// Judge定时任务, 将 Judge发送缓存中的数据 通过rpc连接池 发送到Judge
func forward2JudgeTask(Q *list.SafeListLimited, node string, concurrent int) {
batch := g.Config().Judge.Batch // 一次发送,最多batch条数据
addr := g.Config().Judge.Cluster[node]
sema := nsema.NewSemaphore(concurrent)
for {
items := Q.PopBackBy(batch)
count := len(items)
if count == 0 {
time.Sleep(DefaultSendTaskSleepInterval)
continue
}
judgeItems := make([]*cmodel.JudgeItem, count)
for i := 0; i < count; i++ {
judgeItems[i] = items[i].(*cmodel.JudgeItem)
}
// 同步Call + 有限并发 进行发送
sema.Acquire()
go func(addr string, judgeItems []*cmodel.JudgeItem, count int) {
defer sema.Release()
resp := &cmodel.SimpleRpcResponse{}
var err error
sendOk := false
for i := 0; i < 3; i++ { //最多重试3次
err = JudgeConnPools.Call(addr, "Judge.Send", judgeItems, resp)
if err == nil {
sendOk = true
break
}
time.Sleep(time.Millisecond * 10)
}
// statistics
if !sendOk {
log.Printf("send judge %s:%s fail: %v", node, addr, err)
proc.SendToJudgeFailCnt.IncrBy(int64(count))
} else {
proc.SendToJudgeCnt.IncrBy(int64(count))
}
}(addr, judgeItems, count)
}
}
// Graph定时任务, 进行数据迁移时的 数据冗余发送
func forward2GraphMigratingTask(Q *list.SafeListLimited, node string, addr string, concurrent int) {
batch := g.Config().Graph.Batch // 一次发送,最多batch条数据
sema := nsema.NewSemaphore(concurrent)
for {
items := Q.PopBackBy(batch)
count := len(items)
if count == 0 {
time.Sleep(DefaultSendTaskSleepInterval)
continue
}
graphItems := make([]*cmodel.GraphItem, count)
for i := 0; i < count; i++ {
graphItems[i] = items[i].(*cmodel.GraphItem)
}
sema.Acquire()
go func(addr string, graphItems []*cmodel.GraphItem, count int) {
defer sema.Release()
resp := &cmodel.SimpleRpcResponse{}
var err error
sendOk := false
for i := 0; i < 3; i++ { //最多重试3次
err = GraphMigratingConnPools.Call(addr, "Graph.Send", graphItems, resp)
if err == nil {
sendOk = true
break
}
time.Sleep(time.Millisecond * 10) //发送失败了,睡10ms
}
// statistics
if !sendOk {
log.Printf("send to graph migrating %s:%s fail: %v", node, addr, err)
proc.SendToGraphMigratingFailCnt.IncrBy(int64(count))
} else {
proc.SendToGraphMigratingCnt.IncrBy(int64(count))
}
}(addr, graphItems, count)
}
}
// Tsdb定时任务, 将 Tsdb发送缓存中的数据 通过api连接池 发送到Tsdb
// 单个Cluster配置多个influxdb地址，修改为并发发送。
func forward2InfluxdbTask(Q *list.SafeListLimited, node string, concurrent int) {
cfg := g.Config()
batch := cfg.Influxdb.Batch // 一次发送,最多batch条数据
sema := nsema.NewSemaphore(concurrent * len(cfg.Influxdb.Cluster2[node].Addrs))
retry := cfg.Influxdb.MaxRetry
for {
items := Q.PopBackBy(batch)
count := len(items)
if count == 0 {
time.Sleep(DefaultSendTaskSleepInterval)
continue
}
pts := make([]*client.Point, count)
for i := 0; i < count; i++ {
pts[i] = items[i].(*client.Point)
}
for _, addr := range cfg.Influxdb.Cluster2[node].Addrs {
sema.Acquire()
go coreSend2Influxdb(addr, sema, retry, pts)
}
}
}
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)
}
}