//向monitor注册或取消注册，reg为true表示注册，否则是取消注册，返回true表示成功
func (this *HeartBeat) registerSelf(reg bool) bool {
conn, err := lib.GetConnection(this.monitorAddr)
if err != nil {
return false
}
defer conn.Close()
req := "register"
if !reg {
req = "unregister"
}
m := map[string]string{"req": req, "ip": lib.GetIp(), "port": this.port, "hostname": lib.GetHostname(), "role": this.role}
msg, err := json.Marshal(m)
if err != nil {
loglib.Error("marshal " + req + " info error " + err.Error())
return false
}
_, err = conn.Write(tcp_pack.Pack(msg))
if err != nil {
loglib.Error("send " + req + " info failed" + err.Error())
return false
} else {
plen, ret := tcp_pack.UnPack(conn)
if plen > 0 {
err = json.Unmarshal(ret, &m)
r, ok := m["err"]
if err == nil && ok && r == "0" {
loglib.Info(req + " to monitor success!")
return true
}
loglib.Error(req + " heart beat failed!")
}
}
return false
}
//goroutine
//clear list & zipping & send_to_buffer
func (r Receiver) writeList() {
//收尾工作
defer func() {
if err := recover(); err != nil {
loglib.Error(fmt.Sprintf("receiver panic:%v", err))
}
close(r.sendBuffer)
}()
st := time.Now()
var nLines = 0
var id = r.initId()
ip := lib.GetIp()
var changed = false
for logMap := range r.receiveChan {
logLine := logMap["line"]
changed = false
if logLine == "logfile changed" {
changed = true
} else {
r.logList.PushBack(logLine)
}
nLines = r.logList.Len()
//达到指定行数或发现日志rotate
//因此每小时只有最后一个包比listBufferSize小
//如果quit时包小于listBufferSize就丢弃，重启后再读
if nLines >= r.listBufferSize || changed {
hour := logMap["hour"]
repull, ok := logMap["repull"] //兼容补拉
b := r.clearList()
//r.sendBuffer <- b
ed := time.Now()
elapse := ed.Sub(st)
loglib.Info(fmt.Sprintf("add a pack, id: %s_%d, lines:%d, elapse: %s", hour, id, nLines, elapse))
//route信息
m := make(map[string]string)
m["ip"] = ip
m["hour"] = hour
m["id"] = fmt.Sprintf("%d", id)
m["lines"] = fmt.Sprintf("%d", nLines)
m["stage"] = "make pack"
m["st"] = st.Format("2006-01-02 15:04:05.000")
m["ed"] = ed.Format("2006-01-02 15:04:05.000")
m["elapse"] = elapse.String()
if ok && repull == "1" {
m["repull"] = "1"
}
if changed {
m["done"] = "1"
//这种空包用于给那些日志行数正好是listBufferSize倍数的小时标记结束
//设置repull为1以便空包能够不被拦截
if nLines == 0 {
m["repull"] = "1"
}
}
vbytes := tcp_pack.Packing(b.Bytes(), m, false)
b.Reset()
b.Write(vbytes)
r.sendBuffer <- b
id++
st = time.Now()
nLines = 0
}
if changed {
id = 1 //每小时id刷新
}
}
if nLines > 0 {
loglib.Info(fmt.Sprintf("receiver abandon %d lines", nLines))
}
}
func (t *TcpReceiver) handleConnnection(conn net.Conn, wg *sync.WaitGroup) {
defer func() {
if err := recover(); err != nil {
loglib.Error(fmt.Sprintf("tcp receiver connection panic:%v", err))
}
conn.Close()
wg.Done()
}()
/*
用于标识收到退出信号后，能否直接退出
只要接收信号时，包没有收完，都是可退出的，
发送方会缓存以后重传;
如果收完了就不能直接退出，可能包已传给下一级处理但是
却告诉发送方发送失败
*/
var quit = false //用于标识是否要退出
go lib.HandleQuitSignal(func() {
//关闭连接，避免阻塞在网络io上
conn.Close()
quit = true
})
request := make([]byte, 512*1024) //缓冲为512k
var packLen int = 0
currLen := 0
var b = new(bytes.Buffer)
var content = new(bytes.Buffer)
inAddr := conn.RemoteAddr().String()
parts := strings.Split(inAddr, ":")
inIp := parts[0]
packId := "unkown"
var routeInfo map[string]string
var rePull = false //是否补拉，如果是补拉就不做重复包检验
loglib.Info("incoming: " + inAddr)
outer:
for !quit {
st := time.Now()
if packLen == 0 {
conn.SetReadDeadline(time.Now().Add(5 * time.Minute))
time1 := time.Now() //时间打点
// read zlib pack header length
buf := make([]byte, 4)
_, err := conn.Read(buf)
if err != nil {
loglib.Warning(fmt.Sprintf("conn:%s, get header len, tcp receiver read error:%s, elapse:%s", inAddr, err.Error(), time.Now().Sub(time1)))
break
}
l, _ := binary.Uvarint(buf)
headerLen := int(l)
//get pack header
headerBuf := make([]byte, headerLen)
time2 := time.Now()
_, err = conn.Read(headerBuf)
if err != nil {
loglib.Warning(fmt.Sprintf("conn:%s, get header, tcp receiver read error:%s, elapse:%s", inAddr, err.Error(), time.Now().Sub(time2)))
break
}
//是否补拉
route0 := tcp_pack.ParseHeader(headerBuf)
if v, ok := route0["repull"]; ok && v == "1" {
rePull = true
} else {
rePull = false
}
buf = append(buf, headerBuf...)
header, _, err := tcp_pack.ExtractHeader(buf)
if err != nil {
loglib.Error("wrong format header " + string(headerBuf) + " " + err.Error())
conn.Write([]byte("wrong header"))
break
}
packId = tcp_pack.GetPackId(buf)
packLen = header.PackLen
currLen = 0
routeInfo = make(map[string]string)
b = new(bytes.Buffer)
content = new(bytes.Buffer)
loglib.Info(fmt.Sprintf("conn:%s, start receive pack %s, pack len:%d, header len:%d, header elapse:%s", inAddr, packId, packLen, headerLen, time.Now().Sub(time1)))
b.Write(buf)
routeInfo["ip"] = lib.GetIp()
routeInfo["stage"] = "tcp recv"
routeInfo["st"] = st.Format("2006-01-02 15:04:05.000")
}
//读包体的超时
conn.SetReadDeadline(time.Now().Add(5 * time.Minute))
time3 := time.Now()
//read enough bytes
for currLen < packLen {
requestLen, err := conn.Read(request)
if requestLen == 0 || err != nil {
//sender有重发机制，所以可丢弃
packLen = 0 //设为0以便读取新的包
ed := time.Now()
loglib.Warning(fmt.Sprintf("conn:%s, not full! ip:%s, packid:%s, received:%d, end recv:%s, elapse:%s, body elapse:%s, error:%s", inAddr, inIp, packId, currLen, ed, ed.Sub(st), ed.Sub(time3), err.Error()))
break outer //连接出错直接跳出外层循环
}
currLen += requestLen
content.Write(request[:requestLen])
}
if packLen > 0 && currLen >= packLen {
//收完马上应答
_, err := conn.Write([]byte("ok"))
if err != nil {
loglib.Warning(fmt.Sprintf("ip:%s, packid:%s received, but response back error:%s", inIp, packId, err.Error()))
} else {
loglib.Info(fmt.Sprintf("conn:%s, response to packid:%s", inAddr, packId))
}
//避免收到重复包（补拉例外）
appeared, ok, code := t.hasAppeared(content)
if !ok || rePull {
ed := time.Now()
routeInfo["ed"] = ed.Format("2006-01-02 15:04:05.000")
routeInfo["elapse"] = ed.Sub(st).String()
b.Write(content.Bytes())
vbytes := tcp_pack.Packing(b.Bytes(), routeInfo, true)
b = bytes.NewBuffer(vbytes)
t.buffer <- *b
packAppear := PackAppear{time.Now().Unix(), packId}
t.mutex.Lock()
t.footPrint[code] = packAppear //这里挂过
t.mutex.Unlock()
loglib.Info(fmt.Sprintf("conn:%s, finish ip:%s, packid:%s, repull:%v, received:%d, elapse:%s, body elapse:%s", inAddr, inIp, packId, rePull, currLen, ed.Sub(st), ed.Sub(time3)))
} else {
loglib.Info(fmt.Sprintf("conn:%s, pack %s repeat %s already appear at %s", inAddr, packId, appeared.Id, time.Unix(appeared.Time, 0)))
}
packLen = 0
}
}
loglib.Info("conn finish: " + inAddr)
}