// experimental simple auto rebalance :)
func Rebalance() error {
targetQuota, err := getQuotaMap(safeZkConn)
if err != nil {
return errors.Trace(err)
}
livingNodes, err := getLivingNodeInfos(safeZkConn)
if err != nil {
return errors.Trace(err)
}
log.Infof("start rebalance")
for _, node := range livingNodes {
for len(node.CurSlots) > targetQuota[node.GroupId] {
for _, dest := range livingNodes {
if dest.GroupId != node.GroupId && len(dest.CurSlots) < targetQuota[dest.GroupId] {
slot := node.CurSlots[len(node.CurSlots)-1]
// create a migration task
info := &MigrateTaskInfo{
Delay: 0,
SlotId: slot,
NewGroupId: dest.GroupId,
Status: MIGRATE_TASK_PENDING,
CreateAt: strconv.FormatInt(time.Now().Unix(), 10),
}
globalMigrateManager.PostTask(info)
node.CurSlots = node.CurSlots[0 : len(node.CurSlots)-1]
dest.CurSlots = append(dest.CurSlots, slot)
}
}
}
}
log.Infof("rebalance tasks submit finish")
return nil
}
/*
* 对后端返回的ASK，重新向正确的后端发送
* r : 已经返回ASK的请求
*/
func (s *Session) handleAskRequest(r *Request) (*Request, error) {
// reset response
r.Response.Resp, r.Response.Err = nil, nil
usnow := microseconds()
s.LastOpUnix = usnow / 1e6
s.Ops++
// get hash key of the request
hkey := getHashKey(r.Resp, r.OpStr)
log.Infof("ASK: use key: %s to choose backend", string(hkey))
// create a asking request to backend
nr := &Request{
OpStr: "ASKING",
Start: usnow,
Resp: redis.NewString([]byte("ASKING")),
Wait: &sync.WaitGroup{},
Failed: &s.failed,
}
nr, err := s.handleRequestAsking(nr, hkey)
log.Infof("ASK: after dispatch asking request, waiting. error=%s", err)
if err != nil {
return r, err
} else {
s.GetTasks() <- nr
}
// resend the r to backend
err = s.GetDispather().Dispatch(r)
if err == nil {
s.GetTasks() <- r
}
return r, err
}
func (s *Server) waitOnline() bool {
for {
info, err := s.topo.GetProxyInfo(s.info.Id)
if err != nil {
log.PanicErrorf(err, "get proxy info failed: %s", s.info.Id)
}
switch info.State {
case models.PROXY_STATE_MARK_OFFLINE:
log.Infof("mark offline, proxy got offline event: %s", s.info.Id)
s.markOffline()
return false
case models.PROXY_STATE_ONLINE:
s.info.State = info.State
log.Infof("we are online: %s", s.info.Id)
s.rewatchProxy()
return true
}
select {
case <-s.kill:
log.Infof("mark offline, proxy is killed: %s", s.info.Id)
s.markOffline()
return false
default:
}
log.Infof("wait to be online: %s", s.info.Id)
time.Sleep(3 * time.Second)
}
}
func (s *Server) loopEvents() {
ticker := time.NewTicker(time.Second)
defer ticker.Stop()
var tick int = 0
for s.info.State == models.PROXY_STATE_ONLINE {
select {
case <-s.kill:
log.Infof("mark offline, proxy is killed: %s", s.info.Id)
s.markOffline()
case e := <-s.evtbus:
evtPath := getEventPath(e)
log.Infof("got event %s, %v, lastActionSeq %d", s.info.Id, e, s.lastActionSeq)
if strings.Index(evtPath, models.GetActionResponsePath(s.conf.productName)) == 0 {
seq, err := strconv.Atoi(path.Base(evtPath))
if err != nil {
log.ErrorErrorf(err, "parse action seq failed")
} else {
if seq < s.lastActionSeq {
log.Infof("ignore seq = %d", seq)
continue
}
}
}
s.processAction(e)
case <-ticker.C:
if maxTick := s.conf.pingPeriod; maxTick != 0 {
if tick++; tick >= maxTick {
s.router.KeepAlive()
tick = 0
}
}
}
}
}
func (t *MigrateTask) run() error {
log.Infof("migration start: %+v", t.MigrateTaskInfo)
to := t.NewGroupId
t.UpdateStatus(MIGRATE_TASK_MIGRATING)
err := t.migrateSingleSlot(t.SlotId, to)
if err != nil {
log.ErrorErrorf(err, "migrate single slot failed")
t.UpdateStatus(MIGRATE_TASK_ERR)
t.rollbackPremigrate()
return err
}
t.UpdateFinish()
log.Infof("migration finished: %+v", t.MigrateTaskInfo)
return nil
}
func (s *Session) handleRequestAsking(r *Request, extra []byte) (*Request, error) {
log.Infof("ASK: handling asking request to backend")
if err := s.GetDispather().DispatchAsking(r, extra); err != nil {
return nil, err
}
return r, nil
}
func TestProxyOfflineInWaitActionReceiver(t *testing.T) {
log.Infof("test proxy offline when waiting action response")
fakeZkConn := zkhelper.NewConn()
for i := 1; i <= 4; i++ {
CreateProxyInfo(fakeZkConn, productName, &ProxyInfo{
Id: strconv.Itoa(i),
State: PROXY_STATE_ONLINE,
})
go waitForProxyMarkOffline(fakeZkConn, strconv.Itoa(i))
}
lst, _ := ProxyList(fakeZkConn, productName, nil)
assert.Must(len(lst) == 4)
go func() {
time.Sleep(500 * time.Millisecond)
actionPath := path.Join(GetActionResponsePath(productName), fakeZkConn.Seq2Str(1))
//create test response for proxy 4, means proxy 1,2,3 are timeout
fakeZkConn.Create(path.Join(actionPath, "4"), nil,
0, zkhelper.DefaultFileACLs())
}()
err := NewActionWithTimeout(fakeZkConn, productName, ACTION_TYPE_SLOT_CHANGED, nil, "desc", true, 3*1000)
if err != nil {
assert.Must(err.Error() == ErrReceiverTimeout.Error())
}
for i := 1; i <= 3; i++ {
info, _ := GetProxyInfo(fakeZkConn, productName, strconv.Itoa(i))
assert.Must(info.State == PROXY_STATE_OFFLINE)
}
}
func (s *Server) responseAction(seq int64) {
log.Infof("send response seq = %d", seq)
err := s.topo.DoResponse(int(seq), &s.info)
if err != nil {
log.InfoErrorf(err, "send response seq = %d failed", seq)
}
}
func releaseDashboardNode() {
zkPath := fmt.Sprintf("/zk/codis/db_%s/dashboard", globalEnv.ProductName())
if exists, _, _ := safeZkConn.Exists(zkPath); exists {
log.Infof("removing dashboard node")
safeZkConn.Delete(zkPath, 0)
}
}
func (s *Server) processAction(e interface{}) {
if strings.Index(getEventPath(e), models.GetProxyPath(s.topo.ProductName)) == 0 {
info, err := s.topo.GetProxyInfo(s.info.Id)
if err != nil {
log.PanicErrorf(err, "get proxy info failed: %s", s.info.Id)
}
switch info.State {
case models.PROXY_STATE_MARK_OFFLINE:
log.Infof("mark offline, proxy got offline event: %s", s.info.Id)
s.markOffline()
case models.PROXY_STATE_ONLINE:
s.rewatchProxy()
default:
log.Panicf("unknown proxy state %v", info)
}
return
}
//re-watch
nodes := s.rewatchNodes()
seqs, err := models.ExtraSeqList(nodes)
if err != nil {
log.PanicErrorf(err, "get seq list failed")
}
if len(seqs) == 0 || !s.topo.IsChildrenChangedEvent(e) {
return
}
//get last pos
index := -1
for i, seq := range seqs {
if s.lastActionSeq < seq {
index = i
break
}
}
if index < 0 {
return
}
actions := seqs[index:]
for _, seq := range actions {
exist, err := s.topo.Exist(path.Join(s.topo.GetActionResponsePath(seq), s.info.Id))
if err != nil {
log.PanicErrorf(err, "get action failed")
}
if exist {
continue
}
if s.checkAndDoTopoChange(seq) {
s.responseAction(int64(seq))
}
}
s.lastActionSeq = seqs[len(seqs)-1]
}
func (bc *BackendConn) Run() {
log.Infof("backend conn [%p] to %s, start service", bc, bc.addr)
for k := 0; ; k++ {
err := bc.loopWriter()
if err == nil {
break
} else {
for i := len(bc.input); i != 0; i-- {
r := <-bc.input
bc.setResponse(r, nil, err)
}
}
log.WarnErrorf(err, "backend conn [%p] to %s, restart [%d]", bc, bc.addr, k)
time.Sleep(time.Millisecond * 50)
}
log.Infof("backend conn [%p] to %s, stop and exit", bc, bc.addr)
}
func (s *Server) getActionObject(seq int, target interface{}) {
act := &models.Action{Target: target}
err := s.topo.GetActionWithSeqObject(int64(seq), act)
if err != nil {
log.PanicErrorf(err, "get action object failed, seq = %d", seq)
}
log.Infof("action %+v", act)
}
func (s *Server) onGroupChange(groupId int) {
log.Infof("group changed %d", groupId)
for i, g := range s.groups {
if g == groupId {
s.fillSlot(i)
}
}
}
func NewSessionSize(c net.Conn, auth string, bufsize int, timeout int) *Session {
s := &Session{CreateUnix: time.Now().Unix(), auth: auth}
s.Conn = redis.NewConnSize(c, bufsize)
s.Conn.ReaderTimeout = time.Second * time.Duration(timeout)
s.Conn.WriterTimeout = time.Second * 30
log.Infof("session [%p] create: %s", s, s)
return s
}
func New(addr string, debugVarAddr string, conf *Config) *Server {
log.Infof("create proxy with config: %+v", conf)
proxyHost := strings.Split(addr, ":")[0]
debugHost := strings.Split(debugVarAddr, ":")[0]
hostname, err := os.Hostname()
if err != nil {
log.PanicErrorf(err, "get host name failed")
}
if proxyHost == "0.0.0.0" || strings.HasPrefix(proxyHost, "127.0.0.") || proxyHost == "" {
proxyHost = hostname
}
if debugHost == "0.0.0.0" || strings.HasPrefix(debugHost, "127.0.0.") || debugHost == "" {
debugHost = hostname
}
s := &Server{conf: conf, lastActionSeq: -1, groups: make(map[int]int)}
s.topo = NewTopo(conf.productName, conf.zkAddr, conf.fact, conf.provider, conf.zkSessionTimeout)
s.info.Id = conf.proxyId
s.info.State = models.PROXY_STATE_OFFLINE
s.info.Addr = proxyHost + ":" + strings.Split(addr, ":")[1]
s.info.DebugVarAddr = debugHost + ":" + strings.Split(debugVarAddr, ":")[1]
s.info.Pid = os.Getpid()
s.info.StartAt = time.Now().String()
s.kill = make(chan interface{})
log.Infof("proxy info = %+v", s.info)
if l, err := net.Listen(conf.proto, addr); err != nil {
log.PanicErrorf(err, "open listener failed")
} else {
s.listener = l
}
s.router = router.NewWithAuth(conf.passwd)
s.evtbus = make(chan interface{}, 1024)
s.register()
s.wait.Add(1)
go func() {
defer s.wait.Done()
s.serve()
}()
return s
}
func (s *Slot) prepareAsk(r *Request) (*SharedBackendConn, error) {
if s.migrate.bc == nil {
log.Infof("ASK: slot-%04d is not migrating", s.id)
return nil, ErrSlotIsNotMigrating
}
r.slot = &s.wait
r.slot.Add(1)
return s.migrate.bc, nil
}
func (t *MigrateTask) rollbackPremigrate() {
if s, err := models.GetSlot(t.zkConn, t.productName, t.SlotId); err == nil && s.State.Status == models.SLOT_STATUS_PRE_MIGRATE {
s.State.Status = models.SLOT_STATUS_ONLINE
err = s.Update(t.zkConn)
if err != nil {
log.Warn("rollback premigrate failed", err)
} else {
log.Infof("rollback slot %d from premigrate to online\n", s.Id)
}
}
}
func (s *Server) onSlotRangeChange(param *models.SlotMultiSetParam) {
log.Infof("slotRangeChange %+v", param)
for i := param.From; i <= param.To; i++ {
switch param.Status {
case models.SLOT_STATUS_OFFLINE:
s.resetSlot(i)
case models.SLOT_STATUS_ONLINE:
s.fillSlot(i)
default:
log.Panicf("can not handle status %v", param.Status)
}
}
}
func (s *Router) fillSlot(i int, addr, from string, lock bool) {
if !s.isValidSlot(i) {
return
}
slot := s.slots[i]
slot.blockAndWait()
s.putBackendConn(slot.backend.bc)
s.putBackendConn(slot.migrate.bc)
slot.reset()
if len(addr) != 0 {
xx := strings.Split(addr, ":")
if len(xx) >= 1 {
slot.backend.host = []byte(xx[0])
}
if len(xx) >= 2 {
slot.backend.port = []byte(xx[1])
}
slot.backend.addr = addr
slot.backend.bc = s.getBackendConn(addr)
}
if len(from) != 0 {
slot.migrate.from = from
slot.migrate.bc = s.getBackendConn(from)
}
if !lock {
slot.unblock()
}
if slot.migrate.bc != nil {
log.Infof("fill slot %04d, backend.addr = %s, migrate.from = %s",
i, slot.backend.addr, slot.migrate.from)
} else {
log.Infof("fill slot %04d, backend.addr = %s",
i, slot.backend.addr)
}
}
func (s *Session) Serve(d Dispatcher, maxPipeline int) {
if d == nil {
log.Infof("dispatcher is nil")
return
}
// set dispacher for this session
s.SetDispather(d)
var errlist errors.ErrorList
defer func() {
if err := errlist.First(); err != nil {
log.Infof("session [%p] closed: %s, error = %s", s, s, err)
} else {
log.Infof("session [%p] closed: %s, quit", s, s)
}
}()
tasks := make(chan *Request, maxPipeline)
s.SetTasks(tasks)
go func() {
defer func() {
s.Close()
for _ = range tasks {
}
}()
if err := s.loopWriter(tasks); err != nil {
errlist.PushBack(err)
}
}()
defer func(tasks chan<- *Request) {
close(tasks)
s.SetTasks(nil)
}(tasks)
if err := s.loopReader(tasks); err != nil {
errlist.PushBack(err)
}
}
func (s *Session) handleRequest(resp *redis.Resp) (*Request, error) {
opstr, err := getOpStr(resp)
log.Infof("In handleRequest, opstr is %s", string(opstr))
if err != nil {
return nil, err
}
if isNotAllowed(opstr) {
return nil, errors.New(fmt.Sprintf("command <%s> is not allowed", opstr))
}
usnow := microseconds()
s.LastOpUnix = usnow / 1e6
s.Ops++
r := &Request{
OpStr: opstr,
Start: usnow,
Resp: resp,
Wait: &sync.WaitGroup{},
Failed: &s.failed,
}
if opstr == "QUIT" {
return s.handleQuit(r)
}
if opstr == "AUTH" {
return s.handleAuth(r)
}
if !s.authorized {
if s.auth != "" {
r.Response.Resp = redis.NewError([]byte("NOAUTH Authentication required."))
return r, nil
}
s.authorized = true
}
switch opstr {
case "SELECT":
return s.handleSelect(r)
case "PING":
return s.handlePing(r)
case "MGET":
return s.handleRequestMGet(r)
case "MSET":
return s.handleRequestMSet(r)
case "DEL":
return s.handleRequestMDel(r)
}
return r, s.GetDispather().Dispatch(r)
}
func (s *Session) loopWriter(tasks <-chan *Request) error {
p := &FlushPolicy{
Encoder: s.Writer,
MaxBuffered: 32,
MaxInterval: 300,
}
for r := range tasks {
resp, err := s.handleResponse(r)
log.Infof("receive a response %s", string(r.Response.Resp.Value))
// 当resp返回为ASK，需要重新请求后端
if r.isAskResp() {
log.Infof("ASK: handle -ASK req:response %s:%s", string(r.Resp.Value), string(r.Response.Resp.Value))
if _, err = s.handleAskRequest(r); err != nil {
log.Infof("ASK: handleAskRequest error %s", err)
return err
}
// 重新处理请求的返回
log.Infof("ASK: receive new response")
resp, err = s.handleResponse(r)
log.Infof("ASK: new response %s", string(r.Response.Resp.Value))
}
// 如果该请求为proxy主动发送的请求，将response swallow
if r.isAskingReq() {
return nil
}
if err != nil {
return err
}
if err = p.Encode(resp, len(tasks) == 0); err != nil {
return err
}
}
return nil
}
func (s *Slot) prepare(r *Request, key []byte) (*SharedBackendConn, error) {
if s.backend.bc == nil {
log.Infof("slot-%04d is not ready: key = %s", s.id, key)
return nil, ErrSlotIsNotReady
}
if err := s.slotsmgrt(r, key); err != nil {
log.Warnf("slot-%04d migrate from = %s to %s failed: key = %s, error = %s",
s.id, s.migrate.from, s.backend.addr, key, err)
return nil, err
} else {
r.slot = &s.wait
r.slot.Add(1)
return s.backend.bc, nil
}
}
func setLogLevel(level string) {
var lv = log.LEVEL_INFO
switch strings.ToLower(level) {
case "error":
lv = log.LEVEL_ERROR
case "warn", "warning":
lv = log.LEVEL_WARN
case "debug":
lv = log.LEVEL_DEBUG
case "info":
fallthrough
default:
lv = log.LEVEL_INFO
}
log.SetLevel(lv)
log.Infof("set log level to %s", lv)
}
func setLogLevel(level string) {
level = strings.ToLower(level)
var l = log.LEVEL_INFO
switch level {
case "error":
l = log.LEVEL_ERROR
case "warn", "warning":
l = log.LEVEL_WARN
case "debug":
l = log.LEVEL_DEBUG
case "info":
fallthrough
default:
level = "info"
l = log.LEVEL_INFO
}
log.SetLevel(l)
log.Infof("set log level to <%s>", level)
}
func createDashboardNode() error {
// make sure root dir is exists
rootDir := fmt.Sprintf("/zk/codis/db_%s", globalEnv.ProductName())
zkhelper.CreateRecursive(safeZkConn, rootDir, "", 0, zkhelper.DefaultDirACLs())
zkPath := fmt.Sprintf("%s/dashboard", rootDir)
// make sure we're the only one dashboard
if exists, _, _ := safeZkConn.Exists(zkPath); exists {
data, _, _ := safeZkConn.Get(zkPath)
return errors.New("dashboard already exists: " + string(data))
}
content := fmt.Sprintf(`{"addr": "%v", "pid": %v}`, globalEnv.DashboardAddr(), os.Getpid())
pathCreated, err := safeZkConn.Create(zkPath, []byte(content), 0, zkhelper.DefaultFileACLs())
createdDashboardNode = true
log.Infof("dashboard node created: %v, %s", pathCreated, string(content))
log.Warn("********** Attention **********")
log.Warn("You should use `kill {pid}` rather than `kill -9 {pid}` to stop me,")
log.Warn("or the node resisted on zk will not be cleaned when I'm quiting and you must remove it manually")
log.Warn("*******************************")
return errors.Trace(err)
}
func runDashboard(addr string, httpLogFile string) {
log.Infof("dashboard listening on addr: %s", addr)
m := martini.Classic()
f, err := os.OpenFile(httpLogFile, os.O_RDWR|os.O_CREATE|os.O_APPEND, 0666)
if err != nil {
log.PanicErrorf(err, "open http log file failed")
}
defer f.Close()
m.Map(stdlog.New(f, "[martini]", stdlog.LstdFlags))
binRoot, err := filepath.Abs(filepath.Dir(os.Args[0]))
if err != nil {
log.PanicErrorf(err, "get binroot path failed")
}
m.Use(martini.Static(filepath.Join(binRoot, "assets/statics")))
m.Use(render.Renderer(render.Options{
Directory: filepath.Join(binRoot, "assets/template"),
Extensions: []string{".tmpl", ".html"},
Charset: "UTF-8",
IndentJSON: true,
}))
m.Use(cors.Allow(&cors.Options{
AllowOrigins: []string{"*"},
AllowMethods: []string{"POST", "GET", "DELETE", "PUT"},
AllowHeaders: []string{"Origin", "x-requested-with", "Content-Type", "Content-Range", "Content-Disposition", "Content-Description"},
ExposeHeaders: []string{"Content-Length"},
AllowCredentials: false,
}))
m.Get("/api/server_groups", apiGetServerGroupList)
m.Get("/api/overview", apiOverview)
m.Get("/api/redis/:addr/stat", apiRedisStat)
m.Get("/api/redis/:addr/:id/slotinfo", apiGetRedisSlotInfo)
m.Get("/api/redis/group/:group_id/:slot_id/slotinfo", apiGetRedisSlotInfoFromGroupId)
m.Put("/api/server_groups", binding.Json(models.ServerGroup{}), apiAddServerGroup)
m.Put("/api/server_group/(?P<id>[0-9]+)/addServer", binding.Json(models.Server{}), apiAddServerToGroup)
m.Delete("/api/server_group/(?P<id>[0-9]+)", apiRemoveServerGroup)
m.Put("/api/server_group/(?P<id>[0-9]+)/removeServer", binding.Json(models.Server{}), apiRemoveServerFromGroup)
m.Get("/api/server_group/(?P<id>[0-9]+)", apiGetServerGroup)
m.Post("/api/server_group/(?P<id>[0-9]+)/promote", binding.Json(models.Server{}), apiPromoteServer)
m.Get("/api/migrate/status", apiMigrateStatus)
m.Get("/api/migrate/tasks", apiGetMigrateTasks)
m.Post("/api/migrate", binding.Json(migrateTaskForm{}), apiDoMigrate)
m.Post("/api/rebalance", apiRebalance)
m.Get("/api/slot/list", apiGetSlots)
m.Get("/api/slot/:id", apiGetSingleSlot)
m.Post("/api/slots/init", apiInitSlots)
m.Get("/api/slots", apiGetSlots)
m.Post("/api/slot", binding.Json(RangeSetTask{}), apiSlotRangeSet)
m.Get("/api/proxy/list", apiGetProxyList)
m.Get("/api/proxy/debug/vars", apiGetProxyDebugVars)
m.Post("/api/proxy", binding.Json(models.ProxyInfo{}), apiSetProxyStatus)
m.Get("/api/action/gc", apiActionGC)
m.Get("/api/force_remove_locks", apiForceRemoveLocks)
m.Get("/api/remove_fence", apiRemoveFence)
m.Get("/slots", pageSlots)
m.Get("/", func(r render.Render) {
r.Redirect("/admin")
})
zkBuilder := utils.NewConnBuilder(globalEnv.NewZkConn)
safeZkConn = zkBuilder.GetSafeConn()
unsafeZkConn = zkBuilder.GetUnsafeConn()
// create temp node in ZK
if err := createDashboardNode(); err != nil {
log.PanicErrorf(err, "create zk node failed") // do not release dashborad node here
}
// create long live migrate manager
globalMigrateManager = NewMigrateManager(safeZkConn, globalEnv.ProductName())
go func() {
tick := time.Tick(time.Second)
var lastCnt, qps int64
for _ = range tick {
cnt := getAllProxyOps()
if cnt > 0 {
qps = cnt - lastCnt
lastCnt = cnt
} else {
qps = 0
}
atomic.StoreInt64(&proxiesSpeed, qps)
}
}()
m.RunOnAddr(addr)
}
func SetProxyStatus(zkConn zkhelper.Conn, productName string, proxyName string, status string) error {
p, err := GetProxyInfo(zkConn, productName, proxyName)
if err != nil {
return errors.Trace(err)
}
if status != PROXY_STATE_ONLINE && status != PROXY_STATE_MARK_OFFLINE && status != PROXY_STATE_OFFLINE {
return errors.Errorf("%v, %s", ErrUnknownProxyStatus, status)
}
// check slot status before setting proxy online
if status == PROXY_STATE_ONLINE {
slots, err := Slots(zkConn, productName)
if err != nil {
return errors.Trace(err)
}
for _, slot := range slots {
if slot.State.Status != SLOT_STATUS_ONLINE && slot.State.Status != SLOT_STATUS_MIGRATE {
return errors.Errorf("slot %v is not online or migrate", slot)
}
if slot.GroupId == INVALID_ID {
return errors.Errorf("slot %v has invalid group id", slot)
}
}
}
p.State = status
b, _ := json.Marshal(p)
_, err = zkConn.Set(path.Join(GetProxyPath(productName), proxyName), b, -1)
if err != nil {
return errors.Trace(err)
}
if status == PROXY_STATE_MARK_OFFLINE {
// wait for the proxy down
for {
_, _, c, err := zkConn.GetW(path.Join(GetProxyPath(productName), proxyName))
if zkhelper.ZkErrorEqual(err, zk.ErrNoNode) {
return nil
} else if err != nil {
return errors.Trace(err)
}
<-c
info, err := GetProxyInfo(zkConn, productName, proxyName)
log.Info("mark_offline, check proxy status:", proxyName, info, err)
if zkhelper.ZkErrorEqual(err, zk.ErrNoNode) {
log.Info("shutdown proxy successful")
return nil
} else if err != nil {
return errors.Trace(err)
}
if info.State == PROXY_STATE_OFFLINE {
log.Infof("proxy: %s offline success!", proxyName)
return nil
}
}
}
return nil
}
func main() {
fmt.Print(banner)
args, err := docopt.Parse(usage, nil, true, "codis proxy v0.1", true)
if err != nil {
fmt.Println(err)
os.Exit(1)
}
// set config file
if args["-c"] != nil {
configFile = args["-c"].(string)
}
var maxFileFrag = 10
var maxFragSize int64 = bytesize.GB * 1
if s, ok := args["--log-filesize"].(string); ok && s != "" {
v, err := bytesize.Parse(s)
if err != nil {
log.PanicErrorf(err, "invalid max log file size = %s", s)
}
maxFragSize = v
}
// set output log file
if s, ok := args["-L"].(string); ok && s != "" {
f, err := log.NewRollingFile(s, maxFileFrag, maxFragSize)
if err != nil {
log.PanicErrorf(err, "open rolling log file failed: %s", s)
} else {
defer f.Close()
log.StdLog = log.New(f, "")
}
}
log.SetLevel(log.LEVEL_INFO)
log.SetFlags(log.Flags() | log.Lshortfile)
// set log level
if s, ok := args["--log-level"].(string); ok && s != "" {
setLogLevel(s)
}
cpus = runtime.NumCPU()
// set cpu
if args["--cpu"] != nil {
cpus, err = strconv.Atoi(args["--cpu"].(string))
if err != nil {
log.PanicErrorf(err, "parse cpu number failed")
}
}
// set addr
if args["--addr"] != nil {
addr = args["--addr"].(string)
}
// set http addr
if args["--http-addr"] != nil {
httpAddr = args["--http-addr"].(string)
}
checkUlimit(1024)
runtime.GOMAXPROCS(cpus)
http.HandleFunc("/setloglevel", handleSetLogLevel)
go func() {
err := http.ListenAndServe(httpAddr, nil)
log.PanicError(err, "http debug server quit")
}()
log.Info("running on ", addr)
conf, err := proxy.LoadConf(configFile)
if err != nil {
log.PanicErrorf(err, "load config failed")
}
c := make(chan os.Signal, 1)
signal.Notify(c, os.Interrupt, syscall.SIGTERM, os.Kill)
s := proxy.New(addr, httpAddr, conf)
defer s.Close()
stats.PublishJSONFunc("router", func() string {
var m = make(map[string]interface{})
m["ops"] = router.OpCounts()
m["cmds"] = router.GetAllOpStats()
m["info"] = s.Info()
m["build"] = map[string]interface{}{
"version": utils.Version,
"compile": utils.Compile,
}
b, _ := json.Marshal(m)
return string(b)
})
go func() {
<-c
log.Info("ctrl-c or SIGTERM found, bye bye...")
s.Close()
}()
time.Sleep(time.Second)
if err := s.SetMyselfOnline(); err != nil {
log.WarnError(err, "mark myself online fail, you need mark online manually by dashboard")
}
s.Join()
log.Infof("proxy exit!! :(")
}
func (t *MigrateTask) migrateSingleSlot(slotId int, to int) error {
// set slot status
s, err := models.GetSlot(t.zkConn, t.productName, slotId)
if err != nil {
log.ErrorErrorf(err, "get slot info failed")
return err
}
if s.State.Status == models.SLOT_STATUS_OFFLINE {
log.Warnf("status is offline: %+v", s)
return nil
}
from := s.GroupId
if s.State.Status == models.SLOT_STATUS_MIGRATE {
from = s.State.MigrateStatus.From
}
// make sure from group & target group exists
exists, err := models.GroupExists(t.zkConn, t.productName, from)
if err != nil {
return errors.Trace(err)
}
if !exists {
log.Errorf("src group %d not exist when migrate from %d to %d", from, from, to)
return errors.Errorf("group %d not found", from)
}
exists, err = models.GroupExists(t.zkConn, t.productName, to)
if err != nil {
return errors.Trace(err)
}
if !exists {
return errors.Errorf("group %d not found", to)
}
// cannot migrate to itself, just ignore
if from == to {
log.Warnf("from == to, ignore: %+v", s)
return nil
}
// modify slot status
if err := s.SetMigrateStatus(t.zkConn, from, to); err != nil {
log.ErrorErrorf(err, "set migrate status failed")
return err
}
err = t.Migrate(s, from, to, func(p SlotMigrateProgress) {
// on migrate slot progress
if p.Remain%5000 == 0 {
log.Infof("%+v", p)
}
})
if err != nil {
log.ErrorErrorf(err, "migrate slot failed")
return err
}
// migrate done, change slot status back
s.State.Status = models.SLOT_STATUS_ONLINE
s.State.MigrateStatus.From = models.INVALID_ID
s.State.MigrateStatus.To = models.INVALID_ID
if err := s.Update(t.zkConn); err != nil {
log.ErrorErrorf(err, "update zk status failed, should be: %+v", s)
return err
}
return nil
}