示例#1
0
文件: worker.go 项目: zanella/nomad
// invokeScheduler is used to invoke the business logic of the scheduler
func (w *Worker) invokeScheduler(eval *structs.Evaluation, token string) error {
	defer metrics.MeasureSince([]string{"nomad", "worker", "invoke_scheduler", eval.Type}, time.Now())
	// Store the evaluation token
	w.evalToken = token

	// Snapshot the current state
	snap, err := w.srv.fsm.State().Snapshot()
	if err != nil {
		return fmt.Errorf("failed to snapshot state: %v", err)
	}

	// Store the snapshot's index
	w.snapshotIndex, err = snap.LatestIndex()
	if err != nil {
		return fmt.Errorf("failed to determine snapshot's index: %v", err)
	}

	// Create the scheduler, or use the special system scheduler
	var sched scheduler.Scheduler
	if eval.Type == structs.JobTypeCore {
		sched = NewCoreScheduler(w.srv, snap)
	} else {
		sched, err = scheduler.NewScheduler(eval.Type, w.logger, snap, w)
		if err != nil {
			return fmt.Errorf("failed to instantiate scheduler: %v", err)
		}
	}

	// Process the evaluation
	err = sched.Process(eval)
	if err != nil {
		return fmt.Errorf("failed to process evaluation: %v", err)
	}
	return nil
}
示例#2
0
文件: fsm.go 项目: zanella/nomad
// reconcileSummaries re-calculates the queued allocations for every job that we
// created a Job Summary during the snap shot restore
func (n *nomadFSM) reconcileQueuedAllocations(index uint64) error {
	// Get all the jobs
	iter, err := n.state.Jobs()
	if err != nil {
		return err
	}

	snap, err := n.state.Snapshot()
	if err != nil {
		return fmt.Errorf("unable to create snapshot: %v", err)
	}

	// Invoking the scheduler for every job so that we can populate the number
	// of queued allocations for every job
	for {
		rawJob := iter.Next()
		if rawJob == nil {
			break
		}
		job := rawJob.(*structs.Job)
		planner := &scheduler.Harness{
			State: &snap.StateStore,
		}
		// Create an eval and mark it as requiring annotations and insert that as well
		eval := &structs.Evaluation{
			ID:             structs.GenerateUUID(),
			Priority:       job.Priority,
			Type:           job.Type,
			TriggeredBy:    structs.EvalTriggerJobRegister,
			JobID:          job.ID,
			JobModifyIndex: job.JobModifyIndex + 1,
			Status:         structs.EvalStatusPending,
			AnnotatePlan:   true,
		}

		// Create the scheduler and run it
		sched, err := scheduler.NewScheduler(eval.Type, n.logger, snap, planner)
		if err != nil {
			return err
		}

		if err := sched.Process(eval); err != nil {
			return err
		}

		// Get the job summary from the fsm state store
		summary, err := n.state.JobSummaryByID(job.ID)
		if err != nil {
			return err
		}

		// Add the allocations scheduler has made to queued since these
		// allocations are never getting placed until the scheduler is invoked
		// with a real planner
		if l := len(planner.Plans); l != 1 {
			return fmt.Errorf("unexpected number of plans during restore %d. Please file an issue including the logs", l)
		}
		for _, allocations := range planner.Plans[0].NodeAllocation {
			for _, allocation := range allocations {
				tgSummary, ok := summary.Summary[allocation.TaskGroup]
				if !ok {
					return fmt.Errorf("task group %q not found while updating queued count", allocation.TaskGroup)
				}
				tgSummary.Queued += 1
				summary.Summary[allocation.TaskGroup] = tgSummary
			}
		}

		// Add the queued allocations attached to the evaluation to the queued
		// counter of the job summary
		if l := len(planner.Evals); l != 1 {
			return fmt.Errorf("unexpected number of evals during restore %d. Please file an issue including the logs", l)
		}
		for tg, queued := range planner.Evals[0].QueuedAllocations {
			tgSummary, ok := summary.Summary[tg]
			if !ok {
				return fmt.Errorf("task group %q not found while updating queued count", tg)
			}
			tgSummary.Queued += queued
			summary.Summary[tg] = tgSummary
		}

		if err := n.state.UpsertJobSummary(index, summary); err != nil {
			return err
		}
	}
	return nil
}
示例#3
0
// Plan is used to cause a dry-run evaluation of the Job and return the results
// with a potential diff containing annotations.
func (j *Job) Plan(args *structs.JobPlanRequest, reply *structs.JobPlanResponse) error {
	if done, err := j.srv.forward("Job.Plan", args, args, reply); done {
		return err
	}
	defer metrics.MeasureSince([]string{"nomad", "job", "plan"}, time.Now())

	// Validate the arguments
	if args.Job == nil {
		return fmt.Errorf("Job required for plan")
	}

	// Initialize the job fields (sets defaults and any necessary init work).
	args.Job.Canonicalize()

	// Validate the job.
	if err := validateJob(args.Job); err != nil {
		return err
	}

	// Acquire a snapshot of the state
	snap, err := j.srv.fsm.State().Snapshot()
	if err != nil {
		return err
	}

	// Get the original job
	oldJob, err := snap.JobByID(args.Job.ID)
	if err != nil {
		return err
	}

	var index uint64
	var updatedIndex uint64
	if oldJob != nil {
		index = oldJob.JobModifyIndex
		updatedIndex = oldJob.JobModifyIndex + 1
	}

	// Insert the updated Job into the snapshot
	snap.UpsertJob(updatedIndex, args.Job)

	// Create an eval and mark it as requiring annotations and insert that as well
	eval := &structs.Evaluation{
		ID:             structs.GenerateUUID(),
		Priority:       args.Job.Priority,
		Type:           args.Job.Type,
		TriggeredBy:    structs.EvalTriggerJobRegister,
		JobID:          args.Job.ID,
		JobModifyIndex: updatedIndex,
		Status:         structs.EvalStatusPending,
		AnnotatePlan:   true,
	}

	// Create an in-memory Planner that returns no errors and stores the
	// submitted plan and created evals.
	planner := &scheduler.Harness{
		State: &snap.StateStore,
	}

	// Create the scheduler and run it
	sched, err := scheduler.NewScheduler(eval.Type, j.srv.logger, snap, planner)
	if err != nil {
		return err
	}

	if err := sched.Process(eval); err != nil {
		return err
	}

	// Annotate and store the diff
	if plans := len(planner.Plans); plans != 1 {
		return fmt.Errorf("scheduler resulted in an unexpected number of plans: %v", plans)
	}
	annotations := planner.Plans[0].Annotations
	if args.Diff {
		jobDiff, err := oldJob.Diff(args.Job, true)
		if err != nil {
			return fmt.Errorf("failed to create job diff: %v", err)
		}

		if err := scheduler.Annotate(jobDiff, annotations); err != nil {
			return fmt.Errorf("failed to annotate job diff: %v", err)
		}
		reply.Diff = jobDiff
	}

	// Grab the failures
	if len(planner.Evals) != 1 {
		return fmt.Errorf("scheduler resulted in an unexpected number of eval updates: %v", planner.Evals)
	}
	updatedEval := planner.Evals[0]

	// If it is a periodic job calculate the next launch
	if args.Job.IsPeriodic() && args.Job.Periodic.Enabled {
		reply.NextPeriodicLaunch = args.Job.Periodic.Next(time.Now().UTC())
	}

	reply.FailedTGAllocs = updatedEval.FailedTGAllocs
	reply.JobModifyIndex = index
	reply.Annotations = annotations
	reply.CreatedEvals = planner.CreateEvals
	reply.Index = index
	return nil
}