func isolate(w http.ResponseWriter, req *http.Request) {
var steps []string
vars := mux.Vars(req)
n := vars["name"]
log.WithFields(log.Fields{
"RemoteAddr": req.RemoteAddr,
"URL": req.URL,
"Name": n,
}).Info("handle `/isolate/{name}` start")
var err error
defer func() {
log.WithFields(log.Fields{
"RemoteAddr": req.RemoteAddr,
"URL": req.URL,
"Name": n,
}).Info("handle `/isolate/{name}` end")
if err != nil {
w.Write([]byte(err.Error()))
}
if len(sm.steps) != 0 {
s := ""
for _, v := range steps {
s = s + v
}
w.Write([]byte(s))
}
atomic.CompareAndSwapInt64(&sm.state, 2, 0)
}()
if !getLock(2) {
w.WriteHeader(http.StatusNotModified)
err = errors.New("http isolate get lock fail")
return
}
sm.steps = []string{}
var t string
sm.rwMutex.RLock()
_, ok := sm.proxys[n]
if ok {
t = consts.Type_Proxy
} else {
_, ok = sm.topology.DataNodeGroup["default"][n]
if ok {
status := sm.topology.DataNodeGroup["default"][n].Status
if status == consts.Normal {
t = consts.Type_Db
}
}
}
sm.rwMutex.RUnlock()
if t == consts.Type_Proxy {
err := sm.consul.PutVal(prefix+consts.ActionKey, []byte(consts.ActionIsolateProxyVal+","+n))
if err != nil {
w.WriteHeader(http.StatusInternalServerError)
return
}
err = CloseProxy(n)
if err != nil {
w.WriteHeader(http.StatusInternalServerError)
return
}
w.WriteHeader(http.StatusOK)
sm.rwMutex.RLock()
bproxys, _ := json.Marshal(sm.proxys)
sm.rwMutex.RUnlock()
err = sm.consul.PutVal(prefix+consts.ProxyKey, bproxys)
if err != nil {
log.WithFields(log.Fields{
"key": prefix + consts.ProxyKey,
"val": string(bproxys),
"err": err.Error(),
}).Error("push proxy to consul fail, continue")
}
err = presistent.ProxyPresist(string(bproxys))
if err != nil {
log.WithFields(log.Fields{
"val": string(bproxys),
"err": err.Error(),
}).Error("proxy presist fail, continue")
}
err = sm.consul.PutVal(prefix+consts.ActionKey, []byte(""))
if err != nil {
log.WithFields(log.Fields{
"key": prefix + consts.ActionKey,
"val": "",
"err": err.Error(),
}).Error("push action key to consul fail, continue")
}
} else if t == consts.Type_Db {
sm.rwMutex.RLock()
status := sm.topology.DataNodeGroup["default"][n].Status
sm.rwMutex.RUnlock()
if status != consts.Normal {
w.WriteHeader(http.StatusBadRequest)
return
}
err := sm.consul.PutVal(prefix+consts.ActionKey, []byte(consts.ActionIsolateDbVal+","+n))
if err != nil {
w.WriteHeader(http.StatusInternalServerError)
return
}
err, steps = ha.IsolateDB(n, sm.rwMutex, sm.topology, sm.swarm)
sm.steps = steps
if err != nil {
w.WriteHeader(http.StatusInternalServerError)
return
}
w.WriteHeader(http.StatusOK)
sm.rwMutex.RLock()
btopology, _ := json.Marshal(sm.topology)
sm.rwMutex.RUnlock()
err = sm.consul.PutVal(prefix+consts.TopologyKey, btopology)
if err != nil {
log.WithFields(log.Fields{
"key": prefix + consts.TopologyKey,
"val": string(btopology),
"err": err.Error(),
}).Error("push topology to consul fail, continue")
}
err = presistent.TopologyPresist(string(btopology))
if err != nil {
log.WithFields(log.Fields{
"val": string(btopology),
"err": err.Error(),
}).Error("topology presist fail, continue")
}
err = sm.consul.PutVal(prefix+consts.ActionKey, []byte(""))
if err != nil {
log.WithFields(log.Fields{
"key": prefix + consts.ActionKey,
"val": "",
"err": err.Error(),
}).Error("push action key to consul fail, continue")
}
} else {
w.WriteHeader(http.StatusBadRequest)
err = errors.New(http.StatusText(http.StatusBadRequest))
}
}
// no need lock
func NewSwitchManager(context *cli.Context) error {
config, err := NewConfig(context)
if err != nil {
return err
}
sm = &switchManager{
config: config,
rwMutex: new(sync.RWMutex),
steps: []string{},
closed: true,
}
if sm.config.Domain == "" {
return errors.New("need Domain specify")
}
if sm.config.Name == "" {
return errors.New("need Name specify")
}
if sm.config.Port == "" {
return errors.New("need Port specify")
}
if sm.config.ProxyPort == "" {
return errors.New("need ProxyPort specify")
}
prefix = sm.config.Domain + "/" + sm.config.Name + "/"
cs, err := consul.NewConsul(context)
if err != nil {
sm.steps = append(sm.steps, fmt.Sprintf("Step1. New Consul fail:%s\n", err.Error()))
return err
}
sm.steps = append(sm.steps, "Step1. New Consul success\n")
sm.consul = cs
swm, err := swarm.NewSwarm(context, cs)
if err != nil {
sm.steps = append(sm.steps, fmt.Sprintf("Step2. New Swarm fail:%s\n", err.Error()))
return err
}
sm.steps = append(sm.steps, "Step2. New Swarm success\n")
sm.swarm = swm
// init from consul
isInitalized, err := sm.consul.GetVal(prefix + consts.InitKey)
if err == nil && isInitalized == "true" {
log.Info("init from consul")
err = initProxy()
if err != nil {
log.WithFields(log.Fields{
"err:": err.Error(),
}).Error("pull proxy from consul fail, return")
sm.steps = append(sm.steps, fmt.Sprintf("Step3. Init from consul fail:%s\n", err.Error()))
return err
}
err = initTopo()
if err != nil {
log.WithFields(log.Fields{
"err:": err.Error(),
}).Error("pull topology from consul fail, continue")
sm.steps = append(sm.steps, fmt.Sprintf("Step3. Init from consul fail:%s\n", err.Error()))
return err
}
err = initSwarm()
if err != nil {
log.WithFields(log.Fields{
"err:": err.Error(),
}).Error("pull swarm from consul fail, continue")
sm.steps = append(sm.steps, fmt.Sprintf("Step3. Init from consul fail:%s\n", err.Error()))
return err
}
sm.steps = append(sm.steps, "Step3. Init from consul success\n")
} else {
sm.steps = append(sm.steps, "Step3. No init from consul\n")
}
sm.closed = false
sm.state = 0
// recover action no need lock
actionVal, err := sm.consul.GetVal(prefix + consts.ActionKey)
if actionVal != "" {
log.Info("recover action:%s\n", actionVal)
split := strings.Split(actionVal, ",")
actionVal := split[0]
name := split[1]
if actionVal == consts.ActionIsolateProxyVal {
if sm.proxys[name].Status != consts.ProxyClose {
err = CloseProxy(name)
if err != nil {
sm.steps = append(sm.steps, fmt.Sprintf("Step4. Init recover action %s fail:%s\n", actionVal, err.Error()))
return err
}
bproxys, _ := json.Marshal(sm.proxys)
err = sm.consul.PutVal(prefix+consts.ProxyKey, bproxys)
if err != nil {
log.WithFields(log.Fields{
"key": prefix + consts.ProxyKey,
"val": string(bproxys),
"err": err.Error(),
}).Error("push proxy to consul fail, continue")
}
err = presistent.ProxyPresist(string(bproxys))
if err != nil {
log.WithFields(log.Fields{
"val": string(bproxys),
"err": err.Error(),
}).Error("proxy presist fail, continue")
}
}
err := sm.consul.PutVal(prefix+consts.ActionKey, []byte(""))
if err != nil {
log.WithFields(log.Fields{
"key": prefix + consts.ActionKey,
"val": "",
"err": err.Error(),
}).Error("push action key to consul fail, continue")
}
sm.steps = append(sm.steps, "Step4. Init recover action %s success\n")
} else if actionVal == consts.ActionIsolateDbVal {
if sm.topology.DataNodeGroup["default"][name].Status != consts.Abnormal {
err, _ = ha.IsolateDB(name, sm.rwMutex, sm.topology, sm.swarm)
if err != nil {
sm.steps = append(sm.steps, fmt.Sprintf("Step4. Init recover action %s fail:%s\n", actionVal, err.Error()))
return err
}
btopology, _ := json.Marshal(sm.topology)
err = sm.consul.PutVal(prefix+consts.TopologyKey, btopology)
if err != nil {
log.WithFields(log.Fields{
"key": prefix + consts.TopologyKey,
"val": string(btopology),
"err": err.Error(),
}).Error("push topology to consul fail, continue")
}
err = presistent.TopologyPresist(string(btopology))
if err != nil {
log.WithFields(log.Fields{
"val": string(btopology),
"err": err.Error(),
}).Error("topology presist fail, continue")
}
}
err := sm.consul.PutVal(prefix+consts.ActionKey, []byte(""))
if err != nil {
log.WithFields(log.Fields{
"key": prefix + consts.ActionKey,
"val": "",
"err": err.Error(),
}).Error("push action key to consul fail, continue")
}
sm.steps = append(sm.steps, "Step4. Init recover action %s success\n")
} else if actionVal == consts.ActionRecoverDbVal {
if sm.topology.DataNodeGroup["default"][name].Status != consts.Normal {
err, _ = ha.RecoverDb(name, sm.rwMutex, sm.topology, sm.swarm)
if err != nil {
sm.steps = append(sm.steps, fmt.Sprintf("Step4. Init recover action %s fail:%s\n", actionVal, err.Error()))
return err
}
btopology, _ := json.Marshal(sm.topology)
err = sm.consul.PutVal(prefix+consts.TopologyKey, btopology)
if err != nil {
log.WithFields(log.Fields{
"key": prefix + consts.TopologyKey,
"val": string(btopology),
"err": err.Error(),
}).Error("push topology to consul fail, continue")
}
err = presistent.TopologyPresist(string(btopology))
if err != nil {
log.WithFields(log.Fields{
"val": string(btopology),
"err": err.Error(),
}).Error("topology presist fail, continue")
}
}
err := sm.consul.PutVal(prefix+consts.ActionKey, []byte(""))
if err != nil {
log.WithFields(log.Fields{
"key": prefix + consts.ActionKey,
"val": "",
"err": err.Error(),
}).Error("push action key to consul fail, continue")
}
sm.steps = append(sm.steps, "Step4. Init recover action %s success\n")
}
} else {
sm.steps = append(sm.steps, "Step4. No init recover action\n")
}
log.Info("sm init success")
return nil
}
func HealthCheck() {
m := cmap.New()
for {
if atomic.LoadInt64(&sm.state) != 0 {
time.Sleep(sm.config.HealthCheckInterval)
continue
}
sm.rwMutex.RLock()
if sm.topology == nil {
sm.rwMutex.RUnlock()
time.Sleep(sm.config.HealthCheckInterval)
continue
}
for k, v := range sm.topology.DataNodeGroup["default"] {
serviceName := k
dbName := k
if v.Status == consts.Normal {
go func() {
health, err := sm.consul.HealthCheck(serviceName)
if err != nil {
log.WithFields(log.Fields{
"dbName": dbName,
}).Error("consul health check error, direct use swarm dbcheck")
health, err = sm.swarm.DbHealthCheck(serviceName)
if err != nil {
log.WithFields(log.Fields{
"dbName": dbName,
}).Error("swarm DbHealthCheck error")
}
}
if !health || err != nil {
log.WithFields(log.Fields{
"fail times": m.GetFail(dbName),
"dbName": dbName,
}).Warn("health check fail")
if ok := m.Fail(dbName); ok {
return
}
log.Error("toggle health check auto isolate")
if getLock(1) {
sm.steps = []string{}
// maybe already isolated at before goroute?
// no need rwMutex?
if sm.topology.DataNodeGroup["default"][dbName].Status != consts.Normal {
log.WithFields(log.Fields{
"dbName": dbName,
"status": sm.topology.DataNodeGroup["default"][dbName].Status,
}).Warn("db node already isolated")
atomic.CompareAndSwapInt64(&sm.state, 1, 0)
return
}
err, steps := ha.IsolateDB(dbName, sm.rwMutex, sm.topology, sm.swarm)
sm.steps = steps
atomic.CompareAndSwapInt64(&sm.state, 1, 0)
if err != nil {
log.WithFields(log.Fields{
"dbName": dbName,
}).Error("health check auto isolate fail")
}
sm.rwMutex.RLock()
btopology, _ := json.Marshal(sm.topology)
sm.rwMutex.RUnlock()
err = sm.consul.PutVal(prefix+consts.TopologyKey, btopology)
if err != nil {
log.WithFields(log.Fields{
"key": prefix + consts.TopologyKey,
"val": string(btopology),
"err": err.Error(),
}).Error("push topology to consul fail, continue")
}
err = presistent.TopologyPresist(string(btopology))
if err != nil {
log.WithFields(log.Fields{
"val": string(btopology),
"err": err.Error(),
}).Error("topology presist fail, continue")
}
} else {
log.WithFields(log.Fields{
"dbname": dbName,
}).Error("health check auto isolate get lock fail")
}
} else {
m.Success(dbName)
}
}()
}
}
sm.rwMutex.RUnlock()
time.Sleep(sm.config.HealthCheckInterval)
}
}