func StartSocket() {
if !g.Config().Socket.Enable {
return
}
addr := g.Config().Socket.Listen
tcpAddr, err := net.ResolveTCPAddr("tcp", addr)
if err != nil {
log.Fatalf("net.ResolveTCPAddr fail: %s", err)
}
listener, err := net.ListenTCP("tcp", tcpAddr)
if err != nil {
log.Fatalf("listen %s fail: %s", addr, err)
} else {
log.Println("socket listening", addr)
}
defer listener.Close()
for {
conn, err := listener.Accept()
if err != nil {
log.Println("listener.Accept occur error:", err)
continue
}
go socketTelnetHandle(conn)
}
}
func StartRpc() {
if !g.Config().Rpc.Enabled {
return
}
addr := g.Config().Rpc.Listen
tcpAddr, err := net.ResolveTCPAddr("tcp", addr)
if err != nil {
log.Fatalf("net.ResolveTCPAddr fail: %s", err)
}
listener, err := net.ListenTCP("tcp", tcpAddr)
if err != nil {
log.Fatalf("listen %s fail: %s", addr, err)
} else {
log.Println("rpc listening", addr)
}
server := rpc.NewServer()
server.Register(new(Transfer))
for {
conn, err := listener.Accept()
if err != nil {
log.Println("listener.Accept occur error:", err)
continue
}
// go rpc.ServeConn(conn)
go server.ServeCodec(jsonrpc.NewServerCodec(conn))
}
}
func configCommonRoutes() {
http.HandleFunc("/health", func(w http.ResponseWriter, r *http.Request) {
w.Write([]byte("ok\n"))
})
http.HandleFunc("/version", func(w http.ResponseWriter, r *http.Request) {
w.Write([]byte(fmt.Sprintf("%s\n", g.VERSION)))
})
http.HandleFunc("/workdir", func(w http.ResponseWriter, r *http.Request) {
w.Write([]byte(fmt.Sprintf("%s\n", file.SelfDir())))
})
http.HandleFunc("/config", func(w http.ResponseWriter, r *http.Request) {
RenderDataJson(w, g.Config())
})
http.HandleFunc("/config/reload", func(w http.ResponseWriter, r *http.Request) {
if strings.HasPrefix(r.RemoteAddr, "127.0.0.1") {
g.ParseConfig(g.ConfigFile)
RenderDataJson(w, "ok")
} else {
RenderDataJson(w, "no privilege")
}
})
}
func socketTelnetHandle(conn net.Conn) {
defer conn.Close()
items := []*cmodel.MetaData{}
buf := bufio.NewReader(conn)
cfg := g.Config()
timeout := time.Duration(cfg.Socket.Timeout) * time.Second
for {
conn.SetReadDeadline(time.Now().Add(timeout))
line, err := buf.ReadString('\n')
if err != nil {
break
}
line = strings.Trim(line, "\n")
if line == "quit" {
break
}
if line == "" {
continue
}
t := strings.Fields(line)
if len(t) < 2 {
continue
}
cmd := t[0]
if cmd != "update" {
continue
}
item, err := convertLine2MetaData(t[1:])
if err != nil {
continue
}
items = append(items, item)
}
// statistics
count := int64(len(items))
pfc.Meter("SocketRecv", count)
pfc.Meter("Recv", count)
if cfg.Transfer.Enabled {
sender.Push2SendQueue(items)
}
return
}
func startHttpServer() {
if !g.Config().Http.Enable {
return
}
addr := g.Config().Http.Listen
if addr == "" {
return
}
configCommonRoutes()
configProcHttpRoutes()
configApiHttpRoutes()
s := &http.Server{
Addr: addr,
MaxHeaderBytes: 1 << 30,
}
log.Println("http.startHttpServer ok, listening", addr)
log.Fatalln(s.ListenAndServe())
}
func initConnPools() {
cfg := g.Config()
// init transfer global configs
addrs := make([]string, 0)
for hn, addr := range cfg.Transfer.Cluster {
TransferHostnames = append(TransferHostnames, hn)
addrs = append(addrs, addr)
TransferMap[hn] = addr
}
// init transfer send cnt
for hn, addr := range cfg.Transfer.Cluster {
TransferSendCnt[hn] = nproc.NewSCounterQps(hn + ":" + addr)
TransferSendFailCnt[hn] = nproc.NewSCounterQps(hn + ":" + addr)
}
// init conn pools
SenderConnPools = cpool.CreateSafeRpcConnPools(cfg.Transfer.MaxConns, cfg.Transfer.MaxIdle,
cfg.Transfer.ConnTimeout, cfg.Transfer.CallTimeout, addrs)
}
// process new metric values
func RecvMetricValues(args []*cmodel.MetricValue, reply *g.TransferResp, from string) error {
start := time.Now()
reply.ErrInvalid = 0
items := []*cmodel.MetaData{}
for _, v := range args {
if v == nil {
reply.ErrInvalid += 1
continue
}
// 历史遗留问题.
// 老版本agent上报的metric=kernel.hostname的数据,其取值为string类型,现在已经不支持了;所以,这里硬编码过滤掉
if v.Metric == "kernel.hostname" {
reply.ErrInvalid += 1
continue
}
if v.Metric == "" || v.Endpoint == "" {
reply.ErrInvalid += 1
continue
}
if v.Type != g.COUNTER && v.Type != g.GAUGE && v.Type != g.DERIVE {
reply.ErrInvalid += 1
continue
}
if v.Value == "" {
reply.ErrInvalid += 1
continue
}
if v.Step <= 0 {
reply.ErrInvalid += 1
continue
}
if len(v.Metric)+len(v.Tags) > 510 {
reply.ErrInvalid += 1
continue
}
errtags, tags := cutils.SplitTagsString(v.Tags)
if errtags != nil {
reply.ErrInvalid += 1
continue
}
// TODO 呵呵,这里需要再优雅一点
now := start.Unix()
if v.Timestamp <= 0 || v.Timestamp > now*2 {
v.Timestamp = now
}
fv := &cmodel.MetaData{
Metric: v.Metric,
Endpoint: v.Endpoint,
Timestamp: v.Timestamp,
Step: v.Step,
CounterType: v.Type,
Tags: tags, //TODO tags键值对的个数,要做一下限制
}
valid := true
var vv float64
var err error
switch cv := v.Value.(type) {
case string:
vv, err = strconv.ParseFloat(cv, 64)
if err != nil {
valid = false
}
case float64:
vv = cv
case int64:
vv = float64(cv)
default:
valid = false
}
if !valid {
reply.ErrInvalid += 1
continue
}
fv.Value = vv
items = append(items, fv)
}
// statistics
cnt := int64(len(items))
pfc.Meter("Recv", cnt)
if from == "rpc" {
pfc.Meter("RpcRecv", cnt)
} else if from == "http" {
pfc.Meter("HttpRecv", cnt)
}
cfg := g.Config()
if cfg.Transfer.Enabled {
sender.Push2SendQueue(items)
}
reply.Msg = "ok"
reply.Total = len(args)
reply.Latency = (time.Now().UnixNano() - start.UnixNano()) / 1000000
return nil
}
func forward2TransferTask(Q *nlist.SafeListLimited, concurrent int32) {
cfg := g.Config()
batch := int(cfg.Transfer.Batch)
maxConns := int64(cfg.Transfer.MaxConns)
retry := int(cfg.Transfer.Retry)
if retry < 1 {
retry = 1
}
sema := nsema.NewSemaphore(int(concurrent))
transNum := len(TransferHostnames)
for {
items := Q.PopBackBy(batch)
count := len(items)
if count == 0 {
time.Sleep(time.Millisecond * 50)
continue
}
transItems := make([]*cmodel.MetricValue, count)
for i := 0; i < count; i++ {
transItems[i] = convert(items[i].(*cmodel.MetaData))
}
sema.Acquire()
go func(transItems []*cmodel.MetricValue, count int) {
defer sema.Release()
var err error
// 随机遍历transfer列表，直到数据发送成功 或者 遍历完;随机遍历，可以缓解慢transfer
resp := &g.TransferResp{}
sendOk := false
for j := 0; j < retry && !sendOk; j++ {
rint := rand.Int()
for i := 0; i < transNum && !sendOk; i++ {
idx := (i + rint) % transNum
host := TransferHostnames[idx]
addr := TransferMap[host]
// 过滤掉建连缓慢的host, 否则会严重影响发送速率
cc := pfc.GetCounterCount(host)
if cc >= maxConns {
continue
}
pfc.Counter(host, 1)
err = SenderConnPools.Call(addr, "Transfer.Update", transItems, resp)
pfc.Counter(host, -1)
if err == nil {
sendOk = true
// statistics
TransferSendCnt[host].IncrBy(int64(count))
} else {
// statistics
TransferSendFailCnt[host].IncrBy(int64(count))
}
}
}
// statistics
if !sendOk {
if cfg.Debug {
log.Printf("send to transfer fail, connpool:%v", SenderConnPools.Proc())
}
pfc.Meter("SendFail", int64(count))
} else {
pfc.Meter("Send", int64(count))
}
}(transItems, count)
}
}
func startSendTasks() {
cfg := g.Config()
concurrent := cfg.Transfer.MaxConns * int32(len(cfg.Transfer.Cluster))
go forward2TransferTask(SenderQueue, concurrent)
}
func initConnPools() {
cfg := g.Config()
SenderConnPools = cpool.CreateSafeRpcConnPools(cfg.Transfer.MaxConns, cfg.Transfer.MaxIdle,
cfg.Transfer.ConnTimeout, cfg.Transfer.CallTimeout, []string{cfg.Transfer.Addr})
}