circuitbreaker/process.go

package main

import (
	"context"
	"errors"
	"fmt"
	"time"

	"github.com/lightningnetwork/lnd/lnwire"
	"github.com/lightningnetwork/lnd/routing/route"
	"go.uber.org/zap"
	"golang.org/x/sync/errgroup"
)

var (
	rpcTimeout                 = 10 * time.Second
	defaultPeerRefreshInterval = 10 * time.Minute

	errChannelNotFound = errors.New("channel not found")
)

const burstSize = 10

type lndclient interface {
	getInfo() (*info, error)

	listChannels() (map[uint64]*channel, error)

	listClosedChannels() (map[uint64]*channel, error)

	getNodeAlias(key route.Vertex) (string, error)

	subscribeHtlcEvents(ctx context.Context) (htlcEventsClient, error)

	htlcInterceptor(ctx context.Context) (htlcInterceptorClient, error)

	getPendingIncomingHtlcs(ctx context.Context, peer *route.Vertex) (
		map[route.Vertex]map[circuitKey]*inFlightHtlc, error)
}

type circuitKey struct {
	channel uint64
	htlc    uint64
}

type interceptEvent struct {
	circuitKey
	incomingMsat lnwire.MilliSatoshi
	outgoingMsat lnwire.MilliSatoshi
	resume       func(bool) error
}

type resolvedEvent struct {
	incomingCircuitKey circuitKey
	outgoingCircuitKey circuitKey
	settled            bool
	timestamp          time.Time
}

type rateCounters struct {
	counters map[route.Vertex]*peerState
}

type rateCountersRequest struct {
	counters chan *rateCounters
}

type process struct {
	db     *Db
	client lndclient
	limits *Limits
	log    *zap.SugaredLogger

	interceptChan           chan interceptEvent
	resolveChan             chan resolvedEvent
	updateLimitChan         chan updateLimitEvent
	rateCountersRequestChan chan rateCountersRequest
	newPeerChan             chan route.Vertex

	identity route.Vertex
	chanMap  map[uint64]*channel
	aliasMap map[route.Vertex]string

	peerCtrls map[route.Vertex]*peerController

	burstSize           int
	peerRefreshInterval time.Duration

	// Testing hook
	resolvedCallback func()
}

func NewProcess(client lndclient, log *zap.SugaredLogger, limits *Limits, db *Db) *process {
	return &process{
		db:                      db,
		log:                     log,
		client:                  client,
		interceptChan:           make(chan interceptEvent),
		resolveChan:             make(chan resolvedEvent),
		updateLimitChan:         make(chan updateLimitEvent),
		rateCountersRequestChan: make(chan rateCountersRequest),
		newPeerChan:             make(chan route.Vertex),
		chanMap:                 make(map[uint64]*channel),
		aliasMap:                make(map[route.Vertex]string),
		peerCtrls:               make(map[route.Vertex]*peerController),
		limits:                  limits,
		burstSize:               burstSize,
		peerRefreshInterval:     defaultPeerRefreshInterval,
	}
}

type updateLimitEvent struct {
	limit *Limit
	peer  *route.Vertex
}

func (p *process) UpdateLimit(ctx context.Context, peer *route.Vertex,
	limit *Limit) error {

	if peer == nil && limit == nil {
		return errors.New("cannot clear default limit")
	}

	update := updateLimitEvent{
		limit: limit,
		peer:  peer,
	}

	select {
	case p.updateLimitChan <- update:
		return nil

	case <-ctx.Done():
		return ctx.Err()
	}
}

func (p *process) Run(ctx context.Context) error {
	p.log.Info("CircuitBreaker started")

	info, err := p.client.getInfo()
	if err != nil {
		return err
	}
	p.identity = info.nodeKey

	p.log.Infow("Connected to lnd node",
		"pubkey", p.identity.String())

	group, ctx := errgroup.WithContext(ctx)

	stream, err := p.client.subscribeHtlcEvents(ctx)
	if err != nil {
		return err
	}

	interceptor, err := p.client.htlcInterceptor(ctx)
	if err != nil {
		return err
	}

	p.log.Info("Interceptor/notification handlers registered")

	group.Go(func() error {
		err := p.processHtlcEvents(ctx, stream)
		if err != nil {
			return fmt.Errorf("htlc events error: %w", err)
		}

		return nil
	})

	group.Go(func() error {
		err := p.processInterceptor(ctx, interceptor)
		if err != nil {
			return fmt.Errorf("interceptor error: %w", err)
		}

		return err
	})

	group.Go(func() error {
		return p.peerRefreshLoop(ctx)
	})

	group.Go(func() error {
		return p.runEventLoop(ctx)
	})

	return group.Wait()
}

func (p *process) peerRefreshLoop(ctx context.Context) error {
	notifiedPeers := make(map[route.Vertex]struct{})

	for {
		// Get all peers.
		channels, err := p.client.listChannels()
		if err != nil {
			return err
		}

		// Notify the main event loop of the new ones.
		for _, ch := range channels {
			if _, ok := notifiedPeers[ch.peer]; ok {
				continue
			}

			notifiedPeers[ch.peer] = struct{}{}

			select {
			case p.newPeerChan <- ch.peer:
			case <-ctx.Done():
				return ctx.Err()
			}
		}

		// Poll delay.
		select {
		case <-time.After(p.peerRefreshInterval):
		case <-ctx.Done():
			return ctx.Err()
		}
	}
}

func (p *process) getPeerController(ctx context.Context, peer route.Vertex,
	startGo func(func() error)) *peerController {

	ctrl, ok := p.peerCtrls[peer]
	if ok {
		return ctrl
	}

	// If the peer does not yet exist, initialize it with no pending htlcs.
	htlcs := make(map[circuitKey]*inFlightHtlc)

	return p.createPeerController(ctx, peer, startGo, htlcs)
}

func (p *process) createPeerController(ctx context.Context, peer route.Vertex,
	startGo func(func() error),
	htlcs map[circuitKey]*inFlightHtlc) *peerController {

	peerCfg, ok := p.limits.PerPeer[peer]
	if !ok {
		peerCfg = p.limits.Default
	}

	cfg := &peerControllerCfg{
		logger:    p.log,
		limit:     peerCfg,
		burstSize: p.burstSize,
		htlcs:     htlcs,
		lnd:       p.client,
		pubKey:    peer,
		now:       time.Now,
		htlcCompleted: func(ctx context.Context, htlc *HtlcInfo) error {
			// If the add time of a htlc is zero, it was resumed after a LND
			// restart. We don't store these htlcs because they have
			// incomplete information (missing add time and amounts).
			if htlc.addTime.IsZero() {
				log.Debug("Not storing incomplete htlc resumed after "+
					"restart: %v (%v) -> %v (%v)",
					htlc.incomingCircuit.channel,
					htlc.incomingCircuit.htlc,
					htlc.outgoingCircuit.channel,
					htlc.outgoingCircuit.htlc)

				return nil
			}

			return p.db.RecordHtlcResolution(ctx, htlc)
		},
	}
	ctrl := newPeerController(cfg)

	startGo(func() error {
		return ctrl.run(ctx)
	})

	p.peerCtrls[peer] = ctrl

	return ctrl
}

func (p *process) runEventLoop(ctx context.Context) error {
	group, ctx := errgroup.WithContext(ctx)

	// Event loop will spin up new goroutines using the group that is passed in here.
	// We run it in the same group so that both errors in eventLoop and those in
	// the goroutines that is spins will will prompt exit.
	group.Go(func() error {
		return p.eventLoop(ctx, group)
	})

	return group.Wait()
}

func (p *process) eventLoop(ctx context.Context, group *errgroup.Group) error {
	// Retrieve all pending htlcs from lnd.
	htlcsPerPeer, err := p.client.getPendingIncomingHtlcs(ctx, nil)
	if err != nil {
		return err
	}

	// Initialize peer controllers with currently pending htlcs.
	for peer, htlcs := range htlcsPerPeer {
		p.createPeerController(ctx, peer, group.Go, htlcs)
	}

	for {
		select {
		case interceptEvent := <-p.interceptChan:
			chanInfo, err := p.getChanInfo(interceptEvent.channel)
			if err != nil {
				return err
			}

			ctrl := p.getPeerController(ctx, chanInfo.peer, group.Go)

			peerEvent := peerInterceptEvent{
				interceptEvent: interceptEvent,
				peerInitiated:  !chanInfo.initiator,
			}
			if err := ctrl.process(ctx, peerEvent); err != nil {
				return err
			}

		case resolvedEvent := <-p.resolveChan:
			chanInfo, err := p.getChanInfo(
				resolvedEvent.incomingCircuitKey.channel,
			)
			if err != nil {
				return err
			}

			ctrl := p.getPeerController(ctx, chanInfo.peer, group.Go)

			// Lookup the outgoing peer to supplement the information on the
			// resolved event. Here we handle a channel lookup error
			// differently to the incoming channel, because it's possible
			// we were forwarded a HTLC with a bogus outgoing channel. If
			// this is the case, LND would have failed the HTLC back even if
			// we let it through. We catch and log that error, rather than
			// exiting like we do with incoming channels (where we reasonably
			// expect to find the channel). We still enforce channel lookup
			// for successful HTLCs, because then we know that the channel
			// does exist and should be found.
			var outgoingPeer *route.Vertex
			chanInfo, err = p.getChanInfo(
				resolvedEvent.outgoingCircuitKey.channel,
			)
			switch {
			case errors.Is(err, errChannelNotFound) && !resolvedEvent.settled:
				log.Debugf("Channel not found for failed htlc: %v",
					resolvedEvent.outgoingCircuitKey.channel)

			case err != nil:
				return err

			default:
				outgoingPeer = &chanInfo.peer
			}

			if err := ctrl.resolved(ctx, peerResolvedEvent{
				resolvedEvent: resolvedEvent,
				outgoingPeer:  outgoingPeer,
			}); err != nil {
				return err
			}

			if p.resolvedCallback != nil {
				p.resolvedCallback()
			}

		case update := <-p.updateLimitChan:
			switch {
			// Update sets default limit.
			case update.peer == nil:
				p.limits.Default = *update.limit

				// Update all controllers that have no specific limit.
				for node, ctrl := range p.peerCtrls {
					_, ok := p.limits.PerPeer[node]
					if ok {
						continue
					}

					err := ctrl.updateLimit(ctx, *update.limit)
					if err != nil {
						return err
					}
				}

			// Update sets specific limit.
			case update.limit != nil:
				p.limits.PerPeer[*update.peer] = *update.limit

				// Update specific controller if it exists.
				ctrl, ok := p.peerCtrls[*update.peer]
				if ok {
					err := ctrl.updateLimit(ctx, *update.limit)
					if err != nil {
						return err
					}
				}

			// Update clears limit.
			case update.limit == nil:
				delete(p.limits.PerPeer, *update.peer)

				// Apply default limit to peer controller.
				ctrl, ok := p.peerCtrls[*update.peer]
				if ok {
					err := ctrl.updateLimit(ctx, p.limits.Default)
					if err != nil {
						return err
					}
				}
			}

		case req := <-p.rateCountersRequestChan:
			allCounts := make(map[route.Vertex]*peerState)
			for node, ctrl := range p.peerCtrls {
				state, err := ctrl.state(ctx)
				if err != nil {
					return err
				}

				allCounts[node] = state
			}

			req.counters <- &rateCounters{
				counters: allCounts,
			}

		case <-ctx.Done():
			return ctx.Err()

		// A new or existing peer has been reported.
		case newPeer := <-p.newPeerChan:
			p.log.Infow("New peer notification received", "peer", newPeer)

			// Try to get the existing peer controller. If it doesn't exist, it
			// will be created. This causes the peer to be reported over grpc.
			_ = p.getPeerController(ctx, newPeer, group.Go)
		}
	}
}

func (p *process) getRateCounters(ctx context.Context) (
	map[route.Vertex]*peerState, error) {

	replyChan := make(chan *rateCounters)

	select {
	case p.rateCountersRequestChan <- rateCountersRequest{
		counters: replyChan,
	}:
	case <-ctx.Done():
		return nil, ctx.Err()
	}

	select {
	case reply := <-replyChan:
		return reply.counters, nil

	case <-ctx.Done():
		return nil, ctx.Err()
	}
}

func (p *process) processHtlcEvents(ctx context.Context,
	stream htlcEventsClient) error {

	for {
		event, err := stream.recv()
		if err != nil {
			return err
		}

		select {
		case p.resolveChan <- *event:
		case <-ctx.Done():
			return ctx.Err()
		}
	}
}

func (p *process) processInterceptor(ctx context.Context,
	interceptor htlcInterceptorClient) error {

	for {
		event, err := interceptor.recv()
		if err != nil {
			return err
		}

		key := event.circuitKey

		resume := func(resume bool) error {
			return interceptor.send(&interceptResponse{
				key:    key,
				resume: resume,
			})
		}

		select {
		case p.interceptChan <- interceptEvent{
			circuitKey:   key,
			incomingMsat: event.incomingMsat,
			outgoingMsat: event.outgoingMsat,
			resume:       resume,
		}:

		case <-ctx.Done():
			return ctx.Err()
		}
	}
}

func (p *process) getChanInfo(channel uint64) (*channel, error) {
	// Try to look up from the cache.
	ch, ok := p.chanMap[channel]
	if ok {
		return ch, nil
	}

	// Cache miss. Retrieve all channels and update the cache.
	channels, err := p.client.listChannels()
	if err != nil {
		return nil, err
	}

	for chanId, ch := range channels {
		p.chanMap[chanId] = ch
	}

	// Try looking up the channel again.
	ch, ok = p.chanMap[channel]
	if ok {
		return ch, nil
	}

	// If the channel is not open, fall back to checking our closed
	// channels.
	closedChannels, err := p.client.listClosedChannels()
	if err != nil {
		return nil, err
	}

	// Add to cache and try again.
	for chanId, ch := range closedChannels {
		p.chanMap[chanId] = ch
	}

	ch, ok = p.chanMap[channel]
	if ok {
		return ch, nil
	}

	// Channel not found.
	return nil, fmt.Errorf("%w: %v", errChannelNotFound, channel)
}