remove cpu reservation (#552)

* remove cpu reservation

* remove IsIdle

pkg/service/process.go

@@ -165,13 +165,6 @@ func (s *Service) awaitCleanup(p *Process) {
 	s.mu.Unlock()
 }
 
-func (s *Service) IsIdle() bool {
-	s.mu.RLock()
-	defer s.mu.RUnlock()
-
-	return len(s.activeHandlers) == 0
-}
-
 func (s *Service) getGRPCClient(egressID string) (ipc.EgressHandlerClient, error) {
 	s.mu.RLock()
 	defer s.mu.RUnlock()

pkg/service/service.go

@@ -163,7 +163,7 @@ func (s *Service) StartEgressAffinity(ctx context.Context, req *rpc.StartEgressR
 		return -1
 	}
 
-	if s.IsIdle() {
+	if s.GetRequestCount() == 0 {
 		// group multiple track and track composite requests.
 		// if this instance is idle and another is already handling some, the request will go to that server.
 		// this avoids having many instances with one track request each, taking availability from room composite.
@@ -194,7 +194,6 @@ func (s *Service) ListActiveEgress(ctx context.Context, _ *rpc.ListActiveEgressR
 func (s *Service) Status() ([]byte, error) {
 	info := map[string]interface{}{
 		"CpuLoad": s.GetCPULoad(),
-		"CpuHold": s.GetCPUHold(),
 	}
 
 	s.mu.RLock()
@@ -207,7 +206,7 @@ func (s *Service) Status() ([]byte, error) {
 }
 
 func (s *Service) promIsIdle() float64 {
-	if s.IsIdle() {
+	if s.GetRequestCount() == 0 {
 		return 1
 	}
 	return 0
@@ -234,7 +233,7 @@ func (s *Service) Stop(kill bool) {
 }
 
 func (s *Service) Close() {
-	for !s.IsIdle() {
+	for s.GetRequestCount() > 0 {
 		time.Sleep(shutdownTimer)
 	}
 	logger.Infow("closing server")

pkg/stats/monitor.go

@@ -42,16 +42,14 @@ type Monitor struct {
 	pendingCPUs atomic.Float64
 
 	mu       sync.Mutex
-	counts   map[string]int
-	reserved float64
+	requests atomic.Int32
 }
 
 const cpuHoldDuration = time.Second * 5
 
 func NewMonitor(conf *config.ServiceConfig) *Monitor {
 	return &Monitor{
 		cpuCostConfig: conf.CPUCostConfig,
-		counts:        make(map[string]int),
 	}
 }
 
@@ -147,11 +145,8 @@ func (m *Monitor) GetCPULoad() float64 {
 	return (m.cpuStats.NumCPU() - m.cpuStats.GetCPUIdle()) / m.cpuStats.NumCPU() * 100
 }
 
-func (m *Monitor) GetCPUHold() float64 {
-	m.mu.Lock()
-	defer m.mu.Unlock()
-
-	return m.reserved
+func (m *Monitor) GetRequestCount() int {
+	return int(m.requests.Load())
 }
 
 func (m *Monitor) CanAcceptRequest(req *rpc.StartEgressRequest) bool {
@@ -170,13 +165,15 @@ func (m *Monitor) canAcceptRequest(req *rpc.StartEgressRequest) bool {
 	logger.Debugw("cpu check",
 		"total", total,
 		"available", available,
-		"reserved", m.reserved,
+		"active_requests", m.requests,
 	)
 
-	if m.reserved == 0 {
+	if m.requests.Load() == 0 {
+		// if no requests, use total
 		available = total
-	} else if total-m.reserved < available {
-		available = total - m.reserved
+	} else {
+		// if already running requests, cap usage at 90%
+		available -= 0.9 * total
 	}
 
 	switch req.Request.(type) {
@@ -217,7 +214,7 @@ func (m *Monitor) AcceptRequest(req *rpc.StartEgressRequest) error {
 		cpuHold = m.cpuCostConfig.TrackCpuCost
 	}
 
-	m.reserved += cpuHold
+	m.requests.Inc()
 	m.pendingCPUs.Add(cpuHold)
 	time.AfterFunc(cpuHoldDuration, func() { m.pendingCPUs.Sub(cpuHold) })
 	return nil
@@ -244,37 +241,20 @@ func (m *Monitor) EgressEnded(req *rpc.StartEgressRequest) {
 
 	switch req.Request.(type) {
 	case *rpc.StartEgressRequest_RoomComposite:
-		m.reserved -= m.cpuCostConfig.RoomCompositeCpuCost
 		m.requestGauge.With(prometheus.Labels{"type": types.RequestTypeRoomComposite}).Sub(1)
 	case *rpc.StartEgressRequest_Web:
-		m.reserved -= m.cpuCostConfig.WebCpuCost
 		m.requestGauge.With(prometheus.Labels{"type": types.RequestTypeWeb}).Sub(1)
 	case *rpc.StartEgressRequest_Participant:
-		m.reserved -= m.cpuCostConfig.ParticipantCpuCost
 		m.requestGauge.With(prometheus.Labels{"type": types.RequestTypeParticipant}).Sub(1)
 	case *rpc.StartEgressRequest_TrackComposite:
-		m.reserved -= m.cpuCostConfig.TrackCompositeCpuCost
 		m.requestGauge.With(prometheus.Labels{"type": types.RequestTypeTrackComposite}).Sub(1)
 	case *rpc.StartEgressRequest_Track:
-		m.reserved -= m.cpuCostConfig.TrackCpuCost
 		m.requestGauge.With(prometheus.Labels{"type": types.RequestTypeTrack}).Sub(1)
 	}
-}
 
-func (m *Monitor) EgressAborted(req *rpc.StartEgressRequest) {
-	m.mu.Lock()
-	defer m.mu.Unlock()
+	m.requests.Dec()
+}
 
-	switch req.Request.(type) {
-	case *rpc.StartEgressRequest_RoomComposite:
-		m.reserved -= m.cpuCostConfig.RoomCompositeCpuCost
-	case *rpc.StartEgressRequest_Web:
-		m.reserved -= m.cpuCostConfig.WebCpuCost
-	case *rpc.StartEgressRequest_Participant:
-		m.reserved -= m.cpuCostConfig.ParticipantCpuCost
-	case *rpc.StartEgressRequest_TrackComposite:
-		m.reserved -= m.cpuCostConfig.TrackCompositeCpuCost
-	case *rpc.StartEgressRequest_Track:
-		m.reserved -= m.cpuCostConfig.TrackCpuCost
-	}
+func (m *Monitor) EgressAborted(_ *rpc.StartEgressRequest) {
+	m.requests.Dec()
 }

test/integration.go

@@ -102,7 +102,7 @@ type testCase struct {
 func (r *Runner) awaitIdle(t *testing.T) {
 	r.svc.KillAll()
 	for i := 0; i < 30; i++ {
-		if r.svc.GetCPUHold() == 0 {
+		if r.svc.GetRequestCount() == 0 {
 			return
 		}
 		time.Sleep(time.Second)