remove most of the watchdog machinery

daemon/job/snapjob.go

@@ -6,16 +6,12 @@ import (
 	"github.com/prometheus/client_golang/prometheus"
 	"github.com/zrepl/zrepl/config"
 	"github.com/zrepl/zrepl/daemon/filters"
-	"github.com/zrepl/zrepl/daemon/job/reset"
 	"github.com/zrepl/zrepl/daemon/job/wakeup"
 	"github.com/zrepl/zrepl/daemon/logging"
 	"github.com/zrepl/zrepl/daemon/pruner"
 	"github.com/zrepl/zrepl/daemon/snapper"
 	"github.com/zrepl/zrepl/endpoint"
-	"github.com/zrepl/zrepl/util/envconst"
 	"github.com/zrepl/zrepl/zfs"
-	"sync"
-	"time"
 )
 
 type SnapJob struct {
@@ -27,19 +23,10 @@ type SnapJob struct {
 
 	promPruneSecs *prometheus.HistogramVec // labels: prune_side
 
-	tasksMtx sync.Mutex
-	tasks    snap_activeSideTasks
-}
-
-
-type snap_activeSideTasks struct {
-	state ActiveSideState
-
 	pruner *pruner.Pruner
-
-	prunerCancel context.CancelFunc
 }
 
+
 func (j *SnapJob) Name() string { return j.name }
 
 func (j *SnapJob) GetPruner(ctx context.Context, sender *endpoint.Sender) (*pruner.Pruner) {
@@ -48,20 +35,6 @@ func (j *SnapJob) GetPruner(ctx context.Context, sender *endpoint.Sender) (*prun
 }
 
 
-func (j *SnapJob) updateTasks(u func(*snap_activeSideTasks)) snap_activeSideTasks {
-	j.tasksMtx.Lock()
-	defer j.tasksMtx.Unlock()
-	var copy snap_activeSideTasks
-	copy = j.tasks
-	if u == nil {
-		return copy
-	}
-	u(&copy)
-	j.tasks = copy
-	return copy
-}
-
-
 func (j *SnapJob) Type() Type { return TypeSnap }
 
 func (j *SnapJob) RunPeriodic(ctx context.Context, wakeUpCommon chan <- struct{}) {
@@ -107,12 +80,12 @@ type SnapJobStatus struct {
 }
 
 func (j *SnapJob) Status() *Status {
-	tasks := j.updateTasks(nil)
+	//tasks := j.updateTasks(nil)
 
 	s := &SnapJobStatus{}
 	t := j.Type()
-	if tasks.pruner != nil {
+	if j.pruner != nil {
-		s.Pruning = tasks.pruner.Report()
+		s.Pruning = j.pruner.Report()
 	}
 	return &Status{Type: t, JobSpecific: s}
 }
@@ -142,98 +115,17 @@ outer:
 		}
 		invocationCount++
 		invLog := log.WithField("invocation", invocationCount)
-		j.do(WithLogger(ctx, invLog))
+		j.doPrune(WithLogger(ctx, invLog))
 	}
 }
 
-func (j *SnapJob) do(ctx context.Context) {
+func (j *SnapJob) doPrune(ctx context.Context) {
-
 	log := GetLogger(ctx)
 	ctx = logging.WithSubsystemLoggers(ctx, log)
-
-	// allow cancellation of an invocation (this function)
-	ctx, cancelThisRun := context.WithCancel(ctx)
-	defer cancelThisRun()
-	go func() {
-		select {
-		case <-reset.Wait(ctx):
-			log.Info("reset received, cancelling current invocation")
-			cancelThisRun()
-		case <-ctx.Done():
-		}
-	}()
-
-	// The code after this watchdog goroutine is sequential and transitions the state from
-	//   ActiveSideReplicating -> ActiveSidePruneSender -> ActiveSidePruneReceiver -> ActiveSideDone
-	// If any of those sequential tasks 'gets stuck' (livelock, no progress), the watchdog will eventually
-	// cancel its context.
-	// If the task is written to support context cancellation, it will return immediately (in permanent error state),
-	// and the sequential code above transitions to the next state.
-	go func() {
-
-		wdto := envconst.Duration("ZREPL_JOB_WATCHDOG_TIMEOUT", 10*time.Minute)
-		jitter := envconst.Duration("ZREPL_JOB_WATCHDOG_JITTER", 1*time.Second)
-		// shadowing!
-		log := log.WithField("watchdog_timeout", wdto.String())
-
-		log.Debug("starting watchdog")
-		defer log.Debug("watchdog stopped")
-
-		t := time.NewTicker(wdto)
-		defer t.Stop()
-
-		for {
-			select {
-			case <-ctx.Done():
-				return
-			case <-t.C: // fall
-			}
-
-			j.updateTasks(func(tasks *snap_activeSideTasks) {
-				// Since cancelling a task will cause the sequential code to transition to the next state immediately,
-				// we cannot check for its progress right then (no fallthrough).
-				// Instead, we return (not continue because we are in a closure) and give the new state another
-				// ZREPL_JOB_WATCHDOG_TIMEOUT interval to try make some progress.
-
-				log.WithField("state", tasks.state).Debug("watchdog firing")
-
-				const WATCHDOG_ENVCONST_NOTICE = " (adjust ZREPL_JOB_WATCHDOG_TIMEOUT env variable if inappropriate)"
-
-				switch tasks.state {
-				case ActiveSidePruneSender:
-					log.WithField("prune_sender_progress", "TEST DEBUG 123").
-						Debug("check pruner_sender progress")
-					if tasks.pruner.Progress.CheckTimeout(wdto, jitter) {
-						log.Error("pruner_sender did not make progress, cancelling" + WATCHDOG_ENVCONST_NOTICE)
-						tasks.prunerCancel()
-						return
-					}
-				case ActiveSideDone:
-					// ignore, ctx will be Done() in a few milliseconds and the watchdog will exit
-				default:
-					log.WithField("state", tasks.state).
-						Error("watchdog implementation error: unknown active side state")
-				}
-			})
-
-		}
-	}()
-
-		ctx, localCancel := context.WithCancel(ctx)
 	sender := endpoint.NewSender(j.FSFilter())
-		tasks := j.updateTasks(func(tasks *snap_activeSideTasks) {
+	j.pruner = j.GetPruner(ctx, sender)
-			tasks.pruner = j.GetPruner(ctx, sender)
+	log.Info("start pruning")
-			tasks.prunerCancel = localCancel
+	j.pruner.Prune()
-			tasks.state = ActiveSidePruneSender
+	log.Info("finished pruning")
-		})
-
-		log.Info("start pruning sender")
-		tasks.pruner.Prune()
-		log.Info("finished pruning sender")
-		localCancel()
-	j.updateTasks(func(tasks *snap_activeSideTasks) {
-		tasks.state = ActiveSideDone
-	})
-
 }