seeds := p.peerHub.getSeeds()
		if len(seeds) == 0 {
			log.Printf("id=%x participant.run action=bootstrap\n", p.id)
			p.node.Campaign()
			p.node.InitCluster(genId())
			p.node.Add(p.id, p.raftPubAddr, []byte(p.pubAddr))
			p.apply(p.node.Next())
		} else {
			log.Printf("id=%x participant.run action=join seeds=\"%v\"\n", p.id, seeds)
			if err := p.join(); err != nil {
				log.Fatalf("id=%x participant.join err=%q", p.id, err)
			}
		}
	}

	p.rh.start()
	defer p.rh.stop()

	node := p.node
	recv := p.rh.recv

	ticker := time.NewTicker(p.tickDuration)
	defer ticker.Stop()
	v2SyncTicker := time.NewTicker(time.Millisecond * 500)
	defer v2SyncTicker.Stop()

	var proposal chan v2Proposal
	var addNodeC, removeNodeC chan raft.Config
	for {
		if node.HasLeader() {
			proposal = p.proposal
			addNodeC = p.addNodeC
			removeNodeC = p.removeNodeC
		} else {
			proposal = nil
			addNodeC = nil
			removeNodeC = nil
		}
		select {
		case p := <-proposal:
			node.Propose(p)
		case c := <-addNodeC:
			node.UpdateConf(raft.AddNode, &c)
		case c := <-removeNodeC:
			node.UpdateConf(raft.RemoveNode, &c)
		case msg := <-recv:
			node.Step(*msg)
		case <-ticker.C:
			node.Tick()
		case <-v2SyncTicker.C:
			node.Sync()
		case <-stop:
			log.Printf("id=%x participant.stop\n", p.id)
			return
		}
		if s := node.UnstableSnapshot(); !s.IsEmpty() {
			if err := p.Recovery(s.Data); err != nil {
				panic(err)
			}
			log.Printf("id=%x recovered index=%d\n", p.id, s.Index)
		}
		p.apply(node.Next())
		ents := node.UnstableEnts()
		p.save(ents, node.UnstableState())
		p.send(node.Msgs())
		if node.IsRemoved() {
			log.Printf("id=%x participant.end\n", p.id)
			return
		}
		if p.node.EntsLen() > defaultCompact {
			d, err := p.Save()
			if err != nil {
				panic(err)
			}
			p.node.Compact(d)
			snap := p.node.GetSnap()
			log.Printf("id=%x compacted index=%d", p.id, snap.Index)
			if err := p.snapshotter.Save(&snap); err != nil {
				log.Printf("id=%d snapshot err=%v", p.id, err)
				// todo(xiangli): consume the error?
				panic(err)
			}
			if err := p.w.Cut(p.node.Index()); err != nil {
				log.Printf("id=%d wal.cut err=%v", p.id, err)
				// todo(xiangli): consume the error?
				panic(err)
			}
			info := p.node.Info()
			if err = p.w.SaveInfo(&info); err != nil {
				log.Printf("id=%d wal.saveInfo err=%v", p.id, err)
				// todo(xiangli): consume the error?
				panic(err)
			}
		}
	}
}

func (p *participant) cleanup() {
	p.w.Close()
	close(p.stopNotifyc)
	p.peerHub.stop()
}

func (p *participant) raftHandler() http.Handler {
	return p.rh
}

func (p *participant) add(id int64, raftPubAddr string, pubAddr string) error {
	log.Printf("id=%x participant.add nodeId=%x raftPubAddr=%s pubAddr=%s\n", p.id, id, raftPubAddr, pubAddr)
	pp := path.Join(v2machineKVPrefix, fmt.Sprint(id))

	_, err := p.Store.Get(pp, false, false)
	if err == nil {
		return nil
	}
	if v, ok := err.(*etcdErr.Error); !ok || v.ErrorCode != etcdErr.EcodeKeyNotFound {
		log.Printf("id=%x participant.add getErr=\"%v\"\n", p.id, err)
		return err
	}

	w, err := p.Watch(pp, true, false, 0)
	if err != nil {
		log.Printf("id=%x participant.add watchErr=\"%v\"\n", p.id, err)
		return tmpErr
	}

	select {
	case p.addNodeC <- raft.Config{NodeId: id, Addr: raftPubAddr, Context: []byte(pubAddr)}:
	default:
		w.Remove()
		log.Printf("id=%x participant.add proposeErr=\"unable to send out addNode proposal\"\n", p.id)
		return tmpErr
	}

	select {
	case v := <-w.EventChan:
		if v.Action == store.Set {
			return nil
		}
		log.Printf("id=%x participant.add watchErr=\"unexpected action\" action=%s\n", p.id, v.Action)
		return tmpErr
	case <-time.After(6 * defaultHeartbeat * p.tickDuration):
		w.Remove()
		log.Printf("id=%x participant.add watchErr=timeout\n", p.id)
		return tmpErr
	case <-p.stopNotifyc:
		return stopErr
	}
}

func (p *participant) remove(id int64) error {
	log.Printf("id=%x participant.remove nodeId=%x\n", p.id, id)
	pp := path.Join(v2machineKVPrefix, fmt.Sprint(id))

	v, err := p.Store.Get(pp, false, false)
	if err != nil {
		return nil
	}

	select {
	case p.removeNodeC <- raft.Config{NodeId: id}:
	default:
		log.Printf("id=%x participant.remove proposeErr=\"unable to send out removeNode proposal\"\n", p.id)
		return tmpErr
	}

	// TODO(xiangli): do not need to watch if the
	// removal target is self
	w, err := p.Watch(pp, true, false, v.Index()+1)
	if err != nil {
		log.Printf("id=%x participant.remove watchErr=\"%v\"\n", p.id, err)
		return tmpErr
	}

	select {
	case v := <-w.EventChan:
		if v.Action == store.Delete {
			return nil
		}
		log.Printf("id=%x participant.remove watchErr=\"unexpected action\" action=%s\n", p.id, v.Action)
		return tmpErr
	case <-time.After(6 * defaultHeartbeat * p.tickDuration):
		w.Remove()
		log.Printf("id=%x participant.remove watchErr=timeout\n", p.id)
		return tmpErr
	case <-p.stopNotifyc:
		return stopErr
	}
}

func (p *participant) apply(ents []raft.Entry) {
	offset := p.node.Applied() - int64(len(ents)) + 1
	for i, ent := range ents {
		switch ent.Type {
		// expose raft entry type
		case raft.Normal:
			if len(ent.Data) == 0 {
				continue
			}
			p.v2apply(offset+int64(i), ent)
		case raft.ClusterInit:
			p.clusterId = p.node.ClusterId()
			log.Printf("id=%x participant.cluster.setId clusterId=%x\n", p.id, p.clusterId)
		case raft.AddNode:
			cfg := new(raft.Config)
			if err := json.Unmarshal(ent.Data, cfg); err != nil {
				log.Printf("id=%x participant.cluster.addNode unmarshalErr=\"%v\"\n", p.id, err)
				break
			}
			pp := path.Join(v2machineKVPrefix, fmt.Sprint(cfg.NodeId))
			if ev, _ := p.Store.Get(pp, false, false); ev != nil {
				log.Printf("id=%x participant.cluster.addNode err=existed value=%q", p.id, *ev.Node.Value)
				break
			}
			peer, err := p.peerHub.add(cfg.NodeId, cfg.Addr)
			if err != nil {
				log.Printf("id=%x participant.cluster.addNode peerAddErr=\"%v\"\n", p.id, err)
				break
			}
			peer.participate()
			p.Store.Set(pp, false, fmt.Sprintf("raft=%v&etcd=%v", cfg.Addr, string(cfg.Context)), store.Permanent)
			if p.id == cfg.NodeId {
				p.raftPubAddr = cfg.Addr
				p.pubAddr = string(cfg.Context)
			}
			log.Printf("id=%x participant.cluster.addNode nodeId=%x addr=%s context=%s\n", p.id, cfg.NodeId, cfg.Addr, cfg.Context)
		case raft.RemoveNode:
			cfg := new(raft.Config)
			if err := json.Unmarshal(ent.Data, cfg); err != nil {
				log.Printf("id=%x participant.cluster.removeNode unmarshalErr=\"%v\"\n", p.id, err)
				break
			}
			peer, err := p.peerHub.peer(cfg.NodeId)
			if err != nil {
				log.Fatal("id=%x participant.apply getPeerErr=\"%v\"", p.id, err)
			}
			peer.idle()
			pp := path.Join(v2machineKVPrefix, fmt.Sprint(cfg.NodeId))
			p.Store.Delete(pp, false, false)
			log.Printf("id=%x participant.cluster.removeNode nodeId=%x\n", p.id, cfg.NodeId)
		default:
			panic("unimplemented")
		}
	}
}

func (p *participant) save(ents []raft.Entry, state raft.State) {
	for _, ent := range ents {
		if err := p.w.SaveEntry(&ent); err != nil {
			log.Panicf("id=%x participant.save saveEntryErr=%q", p.id, err)
		}
	}
	if !state.IsEmpty() {
		if err := p.w.SaveState(&state); err != nil {
			log.Panicf("id=%x participant.save saveStateErr=%q", p.id, err)
		}
	}
	if err := p.w.Sync(); err != nil {
		log.Panicf("id=%x participant.save syncErr=%q", p.id, err)
	}

}

func (p *participant) send(msgs []raft.Message) {
	for i := range msgs {
		if err := p.peerHub.send(msgs[i]); err != nil {
			log.Printf("id=%x participant.send err=\"%v\"\n", p.id, err)
		}
	}
}

func (p *participant) join() error {
	info := &context{
		MinVersion: store.MinVersion(),
		MaxVersion: store.MaxVersion(),
		ClientURL:  p.pubAddr,
		PeerURL:    p.raftPubAddr,
	}

	max := p.cfg.MaxRetryAttempts
	for attempt := 0; ; attempt++ {
		for seed := range p.peerHub.getSeeds() {
			if err := p.client.AddMachine(seed, fmt.Sprint(p.id), info); err == nil {
				return nil
			} else {
				log.Printf("id=%x participant.join addMachineErr=\"%v\"\n", p.id, err)
			}
		}
		if attempt == max {
			return fmt.Errorf("etcd: cannot join cluster after %d attempts", max)
		}
		time.Sleep(time.Millisecond * time.Duration(p.cfg.RetryInterval*1000))
	}
}

func genId() int64 {
	r := rand.New(rand.NewSource(int64(time.Now().Nanosecond())))
	return r.Int63()
}