func handleCompletedRunOnce(runOnce *models.RunOnce) {
simulationLock.Lock()
data, ok := runOnceTracker[runOnce.Guid]
if !ok {
logger.Error("uknown.runonce.completed", runOnce.Guid, "executor", runOnce.ExecutorID)
simulationLock.Unlock()
return
}
data.CompletionTime = float64(time.Now().UnixNano()) / 1e9
logger.Info("runonce.completed", runOnce.Guid, "executor", runOnce.ExecutorID, "duration", data.CompletionTime-data.DesiredTime)
data.ExecutorIndex, _ = strconv.Atoi(runOnce.ExecutorID)
data.NumCompletions++
numCompletions := data.NumCompletions
simulationLock.Unlock()
if numCompletions == 1 {
defer simulationWait.Done()
logger.Info("runonce.resolve", runOnce.Guid)
err := bbs.ResolveRunOnce(runOnce)
if err != nil {
logger.Error("runonce.resolve.error", runOnce.Guid, err)
return
}
logger.Info("runonce.resolved", runOnce.Guid)
}
}
func handleRunOnce(bbs Bbs.ExecutorBBS, runOnce *models.RunOnce) {
//hesitate
logger.Info("handling.runonce", runOnce.Guid)
sleepForARandomInterval("sleep.claim", 0, 100)
//reserve memory
ok := reserveMemory(runOnce.MemoryMB)
if !ok {
logger.Info("reserve.memory.failed", runOnce.Guid)
return
}
defer releaseMemory(runOnce.MemoryMB)
//mark claimed
logger.Info("claiming.runonce", runOnce.Guid)
err := bbs.ClaimRunOnce(runOnce, *executorID)
if err != nil {
logger.Info("claim.runonce.failed", runOnce.Guid, err)
return
}
logger.Info("claimed.runonce", runOnce.Guid)
//create container
sleepForContainerCreationInterval()
//mark started
logger.Info("starting.runonce", runOnce.Guid)
err = bbs.StartRunOnce(runOnce, "container")
if err != nil {
logger.Error("start.runonce.failed", runOnce.Guid, err)
return
}
logger.Info("started.runonce", runOnce.Guid)
//run
sleepForRunInterval()
//mark completed
logger.Info("completing.runonce", runOnce.Guid)
err = bbs.CompleteRunOnce(runOnce, false, "", "")
if err != nil {
logger.Error("complete.runonce.failed", runOnce.Guid, err)
return
}
logger.Info("completed.runonce", runOnce.Guid)
}
func startAndMonitorExecutor(index int, output *os.File, ready *sync.WaitGroup) {
cmd := commandForExecutor(index, output)
logger.Info("starting.executor", index)
cmd.Start()
time.Sleep(100 * time.Millisecond) //give it a second...
ready.Done()
shuttingDown := false
cleanup.Register(func() {
shuttingDown = true
if cmd.Process != nil {
cmd.Process.Kill()
}
})
restartCount := 0
for {
err := cmd.Wait()
logger.Error("executor.exited", index, err)
if shuttingDown {
return
}
restartCount++
logger.Info("restarting.executor", index, restartCount)
cmd = commandForExecutor(index, output)
cmd.Start()
}
}
func desireAllRunOnces() {
logger.Info("desiring.runonces", nRunOnces)
dt := over / time.Duration(nRunOnces)
allDesired := &sync.WaitGroup{}
for index := 1; index <= nRunOnces; index++ {
allDesired.Add(1)
runOnce := models.RunOnce{
Guid: fmt.Sprintf("%d", index),
MemoryMB: runOnceMemory,
}
innerIndex := index
go func() {
defer allDesired.Done()
logger.Info("desiring.runonce", innerIndex)
registerDesired(innerIndex)
err := bbs.DesireRunOnce(&runOnce)
if err != nil {
logger.Error("desire.runonce.failed", innerIndex, err)
}
logger.Info("desired.runonce", innerIndex)
}()
time.Sleep(dt)
}
allDesired.Wait()
logger.Info("all.runonces.desired")
}
func convergeRunOnces(bbs Bbs.ExecutorBBS) {
statusChannel, releaseLock, err := bbs.MaintainConvergeLock(*convergenceInterval, *executorID)
if err != nil {
logger.Fatal("executor.converge-lock.acquire-failed", err)
}
tasks.Add(1)
for {
select {
case locked, ok := <-statusChannel:
if !ok {
tasks.Done()
return
}
if locked {
t := time.Now()
logger.Info("converging")
bbs.ConvergeRunOnce(*timeToClaimRunOnce)
logger.Info("converged", time.Since(t))
} else {
logger.Error("lost.convergence.lock")
}
case <-stop:
releaseLock <- nil
}
}
}
func handleRunOnces(bbs Bbs.ExecutorBBS) {
tasks.Add(1)
for {
logger.Info("watch.desired")
runOnces, stopWatching, errors := bbs.WatchForDesiredRunOnce()
INNER:
for {
select {
case runOnce, ok := <-runOnces:
if !ok {
logger.Info("watch.desired.closed")
break INNER
}
tasks.Add(1)
go func() {
handleRunOnce(bbs, runOnce)
tasks.Done()
}()
case err, ok := <-errors:
if ok && err != nil {
logger.Error("watch.desired.error", err)
}
break INNER
case <-stop:
stopWatching <- true
tasks.Done()
}
}
}
}
func (d *etcdData) toJson() []byte {
data, err := json.Marshal(d)
if err != nil {
logger.Error("etcd.marshal.etcdData.failed", err)
}
return data
}
func releaseMemory(memory int) {
lock.Lock()
defer lock.Unlock()
currentMemory = currentMemory + memory
if currentMemory > *maxMemory {
logger.Error("bookkeeping.fail", "current memory exceeds original max memory... how?")
currentMemory = *maxMemory
}
}
func monitorRunOnces(out io.Writer) {
go func() {
ticker := time.NewTicker(time.Second)
for {
<-ticker.C
t := time.Now()
logger.Info("fetch.etcd.runonce.data")
runOnceNodes, err := etcdAdapter.ListRecursively(Bbs.RunOnceSchemaRoot)
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.NewRunOnceFromJSON(node.Value)
if err != nil {
logger.Error("etcd.decode.runonce", err)
continue
}
switch runOnce.State {
case models.RunOnceStatePending:
d.Pending++
case models.RunOnceStateClaimed:
d.Claimed++
case models.RunOnceStateRunning:
d.Running++
d.RunningByExecutor[runOnce.ExecutorID]++
case models.RunOnceStateCompleted:
d.Completed++
}
}
logger.Info("fetched.etcd.runonce.data", time.Since(t), d.String())
out.Write(d.toJson())
out.Write([]byte("\n"))
}
}()
}
