func main() {
initEnv()
log.Printf("[GUARDKEEPER]Now begin to start guardkeeper...")
if *MONITOR_INTERVAL <= 0 {
log.Println("[GUARDKEEPER]the param: monitorInterval is less than 0!")
os.Exit(2)
}
for {
if !common.FileExists(signPath, signFile) {
log.Println("[GUARDKEEPER]sign file not found yet, path: " + signPath + ", filename: " + signFile + " please wait...")
time.Sleep(time.Duration(*MONITOR_INTERVAL) * time.Second)
continue
}
log.Println("[GUARDKEEPER]found the sign file!")
break
}
// hostIp := common.GetHostIp()
hostName := common.GetHostName()
realIp := common.GetShell("ifconfig eth0 | grep \"inet addr\" | awk '{print $2}' | awk -F: '{print $2}'")
// 选用--net=bridge方式的话，则进行ip替换。
// commandCore := fmt.Sprintf("s/%s\t%s/%s\t%s/g", hostIp, hostName, realIp, hostName)
// common.GetShell("sed \"" + commandCore + "\" /etc/hosts > out.tmp && cat out.tmp > /etc/hosts && rm -f out.tmp")
// 选用--net=none方式的话，那么/etc/hosts里面不会有ip\tport信息。此时直接往里面插就行。
common.GetShell(fmt.Sprintf("echo \"%s\t%s\" >> /etc/hosts", realIp, hostName))
log.Println("[GUARDKEEPER]now begin to execute supervisord!")
log.Fatalln("[GUARDKEEPER]Error...", common.GetShell("/usr/bin/supervisord"))
}
func initEnv() {
// 让命令行的命令生效
flag.Parse()
// 初始化一些初始变量
serviceIp = common.GetHostIp()
serviceName = os.Getenv("SERVICE_NAME")
common.Assert(serviceName)
servicePort = os.Getenv("SERVICE_PORT")
common.Assert(servicePort)
// 直接以容器ID + ServiceName来命名，确保唯一。如果后续发现容器ID不唯一，那么就再加上宿主机IP
// serviceId = strconv.FormatInt(time.Now().Unix(), 10) + "_" + serviceName
serviceId = common.GetHostName() + "_" + serviceName
signFile = "/home/work/data/$(hostname)/kickoff_sign_file"
isNormal = false
// 给一个channel，如果没有严重错误就永久等待
fatalExit = make(chan int)
}
func main() {
initEnv()
// 持续监听指定端口号，采用netstat -nlp | grep ":$PORT"来实现，隔5s来一次
log.Printf("[REGISTRATOR]Now begin to listen service: %s with service port: %s.", serviceName, servicePort)
if *monitorInterval <= 0 {
log.Println("[REGISTRATOR]the param: monitorInterval is less than 0!")
os.Exit(2)
}
// ticker := time.NewTicker(time.Duration(*monitorInterval) * time.Millisecond)
log.Println("[REGISTRATOR]Now begin to listen the port...")
for {
if !common.PortExists(servicePort) { // 说明启动还未成功
log.Println("[REGISTRATOR]Not found port for registration, please wait...")
time.Sleep(time.Duration(*monitorInterval) * time.Millisecond)
continue
}
log.Println("[REGISTRATOR]found the port for registration!")
isNormal = true
break
}
// 启动成功的话，就先初始化consul客户端
log.Println("[REGISTRATOR]Now begin to process registration!")
consulClient, _ = consulApi.NewClient(defaultConfig())
// 然后调用consulapi进行注册
register()
updateState(common.GetHostName(), mongo.CONTAINER_STATE_ALL_UP)
log.Println("[REGISTRATOR]Now finished registration!")
// 启动自动监控功能，当端口不在的时候，能够取消注册
ticker := time.NewTicker(time.Duration(*monitorInterval) * time.Second)
go func() {
for {
select {
case <-ticker.C:
log.Println("[REGISTRATOR]Now in a loop monitor ticket time!")
if common.PortExists(servicePort) {
if !isNormal {
log.Println("[REGISTRATOR]Now begin to register!")
isNormal = true
register()
// 注册信号量
updateState(common.GetHostName(), mongo.CONTAINER_STATE_ALL_UP)
}
} else {
if isNormal {
log.Println("[REGISTRATOR]Now begin to deregister!")
isNormal = false
deregister()
// 取消信号量。如果是容器北山，就已经至少被mask掉了。
updateState(common.GetHostName(), mongo.CONTAINER_STATE_ERROR)
}
}
}
}
}()
log.Fatalln("[REGISTRATOR]Error...", <-fatalExit)
}
func initEnv() {
flag.Parse()
signPath = "/home/work/data/" + common.GetHostName() + "/"
signFile = "kickoff_sign_file"
}