2019独角兽企业重金招聘Python工程师标准>>> 
不懂codis-ha策略的就不要使用了!!!!（否则会死很惨）
在项目中用了codis，发现官方提供的codis-ha 能够提供第一次的master-Slave切换，但是考虑，redis的Slaveof 的 同步策略，并未开放在Slave提升为master后 并未刷新其他Slave，导致新的maser没有Slave，综合本项目的考虑（项目部署环境恶劣）自己添加了一个了refreshSlave方法
servergroup.go
package main
import (
"github.com/juju/errors"
log "github.com/ngaut/logging"
"github.com/wandoulabs/codis/pkg/models"
"time"
)
func GetServerGroups() ([]models.ServerGroup, error) {
var groups []models.ServerGroup
err := callHttp(&groups, genUrl(*apiServer, "/api/server_groups"), "GET", nil)
return groups, err
}
func PingServer(checker AliveChecker, errCtx interface{}, errCh chan<- interface{}) {
err := checker.CheckAlive()
log.Debugf("check %+v, result:%v, errCtx:%+v", checker, err, errCtx)
if err != nil {
errCh <- errCtx
return
}
errCh <- nil
}
func getSlave(master *models.Server) (*models.Server, error) {
var group models.ServerGroup
err := callHttp(&group, genUrl(*apiServer, "/api/server_group/", master.GroupId), "GET", nil)
if err != nil {
return nil, errors.Trace(err)
}
for _, s := range group.Servers {
if s.Type == models.SERVER_TYPE_SLAVE {
return s, nil
}
}
return nil, errors.Errorf("can not find any slave in this group: %v", group)
}
func refreshSlave(master *models.Server) {
var group models.ServerGroup
err := callHttp(&group, genUrl(*apiServer, "/api/server_group/", master.GroupId), "GET", nil)
if err == nil {
for _, s := range group.Servers {
if s.Type == models.SERVER_TYPE_SLAVE {
err := callHttp(nil, genUrl(*apiServer, "/api/server_group/", master.GroupId,"/addServer"), "PUT", s)
if err != nil {
log.Errorf("slave refresh failed: %v error:v%", s , errors.ErrorStack(err))
}
}
}
}else{
log.Errorf("slave refresh failed: %v", errors.Trace(err))
}
}
func handleCrashedServer(s *models.Server) error {
switch s.Type {
case models.SERVER_TYPE_MASTER:
//get slave and do promote
slave, err := getSlave(s)
if err != nil {
log.Warning(errors.ErrorStack(err))
return err
}
log.Infof("try promote %+v", slave)
err = callHttp(nil, genUrl(*apiServer, "/api/server_group/", slave.GroupId, "/promote"), "POST", slave)
if err != nil {
log.Errorf("do promote %v failed %v", slave, errors.ErrorStack(err))
return err
}
refreshSlave(s) //refresh
case models.SERVER_TYPE_SLAVE:
log.Errorf("slave is down: %+v", s)
case models.SERVER_TYPE_OFFLINE:
//no need to handle it
default:
log.Fatalf("unkonwn type %+v", s)
}
return nil
}
//ping codis-server find crashed codis-server
func CheckAliveAndPromote(groups []models.ServerGroup) ([]models.Server, error) {
errCh := make(chan interface{}, 100)
var serverCnt int
for _, group := range groups { //each group
for _, s := range group.Servers { //each server
serverCnt++
rc := acf(s.Addr, 5*time.Second)
news := s
go PingServer(rc, news, errCh)
}
}
//get result
var crashedServer []models.Server
for i := 0; i < serverCnt; i++ {
s := <-errCh
if s == nil { //alive
continue
}
log.Warningf("server maybe crashed %+v", s)
crashedServer = append(crashedServer, *s.(*models.Server))
err := handleCrashedServer(s.(*models.Server))
if err != nil {
return crashedServer, err
}
}
return crashedServer, nil
}