func monitorETCD(etcdAdapter *etcdstoreadapter.ETCDStoreAdapter) {
out, err := os.Create(filepath.Join(outDir, "etcdstats.log"))
if err != nil {
logger.Fatal("etcd.log.creation.failure", err)
}
cleanup.Register(func() {
out.Sync()
})
go func() {
ticker := time.NewTicker(time.Second)
for {
<-ticker.C
t := time.Now()
logger.Info("fetch.etcd.runonce.data")
runOnceNodes, err := etcdAdapter.ListRecursively(Bbs.TaskSchemaRoot)
if err != nil {
logger.Info("fetch.etcd.runOnceNodes.error", err)
}
executorNode, err := etcdAdapter.ListRecursively(Bbs.ExecutorSchemaRoot)
if err != nil {
logger.Info("fetch.etcd.executorNode.error", err)
}
readTime := time.Since(t)
d := etcdData{
Time: float64(time.Now().UnixNano()) / 1e9,
RunningByExecutor: map[string]int{},
PresentExecutors: len(executorNode.ChildNodes),
ReadTime: float64(readTime) / 1e9,
}
for _, node := range runOnceNodes.ChildNodes {
runOnce, err := models.NewTaskFromJSON(node.Value)
if err != nil {
logger.Error("etcd.decode.runonce", err)
continue
}
switch runOnce.State {
case models.TaskStatePending:
d.Pending++
case models.TaskStateClaimed:
d.Claimed++
case models.TaskStateRunning:
d.Running++
d.RunningByExecutor[runOnce.ExecutorID]++
case models.TaskStateCompleted:
d.Completed++
}
}
logger.Info("fetched.etcd.runonce.data", time.Since(t), d.String())
out.Write(d.toJson())
out.Write([]byte("\n"))
}
}()
}
func writeInfo() {
data := fmt.Sprintf(`{ "run_onces": %d, "run_once_memory": %d } `, nTasks, taskMemory)
logger.Info("simulator.running", data)
ioutil.WriteFile(filepath.Join(outDir, "info.json"), []byte(data), 0644)
}
func runSimulation(natsClient yagnats.NATSClient) {
simulationLock = &sync.Mutex{}
simulationWait = &sync.WaitGroup{}
taskTracker = map[string]*taskData{}
msg := stagingMessage{
Count: nTasks,
MemoryMB: taskMemory,
}
payload, err := json.Marshal(msg)
if err != nil {
panic(err)
}
t := time.Now()
logger.Info("simulation.start", nTasks)
simulationWait.Add(nTasks)
_, err = natsClient.Subscribe("info.stager.*.staging-request.desire", func(msg *yagnats.Message) {
var desiredLog struct {
Timestamp time.Time `json:"_timestamp"`
Task models.Task `json:"task"`
}
err := json.Unmarshal(msg.Payload, &desiredLog)
if err != nil {
panic(err)
}
registerDesired(desiredLog.Task.Guid, desiredLog.Timestamp)
})
_, err = natsClient.Subscribe("error.>", func(msg *yagnats.Message) {
var errorLog struct {
Timestamp time.Time `json:"_timestamp"`
Error string `json:"error"`
}
err := json.Unmarshal(msg.Payload, &errorLog)
if err != nil {
panic(err)
}
registerError(msg.Subject+": "+errorLog.Error, errorLog.Timestamp)
})
_, err = natsClient.Subscribe("fatal.>", func(msg *yagnats.Message) {
var errorLog struct {
Timestamp time.Time `json:"_timestamp"`
Error string `json:"error"`
}
err := json.Unmarshal(msg.Payload, &errorLog)
if err != nil {
panic(err)
}
registerError(msg.Subject+": "+errorLog.Error, errorLog.Timestamp)
})
executorIndexes := map[string]int{}
_, err = natsClient.Subscribe("completed-task", func(msg *yagnats.Message) {
defer func() {
e := recover()
if e != nil {
logger.Error("RECOVERED PANIC:", e)
}
}()
var task *models.Task
err := json.Unmarshal(msg.Payload, &task)
if err != nil {
panic(err)
}
simulationLock.Lock()
index, ok := executorIndexes[task.ExecutorID]
if !ok {
index = len(executorIndexes) + 1
executorIndexes[task.ExecutorID] = index
}
data, ok := taskTracker[task.Guid]
if !ok {
logger.Error("uknown.runonce.completed", task.Guid, "executor", task.ExecutorID)
simulationLock.Unlock()
return
}
data.CompletionTime = float64(time.Now().UnixNano()) / 1e9
logger.Info("runonce.completed", task.Guid, "executor", task.ExecutorID, "duration", data.CompletionTime-data.DesiredTime)
data.ExecutorIndex = index
data.NumCompletions++
simulationLock.Unlock()
simulationWait.Done()
})
if err != nil {
panic(err)
}
err = natsClient.PublishWithReplyTo("stage", "completed-task", payload)
if err != nil {
panic(err)
}
cleanup.Register(func() {
dt := time.Since(t)
logger.Info("simulation.end", nTasks, "runtime", dt)
simulationResult(dt)
simulationErrors()
})
simulationWait.Wait()
}