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 getHostname(id string, simpleCut bool) string {
if !simpleCut {
return common.GetShell("docker inspect -f {{.Config.Hostname}} " + id)
} else {
return id[:12] // 这边考虑到了destroy的话就没有hostname了，这时候就通过id进行分割获取。
}
}
func startProcessor(msg *dockerapi.APIEvents, shortId string) {
log.Println("now found a START event!")
cnt, _ := strconv.Atoi(common.GetShell("docker inspect " + shortId + " | grep SERVICE_NAME | wc -l"))
if cnt == 0 {
return
}
log.Println("Step1: begin to process container info restore")
storeContainerInfo(shortId)
log.Println("Step2: begin to process the container network")
ip := getIp(shortId)
fmt.Println("IP is: " + ip)
common.GetShell("sh " + SCRIPT_HOME + START_SCRIPT + " " + shortId + " " + ip)
log.Println("Step3: begin to process log collection")
// TODO 处理logstash、kafka topic相关脚本
}
func main() {
flag.Parse()
if os.Getenv("DOCKER_HOST") == "" {
os.Setenv("DOCKER_HOST", "unix:///var/run/docker.sock")
}
common.GetShell("mkdir -p " + SCRIPT_HOME)
common.GetShell("wget " + *ARGS_SCRIPT_WGET_HOME + START_SCRIPT + " && mv " + START_SCRIPT + " " + SCRIPT_HOME)
common.GetShell("wget " + *ARGS_SCRIPT_WGET_HOME + STOP_SCRIPT + " && mv " + STOP_SCRIPT + " " + SCRIPT_HOME)
common.GetShell("wget " + *ARGS_SCRIPT_WGET_HOME + DELETE_SCRIPT + " && mv " + DELETE_SCRIPT + " " + SCRIPT_HOME)
dockerClient, _ = dockerapi.NewClientFromEnv()
events := make(chan *dockerapi.APIEvents)
assert(dockerClient.AddEventListener(events))
log.Println("Listening for Docker events ...")
consulClient, _ = consulApi.NewClient(defaultConfig())
for msg := range events {
log.Printf("get docker event: %s now... \n", msg)
switch msg.Status {
case "start":
go startProcessor(msg, getHostname(msg.ID, true))
case "die":
go dieProcessor(msg, getHostname(msg.ID, true))
case "destroy":
go destroyProcessor(msg, getHostname(msg.ID, true))
}
}
log.Fatalln("Docker listener closed!")
// TODO 需要一个守护进程。定时扫有没有忘了删除的服务
// 此处的逻辑为： 扫所有consul注册的服务，获取虚拟IP，进一步扫本机所有的容器，
// 去进行一一比对。如果本机容器Name带有"mesos"字样，ip在上面存活的，且PID=0的，
// 则认为服务已经停止，此时删除consul上的节点数据。
// ------ 以上为删除已被停止的容器服务注册信息 -------
// 每个宿主机上各自定时扫描汇报给cloud-server自己机器上的所有存活节点。
// 由cloud-server抓取consul-server进行比对。取出不存在列表中的consul服务信息，
// 再等待下一次比对。如果下一次比对仍然没有，则判定为不存在。此时进行删除。
// ------ 以上为删除已不存在的容器服务注册信息 -------
}
func destroyProcessor(msg *dockerapi.APIEvents, shortId string) {
log.Println("now found a DESTROY event! ")
log.Println("Step1: begin to clear up all info abt this container")
// 即使没有start和stop，去调用delete脚本也不会有问题。
common.GetShell("sh " + SCRIPT_HOME + DELETE_SCRIPT + " " + shortId)
delIp(shortId)
log.Println("Step2: begin to mask container info")
maskContainerInfo(shortId)
log.Println("Step3: begin to process log collection")
// TODO
}
func dieProcessor(msg *dockerapi.APIEvents, shortId string) {
log.Println("now found a DIE event! ")
log.Println("Step1: deregister")
needDeregister := globalDeregister(msg, shortId)
if !needDeregister {
log.Println("No need to deregister. No Service Found")
return
}
log.Println("Step2: begin to clear link")
common.GetShell("sh " + SCRIPT_HOME + STOP_SCRIPT + " " + shortId)
log.Println("Step3: begin to process log collection")
}
func defaultConfig() *consulApi.Config {
// 拿ip最后一位设置为1，即为宿主机ip。默认宿主机上必须有consul
// Deprecated。采用OVS+none划分VLAN的方式重做二层网络。采用读取信号量文件来拿
// hostIp := serviceIp[:strings.LastIndex(serviceIp, ".")] + ".1:5000"
hostIp := common.GetShell("cat "+signFile+" | head -n1") + ":8500"
// hostIp := "10.14.5.14:8500"
config := &consulApi.Config{
Address: hostIp,
Scheme: "http",
HttpClient: cleanhttp.DefaultClient(),
}
return config
}