client/controller: record context error and add slowlog about token b…

…ucket request Signed-off-by: Shuning Chen <[email protected]>
tikv · Jun 27, 2024 · 4e5bdb3 · 4e5bdb3
1 parent 114cb56
commit 4e5bdb3
Show file tree

Hide file tree

Showing 3 changed files with 47 additions and 27 deletions.
diff --git a/client/resource_group/controller/controller.go b/client/resource_group/controller/controller.go
@@ -38,10 +38,11 @@ import (
 )
 
 const (
-	controllerConfigPath    = "resource_group/controller"
-	maxNotificationChanLen  = 200
-	needTokensAmplification = 1.1
-	trickleReserveDuration  = 1250 * time.Millisecond
+	controllerConfigPath     = "resource_group/controller"
+	maxNotificationChanLen   = 200
+	needTokensAmplification  = 1.1
+	trickleReserveDuration   = 1250 * time.Millisecond
+	slowNotifyFilterDuration = 100 * time.Millisecond
 
 	watchRetryInterval = 30 * time.Second
 )
@@ -139,7 +140,7 @@ type ResourceGroupsController struct {
 	calculators []ResourceCalculator
 
 	// When a signal is received, it means the number of available token is low.
-	lowTokenNotifyChan chan struct{}
+	lowTokenNotifyChan chan NotifyMsg
 	// When a token bucket response received from server, it will be sent to the channel.
 	tokenResponseChan chan []*rmpb.TokenBucketResponse
 	// When the token bucket of a resource group is updated, it will be sent to the channel.
@@ -181,7 +182,7 @@ func NewResourceGroupController(
 		clientUniqueID:        clientUniqueID,
 		provider:              provider,
 		ruConfig:              ruConfig,
-		lowTokenNotifyChan:    make(chan struct{}, 1),
+		lowTokenNotifyChan:    make(chan NotifyMsg, 1),
 		tokenResponseChan:     make(chan []*rmpb.TokenBucketResponse, 1),
 		tokenBucketUpdateChan: make(chan *groupCostController, maxNotificationChanLen),
 		opts:                  opts,
@@ -287,7 +288,7 @@ func (c *ResourceGroupsController) Start(ctx context.Context) {
 				c.executeOnAllGroups((*groupCostController).updateRunState)
 				c.executeOnAllGroups((*groupCostController).updateAvgRequestResourcePerSec)
 				if len(c.run.currentRequests) == 0 {
-					c.collectTokenBucketRequests(c.loopCtx, FromPeriodReport, periodicReport /* select resource groups which should be reported periodically */)
+					c.collectTokenBucketRequests(c.loopCtx, FromPeriodReport, periodicReport /* select resource groups which should be reported periodically */, NotifyMsg{})
 				}
 			case <-watchRetryTimer.C:
 				if !c.ruConfig.isSingleGroupByKeyspace && watchMetaChannel == nil {
@@ -325,11 +326,11 @@ func (c *ResourceGroupsController) Start(ctx context.Context) {
 					c.handleTokenBucketResponse(resp)
 				}
 				c.run.currentRequests = nil
-			case <-c.lowTokenNotifyChan:
+			case notifyMsg := <-c.lowTokenNotifyChan:
 				c.executeOnAllGroups((*groupCostController).updateRunState)
 				c.executeOnAllGroups((*groupCostController).updateAvgRequestResourcePerSec)
 				if len(c.run.currentRequests) == 0 {
-					c.collectTokenBucketRequests(c.loopCtx, FromLowRU, lowToken /* select low tokens resource group */)
+					c.collectTokenBucketRequests(c.loopCtx, FromLowRU, lowToken /* select low tokens resource group */, notifyMsg)
 				}
 				if c.run.inDegradedMode {
 					c.executeOnAllGroups((*groupCostController).applyDegradedMode)
@@ -508,7 +509,7 @@ func (c *ResourceGroupsController) handleTokenBucketResponse(resp []*rmpb.TokenB
 	}
 }
 
-func (c *ResourceGroupsController) collectTokenBucketRequests(ctx context.Context, source string, typ selectType) {
+func (c *ResourceGroupsController) collectTokenBucketRequests(ctx context.Context, source string, typ selectType, notifyMsg NotifyMsg) {
 	c.run.currentRequests = make([]*rmpb.TokenBucketRequest, 0)
 	c.groupsController.Range(func(_, value any) bool {
 		gc := value.(*groupCostController)
@@ -520,11 +521,11 @@ func (c *ResourceGroupsController) collectTokenBucketRequests(ctx context.Contex
 		return true
 	})
 	if len(c.run.currentRequests) > 0 {
-		c.sendTokenBucketRequests(ctx, c.run.currentRequests, source)
+		c.sendTokenBucketRequests(ctx, c.run.currentRequests, source, notifyMsg)
 	}
 }
 
-func (c *ResourceGroupsController) sendTokenBucketRequests(ctx context.Context, requests []*rmpb.TokenBucketRequest, source string) {
+func (c *ResourceGroupsController) sendTokenBucketRequests(ctx context.Context, requests []*rmpb.TokenBucketRequest, source string, notifyMsg NotifyMsg) {
 	now := time.Now()
 	req := &rmpb.TokenBucketsRequest{
 		Requests:              requests,
@@ -542,15 +543,22 @@ func (c *ResourceGroupsController) sendTokenBucketRequests(ctx context.Context,
 		if err != nil {
 			// Don't log any errors caused by the stopper canceling the context.
 			if !errors.ErrorEqual(err, context.Canceled) {
-				log.L().Sugar().Infof("[resource group controller] token bucket rpc error: %v", err)
+				log.Info("[resource group controller] token bucket rpc error", zap.Error(err))
 			}
 			resp = nil
 			failedTokenRequestDuration.Observe(latency.Seconds())
 		} else {
 			successfulTokenRequestDuration.Observe(latency.Seconds())
 		}
+		if !notifyMsg.StartTime.IsZero() && time.Since(notifyMsg.StartTime) > slowNotifyFilterDuration {
+			log.Warn("[resource group controller] slow token bucket request", zap.String("source", source), zap.Duration("cost", time.Since(notifyMsg.StartTime)))
+		}
 		logControllerTrace("[resource group controller] token bucket response", zap.Time("now", time.Now()), zap.Any("resp", resp), zap.String("source", source), zap.Duration("latency", latency))
-		c.tokenResponseChan <- resp
+		select {
+		case c.tokenResponseChan <- resp:
+		case <-ctx.Done():
+			log.Error("[resource group controller] send token bucket response failed", zap.Error(ctx.Err()))
+		}
 	}()
 }
 
@@ -644,7 +652,7 @@ type groupCostController struct {
 	// fast path to make once token limit with un-limit burst.
 	burstable *atomic.Bool
 
-	lowRUNotifyChan       chan<- struct{}
+	lowRUNotifyChan       chan<- NotifyMsg
 	tokenBucketUpdateChan chan<- *groupCostController
 
 	// run contains the state that is updated by the main loop.
@@ -734,7 +742,7 @@ type tokenCounter struct {
 func newGroupCostController(
 	group *rmpb.ResourceGroup,
 	mainCfg *RUConfig,
-	lowRUNotifyChan chan struct{},
+	lowRUNotifyChan chan NotifyMsg,
 	tokenBucketUpdateChan chan *groupCostController,
 ) (*groupCostController, error) {
 	switch group.Mode {
@@ -853,7 +861,7 @@ func (gc *groupCostController) updateRunState() {
 	}
 	*gc.run.consumption = *gc.mu.consumption
 	gc.mu.Unlock()
-	logControllerTrace("[resource group controller] update run state", zap.Any("request-unit-consumption", gc.run.consumption))
+	logControllerTrace("[resource group controller] update run state", zap.String("name", gc.name), zap.Any("request-unit-consumption", gc.run.consumption))
 	gc.run.now = newTime
 }
 

diff --git a/client/resource_group/controller/limiter.go b/client/resource_group/controller/limiter.go
@@ -75,7 +75,7 @@ type Limiter struct {
 	// last is the last time the limiter's tokens field was updated
 	last                time.Time
 	notifyThreshold     float64
-	lowTokensNotifyChan chan<- struct{}
+	lowTokensNotifyChan chan<- NotifyMsg
 	// To prevent too many chan sent, the notifyThreshold is set to 0 after notify.
 	// So the notifyThreshold cannot show whether the limiter is in the low token state,
 	// isLowProcess is used to check it.
@@ -88,6 +88,11 @@ type Limiter struct {
 	metrics *limiterMetricsCollection
 }
 
+// NotifyMsg is a message to notify the low token state.
+type NotifyMsg struct {
+	StartTime time.Time
+}
+
 // limiterMetricsCollection is a collection of metrics for a limiter.
 type limiterMetricsCollection struct {
 	lowTokenNotifyCounter prometheus.Counter
@@ -102,7 +107,7 @@ func (lim *Limiter) Limit() Limit {
 
 // NewLimiter returns a new Limiter that allows events up to rate r and permits
 // bursts of at most b tokens.
-func NewLimiter(now time.Time, r Limit, b int64, tokens float64, lowTokensNotifyChan chan<- struct{}) *Limiter {
+func NewLimiter(now time.Time, r Limit, b int64, tokens float64, lowTokensNotifyChan chan<- NotifyMsg) *Limiter {
 	lim := &Limiter{
 		limit:               r,
 		last:                now,
@@ -116,7 +121,7 @@ func NewLimiter(now time.Time, r Limit, b int64, tokens float64, lowTokensNotify
 
 // NewLimiterWithCfg returns a new Limiter that allows events up to rate r and permits
 // bursts of at most b tokens.
-func NewLimiterWithCfg(name string, now time.Time, cfg tokenBucketReconfigureArgs, lowTokensNotifyChan chan<- struct{}) *Limiter {
+func NewLimiterWithCfg(name string, now time.Time, cfg tokenBucketReconfigureArgs, lowTokensNotifyChan chan<- NotifyMsg) *Limiter {
 	lim := &Limiter{
 		name:                name,
 		limit:               Limit(cfg.NewRate),
@@ -144,6 +149,7 @@ type Reservation struct {
 	// This is the Limit at reservation time, it can change later.
 	limit           Limit
 	remainingTokens float64
+	err             error
 }
 
 // OK returns whether the limiter can provide the requested number of tokens
@@ -218,7 +224,8 @@ func (lim *Limiter) Reserve(ctx context.Context, waitDuration time.Duration, now
 	select {
 	case <-ctx.Done():
 		return &Reservation{
-			ok: false,
+			ok:  false,
+			err: ctx.Err(),
 		}
 	default:
 	}
@@ -255,7 +262,7 @@ func (lim *Limiter) notify() {
 	lim.notifyThreshold = 0
 	lim.isLowProcess = true
 	select {
-	case lim.lowTokensNotifyChan <- struct{}{}:
+	case lim.lowTokensNotifyChan <- NotifyMsg{StartTime: time.Now()}:
 		if lim.metrics != nil {
 			lim.metrics.lowTokenNotifyCounter.Inc()
 		}
@@ -414,7 +421,9 @@ func (lim *Limiter) reserveN(now time.Time, n float64, maxFutureReserve time.Dur
 				zap.Float64("notify-threshold", lim.notifyThreshold),
 				zap.Bool("is-low-process", lim.isLowProcess),
 				zap.Int64("burst", lim.burst),
-				zap.Int("remaining-notify-times", lim.remainingNotifyTimes))
+				zap.Int("remaining-notify-times", lim.remainingNotifyTimes),
+				zap.String("name", lim.name))
+
 		}
 		lim.last = last
 		if lim.limit == 0 {
@@ -495,6 +504,9 @@ func WaitReservations(ctx context.Context, now time.Time, reservations []*Reserv
 	for _, res := range reservations {
 		if !res.ok {
 			cancel()
+			if res.err != nil {
+				return res.needWaitDuration, res.err
+			}
 			return res.needWaitDuration, errs.ErrClientResourceGroupThrottled.FastGenByArgs(res.needWaitDuration, res.limit, res.remainingTokens)
 		}
 		delay := res.DelayFrom(now)

diff --git a/client/resource_group/controller/limiter_test.go b/client/resource_group/controller/limiter_test.go
@@ -83,7 +83,7 @@ func checkTokens(re *require.Assertions, lim *Limiter, t time.Time, expected flo
 }
 
 func TestSimpleReserve(t *testing.T) {
-	lim := NewLimiter(t0, 1, 0, 2, make(chan struct{}, 1))
+	lim := NewLimiter(t0, 1, 0, 2, make(chan NotifyMsg, 1))
 
 	runReserveMax(t, lim, request{t0, 3, t1, true})
 	runReserveMax(t, lim, request{t0, 3, t4, true})
@@ -103,7 +103,7 @@ func TestSimpleReserve(t *testing.T) {
 
 func TestReconfig(t *testing.T) {
 	re := require.New(t)
-	lim := NewLimiter(t0, 1, 0, 2, make(chan struct{}, 1))
+	lim := NewLimiter(t0, 1, 0, 2, make(chan NotifyMsg, 1))
 
 	runReserveMax(t, lim, request{t0, 4, t2, true})
 	args := tokenBucketReconfigureArgs{
@@ -126,7 +126,7 @@ func TestReconfig(t *testing.T) {
 }
 
 func TestNotify(t *testing.T) {
-	nc := make(chan struct{}, 1)
+	nc := make(chan NotifyMsg, 1)
 	lim := NewLimiter(t0, 1, 0, 0, nc)
 
 	args := tokenBucketReconfigureArgs{
@@ -147,7 +147,7 @@ func TestCancel(t *testing.T) {
 	ctx := context.Background()
 	ctx1, cancel1 := context.WithDeadline(ctx, t2)
 	re := require.New(t)
-	nc := make(chan struct{}, 1)
+	nc := make(chan NotifyMsg, 1)
 	lim1 := NewLimiter(t0, 1, 0, 10, nc)
 	lim2 := NewLimiter(t0, 1, 0, 0, nc)