[#1210] reputation: Resolve race condition

Make all epoch independent in reputation process. Do not reset any timers
related to reputation. Make it possible to finish iteration after the
unexpected `NewEpoch` event.

Signed-off-by: Pavel Karpy <carpawell@nspcc.ru>

cmd/neofs-node/config.go

@@ -31,7 +31,6 @@ import (
 	nmClient "github.com/nspcc-dev/neofs-node/pkg/morph/client/netmap"
 	"github.com/nspcc-dev/neofs-node/pkg/morph/event"
 	netmap2 "github.com/nspcc-dev/neofs-node/pkg/morph/event/netmap"
-	"github.com/nspcc-dev/neofs-node/pkg/morph/timer"
 	"github.com/nspcc-dev/neofs-node/pkg/network"
 	"github.com/nspcc-dev/neofs-node/pkg/network/cache"
 	"github.com/nspcc-dev/neofs-node/pkg/services/control"
@@ -134,8 +133,7 @@ type cfgMorph struct {
 
 	disableCache bool
 
-	blockTimers     []*timer.BlockTimer // all combined timers
+	eigenTrustTicker *eigenTrustTickers // timers for EigenTrust iterations
-	eigenTrustTimer *timer.BlockTimer   // timer for EigenTrust iterations
 
 	proxyScriptHash neogoutil.Uint160
 }

cmd/neofs-node/main.go

@@ -93,7 +93,6 @@ func bootUp(c *cfg) {
 	makeAndWaitNotaryDeposit(c)
 	bootstrapNode(c)
 	startWorkers(c)
-	startBlockTimers(c)
 }
 
 func wait(c *cfg) {

cmd/neofs-node/reputation.go

@@ -10,6 +10,7 @@ import (
 	"github.com/nspcc-dev/neofs-node/cmd/neofs-node/reputation/common"
 	intermediatereputation "github.com/nspcc-dev/neofs-node/cmd/neofs-node/reputation/intermediate"
 	localreputation "github.com/nspcc-dev/neofs-node/cmd/neofs-node/reputation/local"
+	"github.com/nspcc-dev/neofs-node/cmd/neofs-node/reputation/ticker"
 	repClient "github.com/nspcc-dev/neofs-node/pkg/morph/client/reputation"
 	"github.com/nspcc-dev/neofs-node/pkg/morph/event"
 	"github.com/nspcc-dev/neofs-node/pkg/morph/event/netmap"
@@ -215,36 +216,40 @@ func initReputationService(c *cfg) {
 	}
 
 	// initialize eigen trust block timer
-	durationMeter := NewEigenTrustDuration(c.cfgNetmap.wrapper)
+	newEigenTrustIterTimer(c)
 
-	newEigenTrustIterTimer(c, durationMeter, func() {
+	addNewEpochAsyncNotificationHandler(
-		epoch, err := c.cfgNetmap.wrapper.Epoch()
+		c,
+		func(e event.Event) {
+			epoch := e.(netmap.NewEpoch).EpochNumber()
+
+			log := c.log.With(zap.Uint64("epoch", epoch))
+
+			duration, err := c.cfgNetmap.wrapper.EpochDuration()
 			if err != nil {
-			c.log.Debug(
+				log.Debug("could not fetch epoch duration", zap.Error(err))
-				"could not get current epoch",
-				zap.String("error", err.Error()),
-			)
-
 				return
 			}
 
+			iterations, err := c.cfgNetmap.wrapper.EigenTrustIterations()
+			if err != nil {
+				log.Debug("could not fetch iteration number", zap.Error(err))
+				return
+			}
+
+			epochTimer, err := ticker.NewIterationsTicker(duration, iterations, func() {
 				eigenTrustController.Continue(
 					eigentrustctrl.ContinuePrm{
 						Epoch: epoch - 1,
 					},
 				)
 			})
-
-	addNewEpochAsyncNotificationHandler(
-		c,
-		func(e event.Event) {
-			durationMeter.Update() // recalculate duration of one iteration round
-
-			err := c.cfgMorph.eigenTrustTimer.Reset() // start iteration rounds again
 			if err != nil {
-				c.log.Warn("can't reset block timer to start eigen trust calculations again",
+				log.Debug("could not create fixed epoch timer", zap.Error(err))
-					zap.String("error", err.Error()))
+				return
 			}
+
+			c.cfgMorph.eigenTrustTicker.addEpochTimer(epoch, epochTimer)
 		},
 	)
 }

cmd/neofs-node/timers.go

@@ -3,83 +3,41 @@ package main
 import (
 	"sync"
 
-	nmClient "github.com/nspcc-dev/neofs-node/pkg/morph/client/netmap"
+	"github.com/nspcc-dev/neofs-node/cmd/neofs-node/reputation/ticker"
-	"github.com/nspcc-dev/neofs-node/pkg/morph/timer"
 )
 
-type (
+type eigenTrustTickers struct {
-	// EigenTrustDuration is a structure that provides duration of one
+	m sync.Mutex
-	// eigen trust iteration round in blocks for block timer.
-	EigenTrustDuration struct {
-		sync.Mutex
 
-		nm  *nmClient.Client
+	timers map[uint64]*ticker.IterationsTicker
-		val uint32
-	}
-)
-
-// NewEigenTrustDuration returns instance of EigenTrustDuration.
-func NewEigenTrustDuration(nm *nmClient.Client) *EigenTrustDuration {
-	return &EigenTrustDuration{
-		nm: nm,
-	}
 }
 
-// Value returns number of blocks between two iterations of EigenTrust
+func (e *eigenTrustTickers) addEpochTimer(epoch uint64, timer *ticker.IterationsTicker) {
-// calculation. This value is not changed between `Update` calls.
+	e.m.Lock()
-func (e *EigenTrustDuration) Value() (uint32, error) {
+	defer e.m.Unlock()
-	e.Lock()
-	defer e.Unlock()
 
-	if e.val == 0 {
+	e.timers[epoch] = timer
-		e.update()
 }
 
-	return e.val, nil
+func (e *eigenTrustTickers) tick() {
-}
+	e.m.Lock()
+	defer e.m.Unlock()
 
-// Update function recalculate duration of EigenTrust iteration based on
+	for epoch, t := range e.timers {
-// NeoFS epoch duration and amount of iteration rounds from global config.
+		if !t.Tick() {
-func (e *EigenTrustDuration) Update() {
+			delete(e.timers, epoch)
-	e.Lock()
-	defer e.Unlock()
-
-	e.update()
-}
-
-func (e *EigenTrustDuration) update() {
-	iterationAmount, err := e.nm.EigenTrustIterations()
-	if err != nil {
-		return
-	}
-
-	epochDuration, err := e.nm.EpochDuration()
-	if err != nil {
-		return
-	}
-
-	e.val = uint32(epochDuration / iterationAmount)
-}
-
-func startBlockTimers(c *cfg) {
-	for i := range c.cfgMorph.blockTimers {
-		if err := c.cfgMorph.blockTimers[i].Reset(); err != nil {
-			fatalOnErr(err)
 		}
 	}
 }
 
 func tickBlockTimers(c *cfg) {
-	for i := range c.cfgMorph.blockTimers {
+	c.cfgMorph.eigenTrustTicker.tick()
-		c.cfgMorph.blockTimers[i].Tick()
-	}
 }
 
-func newEigenTrustIterTimer(c *cfg, it *EigenTrustDuration, handler timer.BlockTickHandler) {
+func newEigenTrustIterTimer(c *cfg) {
-	c.cfgMorph.eigenTrustTimer = timer.NewBlockTimer(
+	c.cfgMorph.eigenTrustTicker = &eigenTrustTickers{
-		it.Value,
+		// it is expected to have max 2 concurrent epoch
-		handler,
+		// in normal mode work
-	)
+		timers: make(map[uint64]*ticker.IterationsTicker, 2),
-
+	}
-	c.cfgMorph.blockTimers = append(c.cfgMorph.blockTimers, c.cfgMorph.eigenTrustTimer)
 }

pkg/services/reputation/eigentrust/controller/calls.go

@@ -17,6 +17,7 @@ type iterContext struct {
 
 	eigentrust.EpochIteration
 
+	iterationNumber uint32
 	last            bool
 }
 
@@ -42,11 +43,20 @@ func (c *Controller) Continue(prm ContinuePrm) {
 
 			iterCtx.Context, iterCtx.cancel = context.WithCancel(context.Background())
 			iterCtx.EpochIteration.SetEpoch(prm.Epoch)
+
+			iterations, err := c.prm.IterationsProvider.EigenTrustIterations()
+			if err != nil {
+				c.opts.log.Error("could not get EigenTrust iteration number",
+					zap.Error(err),
+				)
+			} else {
+				iterCtx.iterationNumber = uint32(iterations)
+			}
 		} else {
 			iterCtx.cancel()
 		}
 
-		iterCtx.last = iterCtx.I() == c.iterationNumber-1
+		iterCtx.last = iterCtx.I() == iterCtx.iterationNumber-1
 
 		err := c.prm.WorkerPool.Submit(func() {
 			c.prm.DaughtersTrustCalculator.Calculate(iterCtx.iterContext)
@@ -66,16 +76,6 @@ func (c *Controller) Continue(prm ContinuePrm) {
 			// number as already processed, but in any case it grows up
 			// In this case and worker pool failure we can mark epoch
 			delete(c.mCtx, prm.Epoch)
-
-			iterations, err := c.prm.IterationsProvider.EigenTrustIterations()
-			if err != nil {
-				c.opts.log.Debug(
-					"could not get iteration numbers",
-					zap.String("error", err.Error()),
-				)
-			} else {
-				c.iterationNumber = uint32(iterations)
-			}
 		}
 	}
 

pkg/services/reputation/eigentrust/controller/controller.go

@@ -46,9 +46,6 @@ type Controller struct {
 
 	opts *options
 
-	// Number of iterations
-	iterationNumber uint32
-
 	mtx  sync.Mutex
 	mCtx map[uint64]*iterContextCancel
 }
@@ -75,11 +72,6 @@ func New(prm Prm, opts ...Option) *Controller {
 		panicOnPrmValue("DaughtersTrustCalculator", prm.DaughtersTrustCalculator)
 	}
 
-	iterations, err := prm.IterationsProvider.EigenTrustIterations()
-	if err != nil {
-		panic(fmt.Errorf("could not init EigenTrust controller: could not get num of iterations: %w", err))
-	}
-
 	o := defaultOpts()
 
 	for _, opt := range opts {
@@ -87,7 +79,6 @@ func New(prm Prm, opts ...Option) *Controller {
 	}
 
 	return &Controller{
-		iterationNumber: uint32(iterations),
 		prm:  prm,
 		opts: o,
 		mCtx: make(map[uint64]*iterContextCancel),