func (self *Meta) HasSeed(seed *topo.Node) bool {
for _, s := range self.seeds {
if s.Addr() == seed.Addr() {
if s.Id == "" {
*s = *seed
}
return true
}
}
return false
}
func checkMasterRole(node *topo.Node, ismaster bool) (bool, error) {
addr := node.Addr()
info, err := redis.FetchInfo(addr, "replication")
if err != nil {
return false, fmt.Errorf("Connect %s failed", addr)
}
if info.Get("role") == "master" {
return true, nil
} else {
return false, nil
}
}
func checkSlaveRepliStatusOk(node *topo.Node) (bool, error) {
addr := node.Addr()
info, err := redis.FetchInfo(addr, "all")
if err != nil {
return false, err
}
if info.Get("role") == "master" {
return false, nil
}
if info.Get("master_link_status") == "down" {
return false, nil
} else if info.Get("loading") == "1" {
return false, nil
} else {
return true, nil
}
}
func (self *Inspector) isFreeNode(seed *topo.Node) (bool, *topo.Node) {
resp, err := redis.ClusterNodesInRegion(seed.Addr(), self.LocalRegion)
if err != nil && strings.HasPrefix(err.Error(), "ERR Wrong CLUSTER subcommand or number of arguments") {
//server version do not support 'cluster nodes extra [region]'
resp, err = redis.ClusterNodes(seed.Addr())
}
if err != nil {
return false, nil
}
numNode := 0
lines := strings.Split(resp, "\n")
for _, line := range lines {
line = strings.TrimSpace(line)
if line == "" || strings.HasPrefix(line, "# ") {
continue
}
numNode++
}
if numNode != 1 {
return false, nil
}
for _, line := range lines {
line = strings.TrimSpace(line)
if line == "" || strings.HasPrefix(line, "# ") {
continue
}
node, myself, err := self.buildNode(line)
if node.Ip == "127.0.0.1" {
node.Ip = seed.Ip
}
// 只看到自己，是主，且没有slots，才认为是FreeNode
if !myself {
return false, nil
}
if err != nil || len(node.Ranges) > 0 || !node.IsMaster() {
return false, nil
} else {
return true, node
}
}
return false, nil
}
func configAofAndRdb(node *topo.Node, state bool) error {
addr := node.Addr()
var err error
var err1 error
var err2 error
if state {
_, err = redis.RedisCli(addr, "config", "set", "appendonly", "yes")
_, err1 = redis.RedisCli(addr, "config", "set", "dbfilename", "dump.rdb")
} else {
_, err = redis.RedisCli(addr, "config", "set", "appendonly", "no")
_, err1 = redis.RedisCli(addr, "config", "set", "dbfilename", "tmp.rdb")
}
_, err2 = redis.RedisCli(addr, "config", "rewrite")
if err != nil {
return err
}
if err1 != nil {
return err1
}
if err2 != nil {
return err2
}
return nil
}
func upgradeMaster(c *cli.Context) {
pid := context.GetAppName()
addr := context.GetLeaderAddr()
url_fr := "http://" + addr + api.FetchReplicaSetsPath
url_fl := "http://" + addr + api.NodeSetAsMasterPath
extraHeader := &utils.ExtraHeader{
User: context.Config.User,
Role: context.Config.Role,
Token: context.Config.Token,
}
resp, err := utils.HttpGet(url_fr, nil, 5*time.Second)
if err != nil {
fmt.Println(err)
return
}
var rss command.FetchReplicaSetsResult
err = utils.InterfaceToStruct(resp.Body, &rss)
if err != nil {
fmt.Println(err)
return
}
sort.Sort(topo.ByMasterId(rss.ReplicaSets))
sort.Sort(topo.ByNodeState(rss.ReplicaSets))
iidx, err := getIdx(IdxServerAddr, pid, "master")
if err != nil {
fmt.Println(err)
return
}
fmt.Printf("Get last idx record: %d\n", iidx)
var old_master *topo.Node
var new_master *topo.Node
//used to check status
var new_slaves []*topo.Node
old_master = nil
new_master = nil
for idx, rs := range rss.ReplicaSets {
if rs.Master.IsArbiter() {
continue
}
if idx <= iidx {
fmt.Printf("Skipping replica(id:%s) (%d/%d) master\n", rs.Master.Id, idx, len(rss.ReplicaSets))
continue
}
//select a slave in the same IDC
old_master = rs.Master
old_master_r := getRegion(old_master)
if old_master_r == "" {
return
}
new_slaves = append(new_slaves, old_master)
fmt.Printf("Upgrading replica(id:%s) (%d/%d) master\n", rs.Master.Id, idx, len(rss.ReplicaSets))
skip := false
for _, s := range rs.Slaves {
re := getRegion(s)
if re == "" {
return
}
if re == old_master_r && !skip {
new_master = s
skip = true
} else {
new_slaves = append(new_slaves, s)
}
}
if new_master == nil {
fmt.Printf("Select new master failed for master(%s) replica\n", old_master.Id)
return
}
//send failover to the new master
req := api.FailoverTakeoverParams{
NodeId: new_master.Id,
}
resp, err := utils.HttpPostExtra(url_fl, req, 10*time.Second, extraHeader)
if err != nil {
fmt.Println(err)
return
}
if resp.Errno != 0 {
fmt.Println(resp.Errmsg)
return
}
//send failover request done,check the new_master role to a real master
for {
ismaster, err := checkMasterRole(new_master, true)
if err != nil {
fmt.Println(err)
time.Sleep(10 * time.Second)
continue
}
if ismaster == true {
//to be a new master
break
} else {
//wait for next check
time.Sleep(10 * time.Second)
}
}
//disable read flag of the all new slaves,including old master
for _, s := range new_slaves {
resp, err = configRead(s, false)
if err != nil {
fmt.Println(err)
return
}
if resp.Errno != 0 {
fmt.Println(resp.Errmsg)
return
}
}
//disable aof and rdb to speed up start
err = configAofAndRdb(old_master, false)
if err != nil {
fmt.Println(err)
return
}
//shutdown server
err = shutdownServer(old_master)
if err != nil {
fmt.Printf("server %s restart\n", old_master.Addr())
}
//check the status of old master
cnt := 1
for {
fmt.Printf("Check slave status %d times\n", cnt)
cnt++
inner := func(nodes []*topo.Node) bool {
rok := true
for _, n := range nodes {
ok, err := checkSlaveRepliStatusOk(n)
if ok {
//replica status ok,enable read flag,ignore result
configRead(n, true)
continue
}
if !ok || err != nil {
rok = false
}
}
return rok
}
ok := inner(new_slaves)
if !ok {
//not ok, wait for next trun check
time.Sleep(10 * time.Second)
} else {
break
}
}
//enable aof and rdb
err = configAofAndRdb(old_master, true)
if err != nil {
fmt.Println(err)
return
}
//save the idx of the process
err = saveIdx(IdxServerAddr, pid, "master", idx)
if err != nil {
fmt.Println(err)
return
}
}
}
func shutdownServer(node *topo.Node) error {
addr := node.Addr()
_, err := redis.RedisCli(addr, "shutdown", "nosave")
return err
}
// 生成ClusterSnapshot
func (self *Inspector) BuildClusterTopo() (*topo.Cluster, []*topo.Node, error) {
self.mutex.Lock()
defer self.mutex.Unlock()
if len(meta.Seeds()) == 0 {
return nil, nil, ErrNoSeed
}
// 过滤掉连接不上的节点
seeds := []*topo.Node{}
for _, s := range meta.Seeds() {
if redis.IsAlive(s.Addr()) {
seeds = append(seeds, s)
} else {
// remove this seed from meta seeds
// will re-add to seeds if join the cluster again
meta.RemoveSeed(s.Addr())
}
}
if len(seeds) == 0 {
return nil, seeds, ErrNoSeed
}
// 顺序选一个节点，获取nodes数据作为基准，再用其他节点的数据与基准做对比
if self.SeedIndex >= len(seeds) {
self.SeedIndex = len(seeds) - 1
}
var seed *topo.Node
for i := 0; i < len(seeds); i++ {
seed = seeds[self.SeedIndex]
self.SeedIndex++
self.SeedIndex %= len(seeds)
if seed.Free {
glog.Info("Seed node is free ", seed.Addr())
} else {
break
}
}
cluster, err := self.initClusterTopo(seed)
if err != nil {
glog.Infof("InitClusterTopo failed")
return nil, seeds, err
}
// 检查所有节点返回的信息是不是相同，如果不同说明正在变化中，直接返回等待重试
if len(seeds) > 1 {
for _, s := range seeds {
if s == seed {
continue
}
err := self.checkClusterTopo(s, cluster)
if err != nil {
free, node := self.isFreeNode(s)
if free {
node.Free = true
glog.Infof("Found free node %s", node.Addr())
cluster.AddNode(node)
} else {
glog.Infof("checkClusterTopo failed")
return cluster, seeds, err
}
} else {
s.Free = false
}
}
}
// 构造LocalRegion视图
for _, s := range cluster.LocalRegionNodes() {
if s.PFailCount() > cluster.NumLocalRegionNode()/2 {
glog.Infof("Found %d/%d PFAIL state on %s, set FAIL",
s.PFailCount(), cluster.NumLocalRegionNode(), s.Addr())
s.SetFail(true)
}
}
if meta.IsClusterLeader() {
cluster.BuildReplicaSets()
}
meta.MergeSeeds(cluster.LocalRegionNodes())
self.ClusterTopo = cluster
return cluster, seeds, nil
}
func (self *Inspector) checkClusterTopo(seed *topo.Node, cluster *topo.Cluster) error {
resp, err := redis.ClusterNodesInRegion(seed.Addr(), self.LocalRegion)
if err != nil && strings.HasPrefix(err.Error(), "ERR Wrong CLUSTER subcommand or number of arguments") {
//server version do not support 'cluster nodes extra [region]'
resp, err = redis.ClusterNodes(seed.Addr())
}
//this may lead to BuildClusterTopo update failed for a time
//the node is step into this state after check IsAlive
if err != nil && strings.HasPrefix(err.Error(), "LOADING") {
return nil
}
if err != nil {
return err
}
var summary topo.SummaryInfo
lines := strings.Split(resp, "\n")
for _, line := range lines {
if strings.HasPrefix(line, "# ") {
summary.ReadLine(line)
continue
}
line = strings.TrimSpace(line)
if line == "" {
continue
}
s, myself, err := self.buildNode(line)
if err == ErrNodeInHandShake || err == ErrNodeNoAddr {
continue
}
// Fix 'cluster nodes extra' & 'cluster nodes extra region' compatiable
if s.Region != self.LocalRegion {
continue
}
if err != nil {
return err
}
if s.Ip == "127.0.0.1" {
s.Ip = seed.Ip
}
node := cluster.FindNode(s.Id)
if node == nil {
if s.PFail {
glog.Warningf("forget dead node %s(%s) should be forgoten", s.Id, s.Addr())
//redis.ClusterForget(seed.Addr(), s.Id)
}
return fmt.Errorf("node not exist %s(%s)", s.Id, s.Addr())
}
// 对比节点数据是否相同
if !node.Compare(s) {
glog.Infof("%#v vs %#v different", s, node)
if s.Tag == "-" && node.Tag != "-" {
// 可能存在处于不被Cluster接受的节点，节点可以看见Cluster，但Cluster看不到它。
// 一种复现情况情况：某个节点已经死了，系统将其Forget，但是OP并未被摘除该节点，
// 而是恢复了该节点。
glog.Warningf("remeet node %s", seed.Addr())
self.MeetNode(seed)
}
return ErrNodesInfoNotSame
}
if len(node.Ranges) == 0 && len(s.Ranges) > 0 {
glog.Warningf("Ranges not equal, use nonempty ranges.")
node.Ranges = s.Ranges
}
if myself {
info, err := redis.FetchClusterInfo(node.Addr())
if err != nil {
return err
}
node.ClusterInfo = info
node.SummaryInfo = summary
}
if len(s.Migrating) != 0 {
node.Migrating = s.Migrating
}
if len(s.Importing) != 0 {
node.Importing = s.Importing
}
if s.PFail {
node.IncrPFailCount()
}
}
return nil
}
func (self *Inspector) initClusterTopo(seed *topo.Node) (*topo.Cluster, error) {
resp, err := redis.ClusterNodesInRegion(seed.Addr(), self.LocalRegion)
if err != nil && strings.HasPrefix(err.Error(), "ERR Wrong CLUSTER subcommand or number of arguments") {
//server version do not support 'cluster nodes extra [region]'
resp, err = redis.ClusterNodes(seed.Addr())
}
if err != nil {
return nil, err
}
cluster := topo.NewCluster(self.LocalRegion)
var summary topo.SummaryInfo
var nodeidx *topo.Node
var cnt int
lines := strings.Split(resp, "\n")
cnt = 0
for _, line := range lines {
if strings.HasPrefix(line, "# ") {
summary.ReadLine(line)
continue
}
line = strings.TrimSpace(line)
if line == "" {
continue
}
node, myself, err := self.buildNode(line)
if err == ErrNodeInHandShake || err == ErrNodeNoAddr {
continue
}
// Fix 'cluster nodes extra' & 'cluster nodes extra region' compatiable
if node.Region != self.LocalRegion {
continue
}
if err != nil {
return nil, err
}
if node.Ip == "127.0.0.1" {
node.Ip = seed.Ip
}
// 遇到myself，读取该节点的ClusterInfo
if myself {
info, err := redis.FetchClusterInfo(node.Addr())
if err != nil {
return nil, err
}
node.ClusterInfo = info
node.SummaryInfo = summary
}
cluster.AddNode(node)
nodeidx = node
cnt++
}
if cnt == 1 {
if nodeidx.IsMaster() && len(nodeidx.Ranges) == 0 {
glog.Infof("Node %s is free node", nodeidx.Addr())
nodeidx.SetFree(true)
}
}
return cluster, nil
}