Merge #109446

109446: kvadmission: fix handling of non-elastic work with flow control r=irfansharif a=aadityasondhi

Previously, when we were skipping flow control for regular work, we were
still encoding raft messages with AC encoding. This would enqueue
essentially no-op AC work items below raft, increasing the possibility
of OOM. This is because we would not be pacing the rate of regular work
by below-raft admission rates but it would be consuming memory in
below-raft work queues.

This change skips encoding raft messages with AC encoding for cases
where we bypass flow control. Additionally, we still subject such work
to above-raft IO admission control, if enabled.

Informs #104154.

Release note: None

Co-authored-by: Aaditya Sondhi <[email protected]>

Author: craig[bot]

pkg/kv/kvserver/flow_control_replica_integration_test.go

@@ -413,9 +413,9 @@ func newMockFlowHandle(
 
 func (m *mockFlowHandle) Admit(
 	ctx context.Context, pri admissionpb.WorkPriority, ct time.Time,
-) error {
+) (bool, error) {
 	m.t.Fatal("unimplemented")
-	return nil
+	return false, nil
 }
 
 func (m *mockFlowHandle) DeductTokensFor(

pkg/kv/kvserver/kvadmission/kvadmission.go

@@ -310,29 +310,41 @@ func (n *controllerImpl) AdmitKVWork(
 	// to continue even when throttling since there are often significant
 	// number of tokens available.
 	if ba.IsWrite() && !ba.IsSingleHeartbeatTxnRequest() {
-		if !bypassAdmission &&
-			kvflowcontrol.Enabled.Get(&n.settings.SV) &&
-			n.settings.Version.IsActive(ctx, clusterversion.V23_2_UseACRaftEntryEntryEncodings) {
+		var admitted bool
+		attemptFlowControl := kvflowcontrol.Enabled.Get(&n.settings.SV) &&
+			n.settings.Version.IsActive(ctx, clusterversion.V23_2_UseACRaftEntryEntryEncodings)
+		if attemptFlowControl && !bypassAdmission {
 			kvflowHandle, found := n.kvflowHandles.Lookup(ba.RangeID)
 			if !found {
 				return Handle{}, nil
 			}
-			if err := kvflowHandle.Admit(ctx, admissionInfo.Priority, timeutil.FromUnixNanos(createTime)); err != nil {
+			var err error
+			admitted, err = kvflowHandle.Admit(ctx, admissionInfo.Priority, timeutil.FromUnixNanos(createTime))
+			if err != nil {
 				return Handle{}, err
+			} else if admitted {
+				// NB: It's possible for us to be waiting for available flow tokens
+				// for a different set of streams that the ones we'll eventually
+				// deduct tokens from, if the range experiences a split between now
+				// and the point of deduction. That's ok, there's no strong
+				// synchronization needed between these two points.
+				ah.raftAdmissionMeta = &kvflowcontrolpb.RaftAdmissionMeta{
+					AdmissionPriority:   int32(admissionInfo.Priority),
+					AdmissionCreateTime: admissionInfo.CreateTime,
+					AdmissionOriginNode: n.nodeID.Get(),
+				}
 			}
-			// NB: It's possible for us to be waiting for available flow tokens
-			// for a different set of streams that the ones we'll eventually
-			// deduct tokens from, if the range experiences a split between now
-			// and the point of deduction. That's ok, there's no strong
-			// synchronization needed between these two points.
-			ah.raftAdmissionMeta = &kvflowcontrolpb.RaftAdmissionMeta{
-				AdmissionPriority:   int32(admissionInfo.Priority),
-				AdmissionCreateTime: admissionInfo.CreateTime,
-				AdmissionOriginNode: n.nodeID.Get(),
-			}
-		} else {
+
+		}
+		// If flow control is disabled or if work bypasses flow control, we still
+		// subject it above-raft, leaseholder-only IO admission control.
+		if !attemptFlowControl || !admitted {
 			storeAdmissionQ := n.storeGrantCoords.TryGetQueueForStore(int32(ba.Replica.StoreID))
 			if storeAdmissionQ != nil {
+				//  NB: Even though we would know here we're bypassing admission (via
+				//  `bypassAdmission`), we still have to explicitly invoke `.Admit()`.
+				//  We do it for correct token accounting (i.e. we deduct tokens without
+				//  blocking).
 				storeWorkHandle, err := storeAdmissionQ.Admit(
 					ctx, admission.StoreWriteWorkInfo{WorkInfo: admissionInfo})
 				if err != nil {

pkg/kv/kvserver/kvflowcontrol/kvflowcontrol.go

@@ -60,12 +60,6 @@ const (
 	// virtually enqueued in below-raft admission queues and dequeued in
 	// priority order, but only empty elastic flow token buckets above-raft will
 	// block further elastic traffic from being admitted.
-	//
-	// TODO(irfansharif): We're potentially risking OOMs doing all this tracking
-	// for regular work, without coalescing state. With a bit of plumbing, for
-	// requests that bypass flow control we could fallback to using the non-AC
-	// raft encodings and avoid the potential OOMs. Address this as part of
-	// #95563.
 	ApplyToElastic ModeT = iota
 	// ApplyToAll uses flow control for both elastic and regular traffic,
 	// i.e. all work will wait for flow tokens to be available.
@@ -117,11 +111,12 @@ type Tokens int64
 // Controller provides flow control for replication traffic in KV, held at the
 // node-level.
 type Controller interface {
-	// Admit seeks admission to replicate data, regardless of size, for work
-	// with the given priority, create-time, and over the given stream. This
-	// blocks until there are flow tokens available or the stream disconnects,
-	// subject to context cancellation.
-	Admit(context.Context, admissionpb.WorkPriority, time.Time, ConnectedStream) error
+	// Admit seeks admission to replicate data, regardless of size, for work with
+	// the given priority, create-time, and over the given stream. This blocks
+	// until there are flow tokens available or the stream disconnects, subject to
+	// context cancellation. This returns true if the request was admitted through
+	// flow control. Ignore the first return type if err != nil
+	Admit(context.Context, admissionpb.WorkPriority, time.Time, ConnectedStream) (admitted bool, _ error)
 	// DeductTokens deducts (without blocking) flow tokens for replicating work
 	// with given priority over the given stream. Requests are expected to
 	// have been Admit()-ed first.
@@ -158,10 +153,11 @@ type Controller interface {
 // given priority, takes log position into account -- see
 // kvflowcontrolpb.AdmittedRaftLogEntries for more details).
 type Handle interface {
-	// Admit seeks admission to replicate data, regardless of size, for work
-	// with the given priority and create-time. This blocks until there are
-	// flow tokens available for all connected streams.
-	Admit(context.Context, admissionpb.WorkPriority, time.Time) error
+	// Admit seeks admission to replicate data, regardless of size, for work with
+	// the given priority and create-time. This blocks until there are flow tokens
+	// available for all connected streams. This returns true if the request was
+	// admitted through flow control. Ignore the first return type if err != nil.
+	Admit(context.Context, admissionpb.WorkPriority, time.Time) (admitted bool, _ error)
 	// DeductTokensFor deducts (without blocking) flow tokens for replicating
 	// work with given priority along connected streams. The deduction is
 	// tracked with respect to the specific raft log position it's expecting it

pkg/kv/kvserver/kvflowcontrol/kvflowcontroller/kvflowcontroller.go

@@ -136,7 +136,7 @@ func (c *Controller) Admit(
 	pri admissionpb.WorkPriority,
 	_ time.Time,
 	connection kvflowcontrol.ConnectedStream,
-) error {
+) (bool, error) {
 	class := admissionpb.WorkClassFromPri(pri)
 	c.metrics.onWaiting(class)
 
@@ -148,12 +148,11 @@ func (c *Controller) Admit(
 		c.mu.Unlock()
 
 		tokens := b.tokens(class)
-		if tokens > 0 ||
-			// In addition to letting requests through when there are tokens
-			// being available, we'll also let them through if we're not
-			// applying flow control to their specific work class.
-			c.mode() == kvflowcontrol.ApplyToElastic && class == admissionpb.RegularWorkClass {
-
+		// In addition to letting requests through when there are tokens
+		// being available, we'll also let them through if we're not
+		// applying flow control to their specific work class.
+		bypass := c.mode() == kvflowcontrol.ApplyToElastic && class == admissionpb.RegularWorkClass
+		if tokens > 0 || bypass {
 			if log.ExpensiveLogEnabled(ctx, 2) {
 				log.Infof(ctx, "admitted request (pri=%s stream=%s tokens=%s wait-duration=%s mode=%s)",
 					pri, connection.Stream(), tokens, c.clock.PhysicalTime().Sub(tstart), c.mode())
@@ -179,7 +178,10 @@ func (c *Controller) Admit(
 
 			b.signal() // signal a waiter, if any
 			c.metrics.onAdmitted(class, c.clock.PhysicalTime().Sub(tstart))
-			return nil
+			if bypass {
+				return false, nil
+			}
+			return true, nil
 		}
 
 		if !logged && log.ExpensiveLogEnabled(ctx, 2) {
@@ -192,12 +194,12 @@ func (c *Controller) Admit(
 		case <-b.wait(): // wait for a signal
 		case <-connection.Disconnected():
 			c.metrics.onBypassed(class, c.clock.PhysicalTime().Sub(tstart))
-			return nil
+			return true, nil
 		case <-ctx.Done():
 			if ctx.Err() != nil {
 				c.metrics.onErrored(class, c.clock.PhysicalTime().Sub(tstart))
 			}
-			return ctx.Err()
+			return false, ctx.Err()
 		}
 	}
 

pkg/kv/kvserver/kvflowcontrol/kvflowcontroller/kvflowcontroller_test.go

@@ -196,7 +196,9 @@ func TestInspectController(t *testing.T) {
 
 	// Set up a single connected stream, s1/t1, and ensure it shows up in the
 	// Inspect() state.
-	require.NoError(t, controller.Admit(ctx, admissionpb.NormalPri, time.Time{}, makeConnectedStream(1)))
+	admitted, err := controller.Admit(ctx, admissionpb.NormalPri, time.Time{}, makeConnectedStream(1))
+	require.NoError(t, err)
+	require.True(t, admitted)
 	require.Len(t, controller.Inspect(ctx), 1)
 	require.Equal(t, controller.Inspect(ctx)[0],
 		makeInspectStream(1, 8<<20 /* 8MiB */, 16<<20 /* 16 MiB */))
@@ -211,7 +213,9 @@ func TestInspectController(t *testing.T) {
 
 	// Connect another stream, s1/s2, and ensure it shows up in the Inspect()
 	// state.
-	require.NoError(t, controller.Admit(ctx, admissionpb.BulkNormalPri, time.Time{}, makeConnectedStream(2)))
+	admitted, err = controller.Admit(ctx, admissionpb.BulkNormalPri, time.Time{}, makeConnectedStream(2))
+	require.NoError(t, err)
+	require.True(t, admitted)
 	require.Len(t, controller.Inspect(ctx), 2)
 	require.Equal(t, controller.Inspect(ctx)[1],
 		makeInspectStream(2, 8<<20 /* 8MiB */, 16<<20 /* 16 MiB */))

pkg/kv/kvserver/kvflowcontrol/kvflowhandle/kvflowhandle.go

@@ -79,7 +79,9 @@ func New(
 var _ kvflowcontrol.Handle = &Handle{}
 
 // Admit is part of the kvflowcontrol.Handle interface.
-func (h *Handle) Admit(ctx context.Context, pri admissionpb.WorkPriority, ct time.Time) error {
+func (h *Handle) Admit(
+	ctx context.Context, pri admissionpb.WorkPriority, ct time.Time,
+) (bool, error) {
 	if h == nil {
 		// TODO(irfansharif): This can happen if we're proposing immediately on
 		// a newly split off RHS that doesn't know it's a leader yet (so we
@@ -92,14 +94,14 @@ func (h *Handle) Admit(ctx context.Context, pri admissionpb.WorkPriority, ct tim
 		// As for cluster settings that disable flow control entirely or only
 		// for regular traffic, that can be dealt with at the caller by not
 		// calling .Admit() and ensuring we use the right raft entry encodings.
-		return nil
+		return false, nil
 	}
 
 	h.mu.Lock()
 	if h.mu.closed {
 		h.mu.Unlock()
 		log.Errorf(ctx, "operating on a closed handle")
-		return nil
+		return false, nil
 	}
 
 	// NB: We're using a copy-on-write scheme elsewhere to maintain this slice
@@ -115,15 +117,20 @@ func (h *Handle) Admit(ctx context.Context, pri admissionpb.WorkPriority, ct tim
 	h.metrics.onWaiting(class)
 	tstart := h.clock.PhysicalTime()
 
+	// NB: We track whether the last stream was subject to flow control, this
+	// helps us decide later if we should be deducting tokens for this work.
+	var admitted bool
 	for _, c := range connections {
-		if err := h.controller.Admit(ctx, pri, ct, c); err != nil {
+		var err error
+		admitted, err = h.controller.Admit(ctx, pri, ct, c)
+		if err != nil {
 			h.metrics.onErrored(class, h.clock.PhysicalTime().Sub(tstart))
-			return err
+			return false, err
 		}
 	}
 
 	h.metrics.onAdmitted(class, h.clock.PhysicalTime().Sub(tstart))
-	return nil
+	return admitted, nil
 }
 
 // DeductTokensFor is part of the kvflowcontrol.Handle interface.

pkg/kv/kvserver/kvflowcontrol/kvflowhandle/kvflowhandle_test.go

@@ -110,7 +110,9 @@ func TestHandleAdmit(t *testing.T) {
 			// the goroutine is blocked.
 			admitCh := make(chan struct{})
 			go func() {
-				require.NoError(t, handle.Admit(ctx, admissionpb.NormalPri, time.Time{}))
+				admitted, err := handle.Admit(ctx, admissionpb.NormalPri, time.Time{})
+				require.NoError(t, err)
+				require.True(t, admitted)
 				close(admitCh)
 			}()
 
@@ -189,16 +191,26 @@ func TestFlowControlMode(t *testing.T) {
 			handle.ConnectStream(ctx, pos(0), stream)
 			handle.DeductTokensFor(ctx, admissionpb.NormalPri, pos(1), kvflowcontrol.Tokens(16<<20 /* 16MiB */))
 
+			mode := tc.mode // copy to avoid nogo error
+
 			// Invoke .Admit() for {regular,elastic} work in a separate
 			// goroutines, and test below whether the goroutines are blocked.
 			regularAdmitCh := make(chan struct{})
 			elasticAdmitCh := make(chan struct{})
 			go func() {
-				require.NoError(t, handle.Admit(ctx, admissionpb.NormalPri, time.Time{}))
+				admitted, err := handle.Admit(ctx, admissionpb.NormalPri, time.Time{})
+				require.NoError(t, err)
+				if mode == kvflowcontrol.ApplyToElastic {
+					require.False(t, admitted)
+				} else {
+					require.True(t, admitted)
+				}
 				close(regularAdmitCh)
 			}()
 			go func() {
-				require.NoError(t, handle.Admit(ctx, admissionpb.BulkNormalPri, time.Time{}))
+				admitted, err := handle.Admit(ctx, admissionpb.BulkNormalPri, time.Time{})
+				require.NoError(t, err)
+				require.True(t, admitted)
 				close(elasticAdmitCh)
 			}()
 

pkg/kv/kvserver/kvflowcontrol/kvflowhandle/noop.go

@@ -26,8 +26,10 @@ type Noop struct{}
 var _ kvflowcontrol.Handle = Noop{}
 
 // Admit is part of the kvflowcontrol.Handle interface.
-func (n Noop) Admit(ctx context.Context, priority admissionpb.WorkPriority, time time.Time) error {
-	return nil
+func (n Noop) Admit(
+	ctx context.Context, priority admissionpb.WorkPriority, time time.Time,
+) (bool, error) {
+	return false, nil
 }
 
 // DeductTokensFor is part of the kvflowcontrol.Handle interface.

pkg/kv/kvserver/replica_proposal_buf_test.go

@@ -1166,8 +1166,8 @@ var _ kvflowcontrol.Handle = &testFlowTokenHandle{}
 
 func (t *testFlowTokenHandle) Admit(
 	ctx context.Context, priority admissionpb.WorkPriority, t2 time.Time,
-) error {
-	return nil
+) (bool, error) {
+	return false, nil
 }
 
 func (t *testFlowTokenHandle) DeductTokensFor(