func retryTask(c context.Context, ds appwrap.Datastore, taskIntf TaskInterface, jobKey *datastore.Key, taskKey *datastore.Key) error {
var job JobInfo
if j, err := getJob(ds, jobKey); err != nil {
return fmt.Errorf("getting job: %s", err)
} else {
job = j
}
time.Sleep(time.Duration(job.RetryCount) * 5 * time.Second)
if err := backoff.Retry(func() error {
var task JobTask
if err := ds.Get(taskKey, &task); err != nil {
return fmt.Errorf("getting task: %s", err)
}
task.Status = TaskStatusPending
if _, err := ds.Put(taskKey, &task); err != nil {
return fmt.Errorf("putting task: %s", err)
} else if err := taskIntf.PostTask(c, task.Url, job.JsonParameters); err != nil {
return fmt.Errorf("enqueuing task: %s", err)
}
logInfo(c, "retrying task %d/%d", task.Retries, job.RetryCount)
return nil
}, mrBackOff()); err != nil {
logInfo(c, "retryTask() failed after backoff attempts")
return err
} else {
return nil
}
}
func createTasks(ds appwrap.Datastore, jobKey *datastore.Key, taskKeys []*datastore.Key, tasks []JobTask, newStage JobStage) error {
now := time.Now()
firstId := taskKeys[0].IntID()
for i := range tasks {
tasks[i].StartTime = now
tasks[i].Job = jobKey
if taskKeys[i].IntID() < firstId {
firstId = taskKeys[i].IntID()
}
}
putSize := 64
i := 0
for i < len(tasks) {
if err := backoff.Retry(func() error {
last := i + putSize
if last > len(tasks) {
last = len(tasks)
}
if _, err := ds.PutMulti(taskKeys[i:last], tasks[i:last]); err != nil {
if putSize > 5 {
putSize /= 2
}
return err
}
i = last
return nil
}, mrBackOff()); err != nil {
return err
}
}
return runInTransaction(ds,
func(ds appwrap.Datastore) error {
var job JobInfo
if err := ds.Get(jobKey, &job); err != nil {
return err
}
job.TaskCount = len(tasks)
job.FirstTaskId = firstId
job.Stage = newStage
_, err := ds.Put(jobKey, &job)
return err
})
}
func updateTask(ds appwrap.Datastore, taskKey *datastore.Key, status TaskStatus, tryIncrement int, info string, result interface{}) (JobTask, error) {
var task JobTask
newCount := -1
err := backoff.Retry(func() error {
if err := ds.Get(taskKey, &task); err != nil {
return err
}
task.UpdatedAt = time.Now()
task.Info = info
// this prevents double incrementing if the Put times out but has actually
// written the value
if newCount == -1 {
newCount = task.Retries + tryIncrement
}
task.Retries = newCount
if status != "" {
task.Status = status
if status == TaskStatusDone || task.Status == TaskStatusFailed {
task.Done = task.Job
}
}
if result != nil {
resultBytes, err := json.Marshal(result)
if err != nil {
return err
}
task.Result = string(resultBytes)
}
_, err := ds.Put(taskKey, &task)
return err
}, mrBackOff())
return task, err
}
func markJobFailed(c context.Context, ds appwrap.Datastore, jobKey *datastore.Key) (prev JobInfo, finalErr error) {
finalErr = runInTransaction(ds, func(ds appwrap.Datastore) error {
prev = JobInfo{}
if err := ds.Get(jobKey, &prev); err != nil {
return err
}
job := prev
job.Stage = StageFailed
_, err := ds.Put(jobKey, &job)
return err
})
if finalErr != nil {
logCritical(c, "marking job failed for key %s failed: %s", jobKey, finalErr)
}
return
}
func createJob(ds appwrap.Datastore, urlPrefix string, writerNames []string, onCompleteUrl string, separateReduceItems bool, jsonParameters string, retryCount int) (*datastore.Key, error) {
if retryCount == 0 {
// default
retryCount = 3
}
key := ds.NewKey(JobEntity, "", 0, nil)
job := JobInfo{
UrlPrefix: urlPrefix,
Stage: StageFormation,
UpdatedAt: time.Now(),
StartTime: time.Now(),
OnCompleteUrl: onCompleteUrl,
SeparateReduceItems: separateReduceItems,
WriterNames: writerNames,
RetryCount: retryCount,
JsonParameters: jsonParameters,
}
return ds.Put(key, &job)
}
// check if the specified job has completed. it should currently be at expectedStage, and if it's been completed
// we advance it to next stage. if it's already at nextStage another process has beaten us to it so we're done
//
// caller needs to check the stage in the final job; if stageChanged is true it will be either nextStage or StageFailed.
// If StageFailed then at least one of the underlying tasks failed and the reason will appear as a taskError{} in err
func jobStageComplete(c context.Context, ds appwrap.Datastore, jobKey *datastore.Key, taskKeys []*datastore.Key, expectedStage, nextStage JobStage) (stageChanged bool, job JobInfo, finalErr error) {
last := len(taskKeys)
tasks := make([]JobTask, 100)
for last > 0 {
first := last - 100
if first < 0 {
first = 0
}
taskCount := last - first
if err := ds.GetMulti(taskKeys[first:last], tasks[0:taskCount]); err != nil {
finalErr = err
return
} else {
for i := 0; i < taskCount; i++ {
if tasks[i].Status == TaskStatusFailed {
logInfo(c, "failed tasks found")
nextStage = StageFailed
last = -1
finalErr = taskError{tasks[i].Info}
break
} else if tasks[i].Status != TaskStatusDone {
return
}
}
if last >= 0 {
last = first
}
}
}
// running this in a transaction ensures only one process advances the stage
if transErr := runInTransaction(ds, func(ds appwrap.Datastore) error {
job = JobInfo{}
if err := ds.Get(jobKey, &job); err != nil {
return err
}
if job.Stage != expectedStage {
// we're not where we expected, so advancing this isn't our responsibility
stageChanged = false
return errMonitorJobConflict
}
job.Stage = nextStage
job.UpdatedAt = time.Now()
_, err := ds.Put(jobKey, &job)
stageChanged = (err == nil)
return err
}); transErr != nil {
finalErr = transErr
}
if finalErr != nil {
logCritical(c, "taskComplete failed: %s", finalErr)
} else {
logInfo(c, "task is complete")
}
return
}