seaweedfs/weed/server/raft_hashicorp.go

187 lines
5.5 KiB
Go
Raw Normal View History

2022-04-04 08:50:56 +00:00
package weed_server
// https://yusufs.medium.com/creating-distributed-kv-database-by-implementing-raft-consensus-using-golang-d0884eef2e28
// https://github.com/Jille/raft-grpc-example/blob/cd5bcab0218f008e044fbeee4facdd01b06018ad/application.go#L18
import (
"fmt"
transport "github.com/Jille/raft-grpc-transport"
"github.com/chrislusf/seaweedfs/weed/glog"
2022-04-05 10:42:19 +00:00
"github.com/chrislusf/seaweedfs/weed/pb"
2022-04-04 08:50:56 +00:00
"github.com/hashicorp/raft"
boltdb "github.com/hashicorp/raft-boltdb"
"google.golang.org/grpc"
"math/rand"
"os"
"path"
2022-04-04 08:50:56 +00:00
"path/filepath"
2022-04-05 10:42:19 +00:00
"sort"
"strings"
2022-04-04 08:50:56 +00:00
"time"
)
const (
2022-04-06 16:17:04 +00:00
ldbFile = "logs.dat"
sdbFile = "stable.dat"
updatePeersTimeout = 15 * time.Minute
)
2022-04-05 10:42:19 +00:00
func getPeerIdx(self pb.ServerAddress, mapPeers map[string]pb.ServerAddress) int {
peers := make([]pb.ServerAddress, 0, len(mapPeers))
for _, peer := range mapPeers {
peers = append(peers, peer)
}
sort.Slice(peers, func(i, j int) bool {
return strings.Compare(string(peers[i]), string(peers[j])) < 0
})
for i, peer := range peers {
if string(peer) == string(self) {
return i
}
}
return -1
}
2022-04-05 08:17:53 +00:00
func (s *RaftServer) AddPeersConfiguration() (cfg raft.Configuration) {
for _, peer := range s.peers {
cfg.Servers = append(cfg.Servers, raft.Server{
Suffrage: raft.Voter,
ID: raft.ServerID(peer),
2022-04-05 08:17:53 +00:00
Address: raft.ServerAddress(peer.ToGrpcAddress()),
})
}
return cfg
}
func (s *RaftServer) UpdatePeers() {
for {
select {
case isLeader := <-s.RaftHashicorp.LeaderCh():
if isLeader {
2022-04-05 10:42:19 +00:00
peerLeader := string(s.serverAddr)
2022-04-05 08:17:53 +00:00
existsPeerName := make(map[string]bool)
for _, server := range s.RaftHashicorp.GetConfiguration().Configuration().Servers {
if string(server.ID) == peerLeader {
continue
}
existsPeerName[string(server.ID)] = true
}
for _, peer := range s.peers {
peerName := string(peer)
if peerName == peerLeader || existsPeerName[peerName] {
2022-04-05 08:17:53 +00:00
continue
}
glog.V(0).Infof("adding new peer: %s", peerName)
2022-04-05 08:17:53 +00:00
s.RaftHashicorp.AddVoter(
raft.ServerID(peerName), raft.ServerAddress(peer.ToGrpcAddress()), 0, 0)
2022-04-05 08:17:53 +00:00
}
for peer, _ := range existsPeerName {
if _, found := s.peers[peer]; !found {
glog.V(0).Infof("removing old peer: %s", peer)
s.RaftHashicorp.RemoveServer(raft.ServerID(peer), 0, 0)
}
}
if _, found := s.peers[peerLeader]; !found {
glog.V(0).Infof("removing old leader peer: %s", peerLeader)
s.RaftHashicorp.RemoveServer(raft.ServerID(peerLeader), 0, 0)
}
}
2022-04-06 16:17:04 +00:00
return
case <-time.After(updatePeersTimeout):
return
2022-04-05 08:17:53 +00:00
}
}
}
2022-04-04 12:51:51 +00:00
func NewHashicorpRaftServer(option *RaftServerOption) (*RaftServer, error) {
2022-04-04 08:50:56 +00:00
s := &RaftServer{
peers: option.Peers,
serverAddr: option.ServerAddr,
dataDir: option.DataDir,
topo: option.Topo,
}
c := raft.DefaultConfig()
2022-04-05 08:50:39 +00:00
c.LocalID = raft.ServerID(s.serverAddr) // TODO maybee the IP:port address will change
2022-04-04 08:50:56 +00:00
c.HeartbeatTimeout = time.Duration(float64(option.HeartbeatInterval) * (rand.Float64()*0.25 + 1))
c.ElectionTimeout = option.ElectionTimeout
2022-04-04 12:51:51 +00:00
if c.LeaderLeaseTimeout > c.HeartbeatTimeout {
c.LeaderLeaseTimeout = c.HeartbeatTimeout
}
2022-04-04 08:50:56 +00:00
if glog.V(4) {
2022-04-04 12:51:51 +00:00
c.LogLevel = "Debug"
2022-04-04 08:50:56 +00:00
} else if glog.V(2) {
2022-04-04 12:51:51 +00:00
c.LogLevel = "Info"
2022-04-04 08:50:56 +00:00
} else if glog.V(1) {
2022-04-04 12:51:51 +00:00
c.LogLevel = "Warn"
2022-04-04 08:50:56 +00:00
} else if glog.V(0) {
2022-04-04 12:51:51 +00:00
c.LogLevel = "Error"
2022-04-04 08:50:56 +00:00
}
if option.RaftBootstrap {
os.RemoveAll(path.Join(s.dataDir, ldbFile))
os.RemoveAll(path.Join(s.dataDir, sdbFile))
os.RemoveAll(path.Join(s.dataDir, "snapshots"))
}
if err := os.MkdirAll(path.Join(s.dataDir, "snapshots"), os.ModePerm); err != nil {
return nil, err
}
2022-04-04 08:50:56 +00:00
baseDir := s.dataDir
ldb, err := boltdb.NewBoltStore(filepath.Join(baseDir, ldbFile))
2022-04-04 08:50:56 +00:00
if err != nil {
return nil, fmt.Errorf(`boltdb.NewBoltStore(%q): %v`, filepath.Join(baseDir, "logs.dat"), err)
}
sdb, err := boltdb.NewBoltStore(filepath.Join(baseDir, sdbFile))
2022-04-04 08:50:56 +00:00
if err != nil {
return nil, fmt.Errorf(`boltdb.NewBoltStore(%q): %v`, filepath.Join(baseDir, "stable.dat"), err)
}
fss, err := raft.NewFileSnapshotStore(baseDir, 3, os.Stderr)
if err != nil {
return nil, fmt.Errorf(`raft.NewFileSnapshotStore(%q, ...): %v`, baseDir, err)
}
s.TransportManager = transport.New(raft.ServerAddress(s.serverAddr), []grpc.DialOption{option.GrpcDialOption})
stateMachine := StateMachine{topo: option.Topo}
2022-04-04 12:51:51 +00:00
s.RaftHashicorp, err = raft.NewRaft(c, &stateMachine, ldb, sdb, fss, s.TransportManager.Transport())
2022-04-04 08:50:56 +00:00
if err != nil {
return nil, fmt.Errorf("raft.NewRaft: %v", err)
}
2022-04-05 08:17:53 +00:00
if option.RaftBootstrap || len(s.RaftHashicorp.GetConfiguration().Configuration().Servers) == 0 {
cfg := s.AddPeersConfiguration()
2022-04-05 10:42:19 +00:00
// Need to get lock, in case all servers do this at the same time.
peerIdx := getPeerIdx(s.serverAddr, s.peers)
timeSpeep := time.Duration(float64(c.LeaderLeaseTimeout) * (rand.Float64()*0.25 + 1) * float64(peerIdx))
glog.V(0).Infof("Bootstrapping idx: %d sleep: %v new cluster: %+v", peerIdx, timeSpeep, cfg)
time.Sleep(timeSpeep)
2022-04-04 12:51:51 +00:00
f := s.RaftHashicorp.BootstrapCluster(cfg)
2022-04-04 08:50:56 +00:00
if err := f.Error(); err != nil {
return nil, fmt.Errorf("raft.Raft.BootstrapCluster: %v", err)
}
2022-04-05 08:17:53 +00:00
} else {
go s.UpdatePeers()
2022-04-04 08:50:56 +00:00
}
2022-04-05 08:17:53 +00:00
2022-04-04 12:51:51 +00:00
ticker := time.NewTicker(c.HeartbeatTimeout * 10)
if glog.V(4) {
go func() {
for {
select {
case <-ticker.C:
cfuture := s.RaftHashicorp.GetConfiguration()
if err = cfuture.Error(); err != nil {
glog.Fatalf("error getting config: %s", err)
}
configuration := cfuture.Configuration()
glog.V(4).Infof("Showing peers known by %s:\n%+v", s.RaftHashicorp.String(), configuration.Servers)
}
}
}()
}
2022-04-04 08:50:56 +00:00
return s, nil
}