introduce final seats for work estimate

staging/src/k8s.io/apiserver/pkg/server/filters/priority-and-fairness_test.go

@@ -654,6 +654,8 @@ func TestApfWithRequestDigest(t *testing.T) {
 		User:        &user.DefaultInfo{Name: "foo"},
 		WorkEstimate: fcrequest.WorkEstimate{
 			InitialSeats:      5,
+			FinalSeats:        7,
+			AdditionalLatency: 3 * time.Second,
 		},
 	}
 

staging/src/k8s.io/apiserver/pkg/util/flowcontrol/fairqueuing/queueset/fifo_list.go

@@ -51,9 +51,9 @@ type fifo interface {
 	// Length returns the number of requests in the list.
 	Length() int
 
-	// SeatsSum returns the total number of seats of all requests
-	// in this list.
-	SeatsSum() int
+	// QueueSum returns the sum of initial seats, final seats, and
+	// additional latency aggregated from all requests in this queue.
+	QueueSum() queueSum
 
 	// Walk iterates through the list in order of oldest -> newest
 	// and executes the specified walkFunc for each request in that order.
@@ -68,7 +68,7 @@ type fifo interface {
 type requestFIFO struct {
 	*list.List
 
-	seatsSum int
+	sum queueSum
 }
 
 func newRequestFIFO() fifo {
@@ -81,19 +81,19 @@ func (l *requestFIFO) Length() int {
 	return l.Len()
 }
 
-func (l *requestFIFO) SeatsSum() int {
-	return l.seatsSum
+func (l *requestFIFO) QueueSum() queueSum {
+	return l.sum
 }
 
 func (l *requestFIFO) Enqueue(req *request) removeFromFIFOFunc {
 	e := l.PushBack(req)
-	l.seatsSum += req.Seats()
+	addToQueueSum(&l.sum, req)
 
 	return func() *request {
 		if e.Value != nil {
 			l.Remove(e)
 			e.Value = nil
-			l.seatsSum -= req.Seats()
+			deductFromQueueSum(&l.sum, req)
 		}
 		return req
 	}
@@ -122,7 +122,7 @@ func (l *requestFIFO) getFirst(remove bool) (*request, bool) {
 
 	request, ok := e.Value.(*request)
 	if remove && ok {
-		l.seatsSum -= request.Seats()
+		deductFromQueueSum(&l.sum, request)
 	}
 	return request, ok
 }
@@ -136,3 +136,15 @@ func (l *requestFIFO) Walk(f walkFunc) {
 		}
 	}
 }
+
+func addToQueueSum(sum *queueSum, req *request) {
+	sum.InitialSeatsSum += req.InitialSeats()
+	sum.MaxSeatsSum += req.MaxSeats()
+	sum.AdditionalSeatSecondsSum += req.AdditionalSeatSeconds()
+}
+
+func deductFromQueueSum(sum *queueSum, req *request) {
+	sum.InitialSeatsSum -= req.InitialSeats()
+	sum.MaxSeatsSum -= req.MaxSeats()
+	sum.AdditionalSeatSecondsSum -= req.AdditionalSeatSeconds()
+}

staging/src/k8s.io/apiserver/pkg/util/flowcontrol/fairqueuing/queueset/fifo_list_test.go

@@ -18,9 +18,11 @@ package queueset
 
 import (
 	"math/rand"
+	"reflect"
 	"testing"
 	"time"
 
+	"github.com/google/go-cmp/cmp"
 	fcrequest "k8s.io/apiserver/pkg/util/flowcontrol/request"
 )
 
@@ -150,59 +152,88 @@ func TestFIFOWithRemoveIsIdempotent(t *testing.T) {
 	verifyOrder(t, orderExpected, remainingRequests)
 }
 
-func TestFIFOSeatsSum(t *testing.T) {
+func TestFIFOQueueWorkEstimate(t *testing.T) {
 	list := newRequestFIFO()
 
-	newRequest := func(width uint) *request {
-		return &request{workEstimate: fcrequest.WorkEstimate{InitialSeats: width}}
+	update := func(we *queueSum, req *request, multiplier int) {
+		we.InitialSeatsSum += multiplier * req.InitialSeats()
+		we.MaxSeatsSum += multiplier * req.MaxSeats()
+		we.AdditionalSeatSecondsSum += SeatSeconds(multiplier) * req.AdditionalSeatSeconds()
+	}
+
+	assert := func(t *testing.T, want, got *queueSum) {
+		if !reflect.DeepEqual(want, got) {
+			t.Errorf("Expected queue work estimate to match, diff: %s", cmp.Diff(want, got))
+		}
+	}
+
+	newRequest := func(initialSeats, finalSeats uint, additionalLatency time.Duration) *request {
+		return &request{workEstimate: fcrequest.WorkEstimate{
+			InitialSeats:      initialSeats,
+			FinalSeats:        finalSeats,
+			AdditionalLatency: additionalLatency,
+		}}
+	}
+	arrival := []*request{
+		newRequest(1, 3, time.Second),
+		newRequest(2, 2, 2*time.Second),
+		newRequest(3, 1, 3*time.Second),
 	}
-	arrival := []*request{newRequest(1), newRequest(2), newRequest(3)}
 	removeFn := make([]removeFromFIFOFunc, 0)
 
-	seatsSum := 0
+	queueSumExpected := queueSum{}
 	for i := range arrival {
-		removeFn = append(removeFn, list.Enqueue(arrival[i]))
+		req := arrival[i]
+		removeFn = append(removeFn, list.Enqueue(req))
+		update(&queueSumExpected, req, 1)
 
-		seatsSum += i + 1
-		if list.SeatsSum() != seatsSum {
-			t.Errorf("Expected seatsSum: %d, but got: %d", seatsSum, list.SeatsSum())
-		}
+		workEstimateGot := list.QueueSum()
+		assert(t, &queueSumExpected, &workEstimateGot)
 	}
 
+	// NOTE: the test expects the request and the remove func to be at the same index
 	for i := range removeFn {
+		req := arrival[i]
 		removeFn[i]()
 
-		seatsSum -= i + 1
-		if list.SeatsSum() != seatsSum {
-			t.Errorf("Expected seatsSum: %d, but got: %d", seatsSum, list.SeatsSum())
-		}
+		update(&queueSumExpected, req, -1)
+
+		workEstimateGot := list.QueueSum()
+		assert(t, &queueSumExpected, &workEstimateGot)
 
 		// check idempotency
 		removeFn[i]()
-		if list.SeatsSum() != seatsSum {
-			t.Errorf("Expected seatsSum: %d, but got: %d", seatsSum, list.SeatsSum())
-		}
+
+		workEstimateGot = list.QueueSum()
+		assert(t, &queueSumExpected, &workEstimateGot)
 	}
 
 	// Check second type of idempotency: Dequeue + removeFn.
 	for i := range arrival {
-		removeFn[i] = list.Enqueue(arrival[i])
-		seatsSum += i + 1
+		req := arrival[i]
+		removeFn[i] = list.Enqueue(req)
+
+		update(&queueSumExpected, req, 1)
 	}
 
 	for i := range arrival {
+		// we expect Dequeue to pop the oldest request that should
+		// have the lowest index as well.
+		req := arrival[i]
+
 		if _, ok := list.Dequeue(); !ok {
 			t.Errorf("Unexpected failed dequeue: %d", i)
 		}
-		seatsSum -= i + 1
-		if list.SeatsSum() != seatsSum {
-			t.Errorf("Expected seatsSum: %d, but got: %d", seatsSum, list.SeatsSum())
-		}
+
+		update(&queueSumExpected, req, -1)
+
+		queueSumGot := list.QueueSum()
+		assert(t, &queueSumExpected, &queueSumGot)
 
 		removeFn[i]()
-		if list.SeatsSum() != seatsSum {
-			t.Errorf("Expected seatsSum: %d, but got: %d", seatsSum, list.SeatsSum())
-		}
+
+		queueSumGot = list.QueueSum()
+		assert(t, &queueSumExpected, &queueSumGot)
 	}
 }
 

staging/src/k8s.io/apiserver/pkg/util/flowcontrol/fairqueuing/queueset/queueset.go

@@ -268,7 +268,7 @@ func (qs *queueSet) StartRequest(ctx context.Context, workEstimate *fqrequest.Wo
 	// Step 0:
 	// Apply only concurrency limit, if zero queues desired
 	if qs.qCfg.DesiredNumQueues < 1 {
-		if !qs.canAccommodateSeatsLocked(int(workEstimate.InitialSeats)) {
+		if !qs.canAccommodateSeatsLocked(workEstimate.MaxSeats()) {
 			klog.V(5).Infof("QS(%s): rejecting request %q %#+v %#+v because %d seats are asked for, %d seats are in use (%d are executing) and the limit is %d",
 				qs.qCfg.Name, fsName, descr1, descr2, workEstimate, qs.totSeatsInUse, qs.totRequestsExecuting, qs.dCfg.ConcurrencyLimit)
 			metrics.AddReject(ctx, qs.qCfg.Name, fsName, "concurrency-limit")
@@ -315,11 +315,28 @@ func ordinaryPromiseFactoryFactory(qs *queueSet) promiseFactory {
 	return promise.NewWriteOnce
 }
 
-// Seats returns the number of seats this request requires.
-func (req *request) Seats() int {
+// MaxSeats returns the maximum number of seats this request requires, it is
+// the maxumum of the two - WorkEstimate.InitialSeats, WorkEstimate.FinalSeats.
+func (req *request) MaxSeats() int {
+	return req.workEstimate.MaxSeats()
+}
+
+func (req *request) InitialSeats() int {
 	return int(req.workEstimate.InitialSeats)
 }
 
+// AdditionalSeatSeconds returns the amount of work in SeatSeconds produced by
+// the final seats and the additional latency associated with a request.
+func (req *request) AdditionalSeatSeconds() SeatSeconds {
+	return SeatsTimesDuration(float64(req.workEstimate.FinalSeats), req.workEstimate.AdditionalLatency)
+}
+
+// InitialSeatSeconds returns the amount of work in SeatSeconds projected
+// by the initial seats for a given estimated service duration.
+func (req *request) InitialSeatSeconds(estimatedServiceDuration time.Duration) SeatSeconds {
+	return SeatsTimesDuration(float64(req.workEstimate.InitialSeats), estimatedServiceDuration)
+}
+
 func (req *request) NoteQueued(inQueue bool) {
 	if req.queueNoteFn != nil {
 		req.queueNoteFn(inQueue)
@@ -423,7 +440,9 @@ func (qs *queueSet) getVirtualTimeRatioLocked() float64 {
 	activeQueues := 0
 	seatsRequested := 0
 	for _, queue := range qs.queues {
-		seatsRequested += (queue.seatsInUse + queue.requests.SeatsSum())
+		// here we want the sum of the maximum width of the requests in this queue since our
+		// goal is to find the maximum rate at which the queue could work.
+		seatsRequested += (queue.seatsInUse + queue.requests.QueueSum().MaxSeatsSum)
 		if queue.requests.Length() > 0 || queue.requestsExecuting > 0 {
 			activeQueues++
 		}
@@ -489,25 +508,24 @@ func (qs *queueSet) shuffleShardLocked(hashValue uint64, descr1, descr2 interfac
 	offset := qs.enqueues % handSize
 	qs.enqueues++
 	bestQueueIdx := -1
-	bestQueueSeatsSum := int(math.MaxInt32)
+	minQueueSeatSeconds := MaxSeatSeconds
 	for i := 0; i < handSize; i++ {
 		queueIdx := hand[(offset+i)%handSize]
 		queue := qs.queues[queueIdx]
-		waitingSeats := queue.requests.SeatsSum()
-		// TODO: Consider taking into account `additional latency` of requests
-		// in addition to their seats.
-		// Ideally, this should be based on projected completion time in the
-		// virtual world of the youngest request in the queue.
-		thisSeatsSum := waitingSeats + queue.seatsInUse
-		klog.V(7).Infof("QS(%s): For request %#+v %#+v considering queue %d of %d seats waiting and %d executing, nextDispatchR=%v", qs.qCfg.Name, descr1, descr2, queueIdx, waitingSeats, queue.seatsInUse, queue.nextDispatchR)
-		if thisSeatsSum < bestQueueSeatsSum {
-			bestQueueIdx, bestQueueSeatsSum = queueIdx, thisSeatsSum
-		}
+		queueSum := queue.requests.QueueSum()
 
+		// this is the total amount of work in seat-seconds for requests
+		// waiting in this queue, we will select the queue with the minimum.
+		thisQueueSeatSeconds := SeatsTimesDuration(float64(queueSum.InitialSeatsSum), qs.estimatedServiceDuration) + queueSum.AdditionalSeatSecondsSum
+		klog.V(7).Infof("QS(%s): For request %#+v %#+v considering queue %d with sum: %#v and %d seats in use, nextDispatchR=%v", qs.qCfg.Name, descr1, descr2, queueIdx, queueSum, queue.seatsInUse, queue.nextDispatchR)
+		if thisQueueSeatSeconds < minQueueSeatSeconds {
+			minQueueSeatSeconds = thisQueueSeatSeconds
+			bestQueueIdx = queueIdx
+		}
 	}
 	if klog.V(6).Enabled() {
 		chosenQueue := qs.queues[bestQueueIdx]
-		klog.V(6).Infof("QS(%s) at t=%s R=%v: For request %#+v %#+v chose queue %d, had seatSum %d & %d requests executing & virtualStart=%v", qs.qCfg.Name, qs.clock.Now().Format(nsTimeFmt), qs.currentR, descr1, descr2, bestQueueIdx, chosenQueue.requests.SeatsSum(), chosenQueue.requestsExecuting, chosenQueue.nextDispatchR)
+		klog.V(6).Infof("QS(%s) at t=%s R=%v: For request %#+v %#+v chose queue %d, with sum: %#v & %d seats in use & nextDispatchR=%v", qs.qCfg.Name, qs.clock.Now().Format(nsTimeFmt), qs.currentR, descr1, descr2, bestQueueIdx, chosenQueue.requests.QueueSum(), chosenQueue.seatsInUse, chosenQueue.nextDispatchR)
 	}
 	return bestQueueIdx
 }
@@ -619,9 +637,9 @@ func (qs *queueSet) dispatchSansQueueLocked(ctx context.Context, workEstimate *f
 		workEstimate:      *workEstimate,
 	}
 	qs.totRequestsExecuting++
-	qs.totSeatsInUse += req.Seats()
+	qs.totSeatsInUse += req.MaxSeats()
 	metrics.AddRequestsExecuting(ctx, qs.qCfg.Name, fsName, 1)
-	metrics.AddRequestConcurrencyInUse(qs.qCfg.Name, fsName, req.Seats())
+	metrics.AddRequestConcurrencyInUse(qs.qCfg.Name, fsName, req.MaxSeats())
 	qs.obsPair.RequestsExecuting.Add(1)
 	if klog.V(5).Enabled() {
 		klog.Infof("QS(%s) at t=%s R=%v: immediate dispatch of request %q %#+v %#+v, qs will have %d executing", qs.qCfg.Name, now.Format(nsTimeFmt), qs.currentR, fsName, descr1, descr2, qs.totRequestsExecuting)
@@ -650,13 +668,13 @@ func (qs *queueSet) dispatchLocked() bool {
 	// problem because other overhead is also included.
 	qs.totRequestsWaiting--
 	qs.totRequestsExecuting++
-	qs.totSeatsInUse += request.Seats()
+	qs.totSeatsInUse += request.MaxSeats()
 	queue.requestsExecuting++
-	queue.seatsInUse += request.Seats()
+	queue.seatsInUse += request.MaxSeats()
 	metrics.AddRequestsInQueues(request.ctx, qs.qCfg.Name, request.fsName, -1)
 	request.NoteQueued(false)
 	metrics.AddRequestsExecuting(request.ctx, qs.qCfg.Name, request.fsName, 1)
-	metrics.AddRequestConcurrencyInUse(qs.qCfg.Name, request.fsName, request.Seats())
+	metrics.AddRequestConcurrencyInUse(qs.qCfg.Name, request.fsName, request.MaxSeats())
 	qs.obsPair.RequestsWaiting.Add(-1)
 	qs.obsPair.RequestsExecuting.Add(1)
 	if klog.V(6).Enabled() {
@@ -665,7 +683,7 @@ func (qs *queueSet) dispatchLocked() bool {
 			request.workEstimate, queue.index, queue.nextDispatchR, queue.requests.Length(), queue.requestsExecuting, queue.seatsInUse, qs.totSeatsInUse)
 	}
 	// When a request is dequeued for service -> qs.virtualStart += G * width
-	queue.nextDispatchR += SeatsTimesDuration(float64(request.Seats()), qs.estimatedServiceDuration)
+	queue.nextDispatchR += request.InitialSeatSeconds(qs.estimatedServiceDuration) + request.AdditionalSeatSeconds()
 	qs.boundNextDispatch(queue)
 	request.decision.Set(decisionExecute)
 	return ok
@@ -716,7 +734,7 @@ func (qs *queueSet) findDispatchQueueLocked() *queue {
 			estimatedWorkInProgress := SeatsTimesDuration(float64(queue.seatsInUse), qs.estimatedServiceDuration)
 			dsMin = ssMin(dsMin, queue.nextDispatchR-estimatedWorkInProgress)
 			dsMax = ssMax(dsMax, queue.nextDispatchR-estimatedWorkInProgress)
-			currentVirtualFinish := queue.nextDispatchR + SeatsTimesDuration(float64(oldestWaiting.Seats()), qs.estimatedServiceDuration)
+			currentVirtualFinish := queue.nextDispatchR + oldestWaiting.InitialSeatSeconds(qs.estimatedServiceDuration) + oldestWaiting.AdditionalSeatSeconds()
 			klog.V(11).InfoS("Considering queue to dispatch", "queueSet", qs.qCfg.Name, "queue", qs.robinIndex, "finishR", currentVirtualFinish)
 			if currentVirtualFinish < minVirtualFinish {
 				minVirtualFinish = currentVirtualFinish
@@ -732,12 +750,12 @@ func (qs *queueSet) findDispatchQueueLocked() *queue {
 		klog.ErrorS(errors.New("selected queue is empty"), "Impossible", "queueSet", qs.qCfg.Name)
 		return nil
 	}
-	if !qs.canAccommodateSeatsLocked(oldestReqFromMinQueue.Seats()) {
+	if !qs.canAccommodateSeatsLocked(oldestReqFromMinQueue.MaxSeats()) {
 		// since we have not picked the queue with the minimum virtual finish
 		// time, we are not going to advance the round robin index here.
 		if klog.V(4).Enabled() {
 			klog.Infof("QS(%s): request %v %v seats %d cannot be dispatched from queue %d, waiting for currently executing requests to complete, %d requests are occupying %d seats and the limit is %d",
-				qs.qCfg.Name, oldestReqFromMinQueue.descr1, oldestReqFromMinQueue.descr2, oldestReqFromMinQueue.Seats(), minQueue.index, qs.totRequestsExecuting, qs.totSeatsInUse, qs.dCfg.ConcurrencyLimit)
+				qs.qCfg.Name, oldestReqFromMinQueue.descr1, oldestReqFromMinQueue.descr2, oldestReqFromMinQueue.MaxSeats(), minQueue.index, qs.totRequestsExecuting, qs.totSeatsInUse, qs.dCfg.ConcurrencyLimit)
 		}
 		return nil
 	}
@@ -799,10 +817,10 @@ func (qs *queueSet) finishRequestLocked(r *request) {
 	releaseSeatsLocked := func() {
 		defer qs.removeQueueIfEmptyLocked(r)
 
-		qs.totSeatsInUse -= r.Seats()
-		metrics.AddRequestConcurrencyInUse(qs.qCfg.Name, r.fsName, -r.Seats())
+		qs.totSeatsInUse -= r.MaxSeats()
+		metrics.AddRequestConcurrencyInUse(qs.qCfg.Name, r.fsName, -r.MaxSeats())
 		if r.queue != nil {
-			r.queue.seatsInUse -= r.Seats()
+			r.queue.seatsInUse -= r.MaxSeats()
 		}
 	}
 
@@ -812,9 +830,9 @@ func (qs *queueSet) finishRequestLocked(r *request) {
 			releaseSeatsLocked()
 			if !klog.V(6).Enabled() {
 			} else if r.queue != nil {
-				klog.Infof("QS(%s) at t=%s R=%v: request %#+v %#+v finished all use of %d seats, adjusted queue %d start R to %v due to service time %.9fs, queue will have %d requests, %d seats waiting & %d requests occupying %d seats",
-					qs.qCfg.Name, now.Format(nsTimeFmt), qs.currentR, r.descr1, r.descr2, r.workEstimate.InitialSeats, r.queue.index,
-					r.queue.nextDispatchR, actualServiceDuration.Seconds(), r.queue.requests.Length(), r.queue.requests.SeatsSum(), r.queue.requestsExecuting, r.queue.seatsInUse)
+				klog.Infof("QS(%s) at t=%s R=%v: request %#+v %#+v finished all use of %d seats, adjusted queue %d start R to %v due to service time %.9fs, queue sum: %#v, %d requests waiting & %d requests occupying %d seats",
+					qs.qCfg.Name, now.Format(nsTimeFmt), qs.currentR, r.descr1, r.descr2, r.workEstimate.MaxSeats(), r.queue.index,
+					r.queue.nextDispatchR, actualServiceDuration.Seconds(), r.queue.requests.QueueSum(), r.queue.requests.Length(), r.queue.requestsExecuting, r.queue.seatsInUse)
 			} else {
 				klog.Infof("QS(%s) at t=%s R=%v: request %#+v %#+v finished all use of %d seats, qs will have %d requests occupying %d seats", qs.qCfg.Name, now.Format(nsTimeFmt), qs.currentR, r.descr1, r.descr2, r.workEstimate.InitialSeats, qs.totRequestsExecuting, qs.totSeatsInUse)
 			}
@@ -824,9 +842,9 @@ func (qs *queueSet) finishRequestLocked(r *request) {
 		additionalLatency := r.workEstimate.AdditionalLatency
 		if !klog.V(6).Enabled() {
 		} else if r.queue != nil {
-			klog.Infof("QS(%s) at t=%s R=%v: request %#+v %#+v finished main use but lingering on %d seats for %v seconds, adjusted queue %d start R to %v due to service time %.9fs, queue will have %d requests waiting & %d executing, still has %d seats waiting & %d executing",
+			klog.Infof("QS(%s) at t=%s R=%v: request %#+v %#+v finished main use but lingering on %d seats for %v seconds, adjusted queue %d start R to %v due to service time %.9fs, queue will have %d requests waiting & %d executing, queue sum: %#v & %d seats in uses",
 				qs.qCfg.Name, now.Format(nsTimeFmt), qs.currentR, r.descr1, r.descr2, r.workEstimate.InitialSeats, additionalLatency.Seconds(), r.queue.index,
-				r.queue.nextDispatchR, actualServiceDuration.Seconds(), r.queue.requests.Length(), r.queue.requestsExecuting, r.queue.requests.SeatsSum(), r.queue.seatsInUse)
+				r.queue.nextDispatchR, actualServiceDuration.Seconds(), r.queue.requests.Length(), r.queue.requestsExecuting, r.queue.requests.QueueSum(), r.queue.seatsInUse)
 		} else {
 			klog.Infof("QS(%s) at t=%s R=%v: request %#+v %#+v finished main use but lingering on %d seats for %v seconds, qs will have %d requests occupying %d seats", qs.qCfg.Name, now.Format(nsTimeFmt), qs.currentR, r.descr1, r.descr2, r.workEstimate.InitialSeats, additionalLatency.Seconds(), qs.totRequestsExecuting, qs.totSeatsInUse)
 		}
@@ -841,9 +859,9 @@ func (qs *queueSet) finishRequestLocked(r *request) {
 			releaseSeatsLocked()
 			if !klog.V(6).Enabled() {
 			} else if r.queue != nil {
-				klog.Infof("QS(%s) at t=%s R=%v: request %#+v %#+v finished lingering on %d seats, queue %d will have %d requests, %d seats waiting & %d requests occupying %d seats",
+				klog.Infof("QS(%s) at t=%s R=%v: request %#+v %#+v finished lingering on %d seats, queue %d will have %d requests, queue sum: %#v, & %d requests occupying %d seats",
 					qs.qCfg.Name, now.Format(nsTimeFmt), qs.currentR, r.descr1, r.descr2, r.workEstimate.InitialSeats, r.queue.index,
-					r.queue.requests.Length(), r.queue.requests.SeatsSum(), r.queue.requestsExecuting, r.queue.seatsInUse)
+					r.queue.requests.Length(), r.queue.requests.QueueSum(), r.queue.requestsExecuting, r.queue.seatsInUse)
 			} else {
 				klog.Infof("QS(%s) at t=%s R=%v: request %#+v %#+v finished lingering on %d seats, qs will have %d requests occupying %d seats", qs.qCfg.Name, now.Format(nsTimeFmt), qs.currentR, r.descr1, r.descr2, r.workEstimate.InitialSeats, qs.totRequestsExecuting, qs.totSeatsInUse)
 			}
@@ -857,7 +875,7 @@ func (qs *queueSet) finishRequestLocked(r *request) {
 
 		// When a request finishes being served, and the actual service time was S,
 		// the queue’s start R is decremented by (G - S)*width.
-		r.queue.nextDispatchR -= SeatsTimesDuration(float64(r.Seats()), qs.estimatedServiceDuration-actualServiceDuration)
+		r.queue.nextDispatchR -= SeatsTimesDuration(float64(r.InitialSeats()), qs.estimatedServiceDuration-actualServiceDuration)
 		qs.boundNextDispatch(r.queue)
 	}
 }

staging/src/k8s.io/apiserver/pkg/util/flowcontrol/fairqueuing/queueset/queueset_test.go

@@ -1354,6 +1354,55 @@ func TestFinishRequestLocked(t *testing.T) {
 	}
 }
 
+func TestRequestSeats(t *testing.T) {
+	tests := []struct {
+		name     string
+		request  *request
+		expected int
+	}{
+		{
+			name:     "",
+			request:  &request{workEstimate: fcrequest.WorkEstimate{InitialSeats: 3, FinalSeats: 3}},
+			expected: 3,
+		},
+		{
+			name:     "",
+			request:  &request{workEstimate: fcrequest.WorkEstimate{InitialSeats: 1, FinalSeats: 3}},
+			expected: 3,
+		},
+		{
+			name:     "",
+			request:  &request{workEstimate: fcrequest.WorkEstimate{InitialSeats: 3, FinalSeats: 1}},
+			expected: 3,
+		},
+	}
+
+	for _, test := range tests {
+		t.Run(test.name, func(t *testing.T) {
+			seatsGot := test.request.MaxSeats()
+			if test.expected != seatsGot {
+				t.Errorf("Expected seats: %d, got %d", test.expected, seatsGot)
+			}
+		})
+	}
+}
+
+func TestRequestAdditionalSeatSeconds(t *testing.T) {
+	request := &request{
+		workEstimate: fcrequest.WorkEstimate{
+			InitialSeats:      3,
+			FinalSeats:        5,
+			AdditionalLatency: 3 * time.Second,
+		},
+	}
+
+	got := request.AdditionalSeatSeconds()
+	want := SeatsTimesDuration(5, 3*time.Second)
+	if want != got {
+		t.Errorf("Expected AdditionalSeatSeconds: %v, but got: %v", want, got)
+	}
+}
+
 func newFIFO(requests ...*request) fifo {
 	l := newRequestFIFO()
 	for i := range requests {

staging/src/k8s.io/apiserver/pkg/util/flowcontrol/fairqueuing/queueset/types.go

@@ -97,6 +97,22 @@ type queue struct {
 	seatsInUse int
 }
 
+// queueSum tracks the sum of initial seats, final seats, and
+// additional latency aggregated from all requests in a given queue
+type queueSum struct {
+	// InitialSeatsSum is the sum of InitialSeats
+	// associated with all requests in a given queue.
+	InitialSeatsSum int
+
+	// MaxSeatsSum is the sum of MaxSeats
+	// associated with all requests in a given queue.
+	MaxSeatsSum int
+
+	// AdditionalSeatSecondsSum is sum of AdditionalSeatsSeconds
+	// associated with all requests in a given queue.
+	AdditionalSeatSecondsSum SeatSeconds
+}
+
 // Enqueue enqueues a request into the queue and
 // sets the removeFromQueueFn of the request appropriately.
 func (q *queue) Enqueue(request *request) {
@@ -129,6 +145,8 @@ func (q *queue) dump(includeDetails bool) debug.QueueDump {
 		i++
 		return true
 	})
+
+	// TODO: change QueueDump to include queueSum stats
 	return debug.QueueDump{
 		VirtualStart:      q.nextDispatchR.ToFloat(), // TODO: change QueueDump to use SeatSeconds
 		Requests:          digest,

staging/src/k8s.io/apiserver/pkg/util/flowcontrol/request/width.go

@@ -34,9 +34,14 @@ const (
 )
 
 type WorkEstimate struct {
-	// InitialSeats represents the number of initial seats associated with this request
+	// InitialSeats is the number of seats occupied while the server is
+	// executing this request.
 	InitialSeats uint
 
+	// FinalSeats is the number of seats occupied at the end,
+	// during the AdditionalLatency.
+	FinalSeats uint
+
 	// AdditionalLatency specifies the additional duration the seats allocated
 	// to this request must be reserved after the given request had finished.
 	// AdditionalLatency should not have any impact on the user experience, the
@@ -44,6 +49,16 @@ type WorkEstimate struct {
 	AdditionalLatency time.Duration
 }
 
+// MaxSeats returns the number of seats this request requires, it is the maximum
+// of the two, WorkEstimate.InitialSeats and WorkEstimate.FinalSeats.
+func (we *WorkEstimate) MaxSeats() int {
+	if we.InitialSeats >= we.FinalSeats {
+		return int(we.InitialSeats)
+	}
+
+	return int(we.FinalSeats)
+}
+
 // objectCountGetterFunc represents a function that gets the total
 // number of objects for a given resource.
 type objectCountGetterFunc func(string) (int64, error)