func NewDSStatsThreadsafe() DSStatsThreadsafe {
s := ds.NewStats()
return DSStatsThreadsafe{m: &sync.RWMutex{}, dsStats: &s}
}
//
// Kicks off the pollers and handlers
//
func Start(opsConfigFile string, staticAppData StaticAppData) {
var toSession *traffic_ops.Session
fetchSuccessCounter := gmx.NewCounter("fetchSuccess")
fetchFailCounter := gmx.NewCounter("fetchFail")
fetchPendingGauge := gmx.NewGauge("fetchPending")
tr := &http.Transport{
TLSClientConfig: &tls.Config{InsecureSkipVerify: true},
}
sharedClient := http.Client{
Timeout: defaultHttpTimeout,
Transport: tr,
}
cacheHealthConfigChannel := make(chan poller.HttpPollerConfig)
cacheHealthChannel := make(chan cache.Result)
cacheHealthTick := make(chan uint64)
cacheHealthPoller := poller.HttpPoller{
TickChan: cacheHealthTick,
ConfigChannel: cacheHealthConfigChannel,
Config: poller.HttpPollerConfig{
Interval: defaultCacheHealthPollingInterval,
},
Fetcher: fetcher.HttpFetcher{
Handler: cache.Handler{ResultChannel: cacheHealthChannel},
Client: sharedClient,
Success: fetchSuccessCounter,
Fail: fetchFailCounter,
Pending: fetchPendingGauge,
},
}
cacheStatConfigChannel := make(chan poller.HttpPollerConfig)
cacheStatChannel := make(chan cache.Result)
cacheStatPoller := poller.HttpPoller{
ConfigChannel: cacheStatConfigChannel,
Config: poller.HttpPollerConfig{
Interval: defaultCacheStatPollingInterval,
},
Fetcher: fetcher.HttpFetcher{
Handler: cache.Handler{ResultChannel: cacheStatChannel},
Client: sharedClient,
Success: fetchSuccessCounter,
Fail: fetchFailCounter,
Pending: fetchPendingGauge,
},
}
sessionChannel := make(chan *traffic_ops.Session)
monitorConfigChannel := make(chan traffic_ops.TrafficMonitorConfigMap)
monitorOpsConfigChannel := make(chan handler.OpsConfig)
monitorConfigPoller := poller.MonitorConfigPoller{
Interval: defaultMonitorConfigPollingInterval,
SessionChannel: sessionChannel,
ConfigChannel: monitorConfigChannel,
OpsConfigChannel: monitorOpsConfigChannel,
}
opsConfigFileChannel := make(chan interface{})
opsConfigFilePoller := poller.FilePoller{
File: opsConfigFile,
ResultChannel: opsConfigFileChannel,
}
opsConfigChannel := make(chan handler.OpsConfig)
opsConfigFileHandler := handler.OpsConfigFileHandler{
ResultChannel: opsConfigFilePoller.ResultChannel,
OpsConfigChannel: opsConfigChannel,
}
peerConfigChannel := make(chan poller.HttpPollerConfig)
peerChannel := make(chan peer.Result)
peerPoller := poller.HttpPoller{
ConfigChannel: peerConfigChannel,
Config: poller.HttpPollerConfig{
Interval: defaultPeerPollingInterval,
},
Fetcher: fetcher.HttpFetcher{
Handler: peer.Handler{ResultChannel: peerChannel},
Client: sharedClient,
Success: fetchSuccessCounter,
Fail: fetchFailCounter,
Pending: fetchPendingGauge,
},
}
go opsConfigFileHandler.Listen()
go opsConfigFilePoller.Poll()
go monitorConfigPoller.Poll()
go cacheHealthPoller.Poll()
go cacheStatPoller.Poll()
go peerPoller.Poll()
dr := make(chan http_server.DataRequest)
healthHistory := map[string][]interface{}{}
statHistory := map[string][]interface{}{}
var opsConfig handler.OpsConfig
var monitorConfig traffic_ops.TrafficMonitorConfigMap
localStates := peer.NewCRStates() // this is the local state as discoverer by this traffic_monitor
peerStates := map[string]peer.Crstates{} // each peer's last state is saved in this map
combinedStates := peer.NewCRStates() // this is the result of combining the localStates and all the peerStates using the var ??
deliveryServiceServers := map[string][]string{}
serverDeliveryServices := map[string]string{}
serverTypes := map[string]ds.StatCacheType{}
deliveryServiceTypes := map[string]ds.StatType{}
deliveryServiceRegexes := map[string][]string{}
serverCachegroups := map[string]string{}
// TODO put stat data in a struct, for brevity
lastHealthEndTimes := map[string]time.Time{}
lastHealthDurations := map[string]time.Duration{}
fetchCount := uint64(0) // note this is the number of individual caches fetched from, not the number of times all the caches were polled.
healthIteration := uint64(0)
errorCount := uint64(0)
events := []Event{}
eventIndex := uint64(0)
dsStats := ds.NewStats()
lastKbpsStats := ds.NewStatsLastKbps()
localCacheStatus := map[ds.CacheName]CacheAvailableStatus{}
for {
select {
case req := <-dr:
defer close(req.Response)
var body []byte
var err error
switch req.Type {
case http_server.TRConfig:
if toSession == nil {
err = fmt.Errorf("Unable to connect to Traffic Ops")
} else if opsConfig.CdnName == "" {
err = fmt.Errorf("No CDN Configured")
} else {
body, err = toSession.CRConfigRaw(opsConfig.CdnName)
}
if err != nil {
err = fmt.Errorf("TR Config: %v", err)
}
case http_server.TRStateDerived:
body, err = peer.CrStatesMarshall(combinedStates)
if err != nil {
err = fmt.Errorf("TR State (derived): %v", err)
}
case http_server.TRStateSelf:
body, err = peer.CrStatesMarshall(localStates)
if err != nil {
err = fmt.Errorf("TR State (self): %v", err)
}
case http_server.CacheStats:
// TODO: add support for ?hc=N query param, stats=, wildcard, individual caches
// add pp and date to the json:
/*
pp: "0=[my-ats-edge-cache-1], hc=[1]",
date: "Thu Oct 09 20:28:36 UTC 2014"
*/
params := req.Parameters
hc := 1
if _, exists := params["hc"]; exists {
v, err := strconv.Atoi(params["hc"][0])
if err == nil {
hc = v
}
}
body, err = cache.StatsMarshall(statHistory, hc)
if err != nil {
err = fmt.Errorf("CacheStats: %v", err)
}
case http_server.DSStats:
body, err = json.Marshal(ds.StatsJSON(dsStats)) // TODO marshall beforehand, for performance? (test to see how often requests are made)
if err != nil {
err = fmt.Errorf("DsStats: %v", err)
}
case http_server.EventLog:
body, err = json.Marshal(JSONEvents{Events: events})
if err != nil {
err = fmt.Errorf("EventLog: %v", err)
}
case http_server.PeerStates:
body, err = json.Marshal(createApiPeerStates(peerStates))
case http_server.StatSummary:
body = []byte("TODO implement")
case http_server.Stats:
body, err = getStats(staticAppData, cacheHealthPoller.Config.Interval, lastHealthDurations, fetchCount, healthIteration, errorCount)
if err != nil {
err = fmt.Errorf("Stats: %v", err)
}
case http_server.ConfigDoc:
opsConfigCopy := opsConfig
// if the password is blank, leave it blank, so callers can see it's missing.
if opsConfigCopy.Password != "" {
opsConfigCopy.Password = "******"
}
body, err = json.Marshal(opsConfigCopy)
if err != nil {
err = fmt.Errorf("Config Doc: %v", err)
}
case http_server.APICacheCount: // TODO determine if this should use peerStates
body = []byte(strconv.Itoa(len(localStates.Caches)))
case http_server.APICacheAvailableCount:
body = []byte(strconv.Itoa(cacheAvailableCount(localStates.Caches)))
case http_server.APICacheDownCount:
body = []byte(strconv.Itoa(cacheDownCount(localStates.Caches)))
case http_server.APIVersion:
s := "traffic_monitor-" + staticAppData.Version + "."
if len(staticAppData.GitRevision) > 6 {
s += staticAppData.GitRevision[:6]
} else {
s += staticAppData.GitRevision
}
body = []byte(s)
case http_server.APITrafficOpsURI:
body = []byte(opsConfig.Url)
case http_server.APICacheStates:
body, err = json.Marshal(createCacheStatuses(serverTypes, statHistory, lastHealthDurations, localStates.Caches, lastKbpsStats, localCacheStatus))
default:
err = fmt.Errorf("Unknown Request Type: %v", req.Type)
}
if err != nil {
errorCount++
log.Printf("ERROR Request Error: %v\n", err)
} else {
req.Response <- body
}
case oc := <-opsConfigFileHandler.OpsConfigChannel:
var err error
opsConfig = oc
listenAddress := ":80" // default
if opsConfig.HttpListener != "" {
listenAddress = opsConfig.HttpListener
}
handleErr := func(err error) {
errorCount++
log.Printf("%v\n", err)
}
err = http_server.Run(dr, listenAddress)
if err != nil {
handleErr(fmt.Errorf("MonitorConfigPoller: error creating HTTP server: %s\n", err))
continue
}
toSession, err = traffic_ops.Login(opsConfig.Url, opsConfig.Username, opsConfig.Password, opsConfig.Insecure)
if err != nil {
handleErr(fmt.Errorf("MonitorConfigPoller: error instantiating Session with traffic_ops: %s\n", err))
continue
}
deliveryServiceServers, serverDeliveryServices, err = getDeliveryServiceServers(toSession, opsConfig.CdnName)
if err != nil {
handleErr(fmt.Errorf("Error getting delivery service servers from Traffic Ops: %v\n", err))
continue
}
deliveryServiceTypes, err = getDeliveryServiceTypes(toSession, opsConfig.CdnName)
if err != nil {
handleErr(fmt.Errorf("Error getting delivery service types from Traffic Ops: %v\n", err))
continue
}
deliveryServiceRegexes, err = getDeliveryServiceRegexes(toSession, opsConfig.CdnName)
if err != nil {
handleErr(fmt.Errorf("Error getting delivery service regexes from Traffic Ops: %v\n", err))
continue
}
serverCachegroups, err = getServerCachegroups(toSession, opsConfig.CdnName)
if err != nil {
handleErr(fmt.Errorf("Error getting server cachegroups from Traffic Ops: %v\n", err))
continue
}
serverTypes, err = getServerTypes(toSession, opsConfig.CdnName)
if err != nil {
handleErr(fmt.Errorf("Error getting server types from Traffic Ops: %v\n", err))
continue
}
// This must be in a goroutine, because the monitorConfigPoller tick sends to a channel this select listens for. Thus, if we block on sends to the monitorConfigPoller, we have a livelock race condition.
go func() {
monitorConfigPoller.OpsConfigChannel <- opsConfig // this is needed for cdnName
monitorConfigPoller.SessionChannel <- toSession
}()
case monitorConfig = <-monitorConfigPoller.ConfigChannel:
healthUrls := map[string]string{}
statUrls := map[string]string{}
peerUrls := map[string]string{}
caches := map[string]string{}
for _, srv := range monitorConfig.TrafficServer {
caches[srv.HostName] = srv.Status
if srv.Status == "ONLINE" {
localStates.Caches[srv.HostName] = peer.IsAvailable{IsAvailable: true}
continue
}
if srv.Status == "OFFLINE" {
localStates.Caches[srv.HostName] = peer.IsAvailable{IsAvailable: false}
continue
}
// seed states with available = false until our polling cycle picks up a result
if _, exists := localStates.Caches[srv.HostName]; !exists {
localStates.Caches[srv.HostName] = peer.IsAvailable{IsAvailable: false}
}
url := monitorConfig.Profile[srv.Profile].Parameters.HealthPollingURL
r := strings.NewReplacer(
"${hostname}", srv.FQDN,
"${interface_name}", srv.InterfaceName,
"application=system", "application=plugin.remap",
"application=", "application=plugin.remap",
)
url = r.Replace(url)
healthUrls[srv.HostName] = url
r = strings.NewReplacer("application=plugin.remap", "application=")
url = r.Replace(url)
statUrls[srv.HostName] = url
}
for _, srv := range monitorConfig.TrafficMonitor {
if srv.Status != "ONLINE" {
continue
}
// TODO: the URL should be config driven. -jse
url := fmt.Sprintf("http://%s:%d/publish/CrStates?raw", srv.IP, srv.Port)
peerUrls[srv.HostName] = url
}
cacheStatPoller.ConfigChannel <- poller.HttpPollerConfig{Urls: statUrls, Interval: defaultCacheStatPollingInterval}
cacheHealthPoller.ConfigChannel <- poller.HttpPollerConfig{Urls: healthUrls, Interval: defaultCacheHealthPollingInterval}
peerPoller.ConfigChannel <- poller.HttpPollerConfig{Urls: peerUrls, Interval: defaultPeerPollingInterval}
for k := range localStates.Caches {
_, exists := monitorConfig.TrafficServer[k]
if !exists {
fmt.Printf("Warning: removing %s from localStates", k)
delete(localStates.Caches, k)
}
}
addStateDeliveryServices(monitorConfig, localStates.Deliveryservice)
case i := <-cacheHealthTick:
healthIteration = i
case healthResult := <-cacheHealthChannel:
fetchCount++
var prevResult cache.Result
if len(healthHistory[healthResult.Id]) != 0 {
prevResult = healthHistory[healthResult.Id][len(healthHistory[healthResult.Id])-1].(cache.Result)
}
health.GetVitals(&healthResult, &prevResult, &monitorConfig)
healthHistory[healthResult.Id] = pruneHistory(append(healthHistory[healthResult.Id], healthResult), defaultMaxHistory)
isAvailable, whyAvailable := health.EvalCache(healthResult, &monitorConfig)
if localStates.Caches[healthResult.Id].IsAvailable != isAvailable {
fmt.Println("Changing state for", healthResult.Id, " was:", prevResult.Available, " is now:", isAvailable, " because:", whyAvailable, " errors:", healthResult.Errors)
e := Event{Index: eventIndex, Time: time.Now().Unix(), Description: whyAvailable, Name: healthResult.Id, Hostname: healthResult.Id, Type: serverTypes[healthResult.Id].String(), Available: isAvailable}
events = append([]Event{e}, events...)
if len(events) > maxEvents {
events = events[:maxEvents-1]
}
eventIndex++
}
localCacheStatus[ds.CacheName(healthResult.Id)] = CacheAvailableStatus{Available: isAvailable, Status: monitorConfig.TrafficServer[healthResult.Id].Status} // TODO move within localStates
localStates.Caches[healthResult.Id] = peer.IsAvailable{IsAvailable: isAvailable}
calculateDeliveryServiceState(deliveryServiceServers, localStates.Caches, localStates.Deliveryservice)
// TODO determine if we should combineCrStates() here
now := time.Now()
var err error
dsStats, lastKbpsStats, err = ds.CreateStats(statHistory, deliveryServiceServers, serverDeliveryServices, deliveryServiceTypes, deliveryServiceRegexes, serverCachegroups, serverTypes, combinedStates, lastKbpsStats, now)
if err != nil {
errorCount++
log.Printf("ERROR getting deliveryservice: %v\n", err)
}
if lastHealthStart, ok := lastHealthEndTimes[healthResult.Id]; ok {
lastHealthDurations[healthResult.Id] = time.Since(lastHealthStart)
}
lastHealthEndTimes[healthResult.Id] = now
fmt.Printf("DEBUG health duration for %s: %v\n", healthResult.Id, lastHealthDurations[healthResult.Id])
// if _, ok := queryIntervalStart[pollI]; !ok {
// log.Printf("ERROR poll start index not found")
// continue
// }
// lastQueryIntervalTime = time.Since(queryIntervalStart[pollI])
case stats := <-cacheStatChannel:
statHistory[stats.Id] = pruneHistory(append(statHistory[stats.Id], stats), defaultMaxHistory)
case crStatesResult := <-peerChannel:
peerStates[crStatesResult.Id] = crStatesResult.PeerStats
combinedStates = combineCrStates(peerStates, localStates)
}
}
}