status-go/peers/topicpool.go
Andrea Maria Piana 58fcf809ea
Bug/fix pending peers segfault (#2004)
* Replace mclock with time in peer pools

we used mclock as golang before 1.9 did not support monotonic clocks,
https://github.com/gavv/monotime, it does now https://golang.org/pkg/time/
so we can fallback on the system implementation, which will return
nanoseconds with a resolution that is system dependent.

* Handle case where peer have same discovered time

In some system the resolution of the clock is not high enough so
multiple peers are added on the same nanosecond.

This result in the peer just added being immediately removed.

This code adds a check making sure we don't assume that a peer is added.

Another approach would be to make sure to include the peer in the list,
so prevent the peer just being added to be evicted, but it's slightly
more complicated and the resolution is generally accurate enough for our
purpose so that peers will be fresh enough if they have the same
discovered time.

It also adds a regression test, I had to use an interface to stub the
clock.

Fixes: https://github.com/status-im/nim-status-client/issues/522

* bump version to 0.55.3
2020-07-24 11:14:05 -04:00

596 lines
16 KiB
Go

package peers
import (
"container/heap"
"sync"
"sync/atomic"
"time"
"github.com/ethereum/go-ethereum/log"
"github.com/ethereum/go-ethereum/p2p"
"github.com/ethereum/go-ethereum/p2p/discv5"
"github.com/ethereum/go-ethereum/p2p/enode"
"github.com/status-im/status-go/discovery"
"github.com/status-im/status-go/params"
)
const (
// notQueuedIndex used to define that item is not queued in the heap queue.
notQueuedIndex = -1
)
// maxCachedPeersMultiplier peers max limit will be multiplied by this number
// to get the maximum number of cached peers allowed.
var maxCachedPeersMultiplier = 1
// maxPendingPeersMultiplier peers max limit will be multiplied by this number
// to get the maximum number of pending peers allowed.
var maxPendingPeersMultiplier = 2
// TopicPoolInterface the TopicPool interface.
type TopicPoolInterface interface {
StopSearch(server *p2p.Server)
BelowMin() bool
SearchRunning() bool
StartSearch(server *p2p.Server) error
ConfirmDropped(server *p2p.Server, nodeID enode.ID) bool
AddPeerFromTable(server *p2p.Server) *discv5.Node
MaxReached() bool
ConfirmAdded(server *p2p.Server, nodeID enode.ID)
isStopped() bool
Topic() discv5.Topic
SetLimits(limits params.Limits)
setStopSearchTimeout(delay time.Duration)
readyToStopSearch() bool
}
type Clock interface {
Now() time.Time
}
type realClock struct{}
func (realClock) Now() time.Time { return time.Now() }
// newTopicPool returns instance of TopicPool.
func newTopicPool(discovery discovery.Discovery, topic discv5.Topic, limits params.Limits, slowMode, fastMode time.Duration, cache *Cache) *TopicPool {
pool := TopicPool{
discovery: discovery,
topic: topic,
limits: limits,
fastMode: fastMode,
slowMode: slowMode,
fastModeTimeout: DefaultTopicFastModeTimeout,
pendingPeers: make(map[enode.ID]*peerInfoItem),
discoveredPeersQueue: make(peerPriorityQueue, 0),
discoveredPeers: make(map[enode.ID]bool),
connectedPeers: make(map[enode.ID]*peerInfo),
cache: cache,
maxCachedPeers: limits.Max * maxCachedPeersMultiplier,
maxPendingPeers: limits.Max * maxPendingPeersMultiplier,
clock: realClock{},
}
heap.Init(&pool.discoveredPeersQueue)
return &pool
}
// TopicPool manages peers for topic.
type TopicPool struct {
discovery discovery.Discovery
// configuration
topic discv5.Topic
limits params.Limits
fastMode time.Duration
slowMode time.Duration
fastModeTimeout time.Duration
mu sync.RWMutex
discWG sync.WaitGroup
poolWG sync.WaitGroup
quit chan struct{}
running int32
currentMode time.Duration
period chan time.Duration
fastModeTimeoutCancel chan struct{}
pendingPeers map[enode.ID]*peerInfoItem // contains found and requested to be connected peers but not confirmed
discoveredPeersQueue peerPriorityQueue // priority queue to find the most recently discovered peers; does not containt peers requested to connect
discoveredPeers map[enode.ID]bool // remembers which peers have already been discovered and are enqueued
connectedPeers map[enode.ID]*peerInfo // currently connected peers
stopSearchTimeout *time.Time
maxPendingPeers int
maxCachedPeers int
cache *Cache
clock Clock
}
func (t *TopicPool) addToPendingPeers(peer *peerInfo) {
if _, ok := t.pendingPeers[peer.NodeID()]; ok {
return
}
t.pendingPeers[peer.NodeID()] = &peerInfoItem{
peerInfo: peer,
index: notQueuedIndex,
}
// maxPendingPeers = 0 means no limits.
if t.maxPendingPeers == 0 || t.maxPendingPeers >= len(t.pendingPeers) {
return
}
var oldestPeer *peerInfo
for _, i := range t.pendingPeers {
if oldestPeer != nil && oldestPeer.discoveredTime.Before(i.peerInfo.discoveredTime) {
continue
}
oldestPeer = i.peerInfo
}
t.removeFromPendingPeers(oldestPeer.NodeID())
}
// addToQueue adds the passed peer to the queue if it is already pending.
func (t *TopicPool) addToQueue(peer *peerInfo) {
if p, ok := t.pendingPeers[peer.NodeID()]; ok {
if _, ok := t.discoveredPeers[peer.NodeID()]; ok {
return
}
heap.Push(&t.discoveredPeersQueue, p)
t.discoveredPeers[peer.NodeID()] = true
}
}
func (t *TopicPool) popFromQueue() *peerInfo {
if t.discoveredPeersQueue.Len() == 0 {
return nil
}
item := heap.Pop(&t.discoveredPeersQueue).(*peerInfoItem)
item.index = notQueuedIndex
delete(t.discoveredPeers, item.peerInfo.NodeID())
return item.peerInfo
}
func (t *TopicPool) removeFromPendingPeers(nodeID enode.ID) {
peer, ok := t.pendingPeers[nodeID]
if !ok {
return
}
delete(t.pendingPeers, nodeID)
if peer.index != notQueuedIndex {
heap.Remove(&t.discoveredPeersQueue, peer.index)
delete(t.discoveredPeers, nodeID)
}
}
func (t *TopicPool) updatePendingPeer(nodeID enode.ID) {
peer, ok := t.pendingPeers[nodeID]
if !ok {
return
}
peer.discoveredTime = t.clock.Now()
if peer.index != notQueuedIndex {
heap.Fix(&t.discoveredPeersQueue, peer.index)
}
}
func (t *TopicPool) movePeerFromPoolToConnected(nodeID enode.ID) {
peer, ok := t.pendingPeers[nodeID]
if !ok {
return
}
t.removeFromPendingPeers(nodeID)
t.connectedPeers[nodeID] = peer.peerInfo
}
// SearchRunning returns true if search is running
func (t *TopicPool) SearchRunning() bool {
return atomic.LoadInt32(&t.running) == 1
}
// MaxReached returns true if we connected with max number of peers.
func (t *TopicPool) MaxReached() bool {
t.mu.RLock()
defer t.mu.RUnlock()
return len(t.connectedPeers) == t.limits.Max
}
// BelowMin returns true if current number of peers is below min limit.
func (t *TopicPool) BelowMin() bool {
t.mu.RLock()
defer t.mu.RUnlock()
return len(t.connectedPeers) < t.limits.Min
}
// maxCachedPeersReached returns true if max number of cached peers is reached.
func (t *TopicPool) maxCachedPeersReached() bool {
if t.maxCachedPeers == 0 {
return true
}
peers := t.cache.GetPeersRange(t.topic, t.maxCachedPeers)
return len(peers) >= t.maxCachedPeers
}
// setStopSearchTimeout sets the timeout to stop current topic search if it's not
// been stopped before.
func (t *TopicPool) setStopSearchTimeout(delay time.Duration) {
if t.stopSearchTimeout != nil {
return
}
now := t.clock.Now().Add(delay)
t.stopSearchTimeout = &now
}
// isStopSearchDelayExpired returns true if the timeout to stop current topic
// search has been accomplished.
func (t *TopicPool) isStopSearchDelayExpired() bool {
if t.stopSearchTimeout == nil {
return false
}
return t.stopSearchTimeout.Before(t.clock.Now())
}
// readyToStopSearch return true if all conditions to stop search are ok.
func (t *TopicPool) readyToStopSearch() bool {
return t.isStopSearchDelayExpired() || t.maxCachedPeersReached()
}
// updateSyncMode changes the sync mode depending on the current number
// of connected peers and limits.
func (t *TopicPool) updateSyncMode() {
newMode := t.slowMode
if len(t.connectedPeers) < t.limits.Min {
newMode = t.fastMode
}
t.setSyncMode(newMode)
}
func (t *TopicPool) setSyncMode(mode time.Duration) {
if mode == t.currentMode {
return
}
t.period <- mode
t.currentMode = mode
// if selected mode is fast mode and fast mode timeout was not set yet,
// do it now
if mode == t.fastMode && t.fastModeTimeoutCancel == nil {
t.fastModeTimeoutCancel = t.limitFastMode(t.fastModeTimeout)
}
// remove fast mode timeout as slow mode is selected now
if mode == t.slowMode && t.fastModeTimeoutCancel != nil {
close(t.fastModeTimeoutCancel)
t.fastModeTimeoutCancel = nil
}
}
func (t *TopicPool) limitFastMode(timeout time.Duration) chan struct{} {
if timeout == 0 {
return nil
}
cancel := make(chan struct{})
t.poolWG.Add(1)
go func() {
defer t.poolWG.Done()
select {
case <-time.After(timeout):
t.mu.Lock()
t.setSyncMode(t.slowMode)
t.mu.Unlock()
case <-cancel:
return
}
}()
return cancel
}
// ConfirmAdded called when peer was added by p2p Server.
// 1. Skip a peer if it not in our peer table
// 2. Add a peer to a cache.
// 3. Disconnect a peer if it was connected after we reached max limit of peers.
// (we can't know in advance if peer will be connected, thats why we allow
// to overflow for short duration)
// 4. Switch search to slow mode if it is running.
func (t *TopicPool) ConfirmAdded(server *p2p.Server, nodeID enode.ID) {
t.mu.Lock()
defer t.mu.Unlock()
peerInfoItem, ok := t.pendingPeers[nodeID]
inbound := !ok || !peerInfoItem.added
log.Debug("peer added event", "peer", nodeID.String(), "inbound", inbound)
if inbound {
return
}
peer := peerInfoItem.peerInfo // get explicit reference
// established connection means that the node
// is a viable candidate for a connection and can be cached
if err := t.cache.AddPeer(peer.node, t.topic); err != nil {
log.Error("failed to persist a peer", "error", err)
}
t.movePeerFromPoolToConnected(nodeID)
// if the upper limit is already reached, drop this peer
if len(t.connectedPeers) > t.limits.Max {
log.Debug("max limit is reached drop the peer", "ID", nodeID, "topic", t.topic)
peer.dismissed = true
t.removeServerPeer(server, peer)
return
}
// make sure `dismissed` is reset
peer.dismissed = false
// A peer was added so check if we can switch to slow mode.
if t.SearchRunning() {
t.updateSyncMode()
}
}
// ConfirmDropped called when server receives drop event.
// 1. Skip peer if it is not in our peer table.
// 2. If disconnect request - we could drop that peer ourselves.
// 3. If connected number will drop below min limit - switch to fast mode.
// 4. Delete a peer from cache and peer table.
// Returns false if peer is not in our table or we requested removal of this peer.
// Otherwise peer is removed and true is returned.
func (t *TopicPool) ConfirmDropped(server *p2p.Server, nodeID enode.ID) bool {
t.mu.Lock()
defer t.mu.Unlock()
// either inbound or connected from another topic
peer, exist := t.connectedPeers[nodeID]
if !exist {
return false
}
log.Debug("disconnect", "ID", nodeID, "dismissed", peer.dismissed)
delete(t.connectedPeers, nodeID)
// Peer was removed by us because exceeded the limit.
// Add it back to the pool as it can be useful in the future.
if peer.dismissed {
t.addToPendingPeers(peer)
// use queue for peers that weren't added to p2p server
t.addToQueue(peer)
return false
}
// If there was a network error, this event will be received
// but the peer won't be removed from the static nodes set.
// That's why we need to call `removeServerPeer` manually.
t.removeServerPeer(server, peer)
if err := t.cache.RemovePeer(nodeID, t.topic); err != nil {
log.Error("failed to remove peer from cache", "error", err)
}
// As we removed a peer, update a sync strategy if needed.
if t.SearchRunning() {
t.updateSyncMode()
}
return true
}
// AddPeerFromTable checks if there is a valid peer in local table and adds it to a server.
func (t *TopicPool) AddPeerFromTable(server *p2p.Server) *discv5.Node {
t.mu.RLock()
defer t.mu.RUnlock()
// The most recently added peer is removed from the queue.
// If it did not expire yet, it will be added to the server.
// TODO(adam): investigate if it's worth to keep the peer in the queue
// until the server confirms it is added and in the meanwhile only adjust its priority.
peer := t.popFromQueue()
if peer != nil && t.clock.Now().Before(peer.discoveredTime.Add(expirationPeriod)) {
t.addServerPeer(server, peer)
return peer.node
}
return nil
}
// StartSearch creates discv5 queries and runs a loop to consume found peers.
func (t *TopicPool) StartSearch(server *p2p.Server) error {
if atomic.LoadInt32(&t.running) == 1 {
return nil
}
if !t.discovery.Running() {
return ErrDiscv5NotRunning
}
atomic.StoreInt32(&t.running, 1)
t.mu.Lock()
defer t.mu.Unlock()
t.quit = make(chan struct{})
t.stopSearchTimeout = nil
// `period` is used to notify about the current sync mode.
t.period = make(chan time.Duration, 2)
// use fast sync mode at the beginning
t.setSyncMode(t.fastMode)
// peers management
found := make(chan *discv5.Node, 5) // 5 reasonable number for concurrently found nodes
lookup := make(chan bool, 10) // sufficiently buffered channel, just prevents blocking because of lookup
for _, peer := range t.cache.GetPeersRange(t.topic, 5) {
log.Debug("adding a peer from cache", "peer", peer)
found <- peer
}
t.discWG.Add(1)
go func() {
if err := t.discovery.Discover(string(t.topic), t.period, found, lookup); err != nil {
log.Error("error searching foro", "topic", t.topic, "err", err)
}
t.discWG.Done()
}()
t.poolWG.Add(1)
go func() {
t.handleFoundPeers(server, found, lookup)
t.poolWG.Done()
}()
return nil
}
func (t *TopicPool) handleFoundPeers(server *p2p.Server, found <-chan *discv5.Node, lookup <-chan bool) {
selfID := discv5.PubkeyID(server.Self().Pubkey())
for {
select {
case <-t.quit:
return
case <-lookup:
case node := <-found:
if node.ID == selfID {
continue
}
if err := t.processFoundNode(server, node); err != nil {
log.Error("failed to process found node", "node", node, "error", err)
}
}
}
}
// processFoundNode called when node is discovered by kademlia search query
// 2 important conditions
// 1. every time when node is processed we need to update discoveredTime.
// peer will be considered as valid later only if it was discovered < 60m ago
// 2. if peer is connected or if max limit is reached we are not a adding peer to p2p server
func (t *TopicPool) processFoundNode(server *p2p.Server, node *discv5.Node) error {
t.mu.Lock()
defer t.mu.Unlock()
pk, err := node.ID.Pubkey()
if err != nil {
return err
}
nodeID := enode.PubkeyToIDV4(pk)
log.Debug("peer found", "ID", nodeID, "topic", t.topic)
// peer is already connected so update only discoveredTime
if peer, ok := t.connectedPeers[nodeID]; ok {
peer.discoveredTime = t.clock.Now()
return nil
}
if _, ok := t.pendingPeers[nodeID]; ok {
t.updatePendingPeer(nodeID)
} else {
t.addToPendingPeers(&peerInfo{
discoveredTime: t.clock.Now(),
node: node,
publicKey: pk,
})
}
log.Debug(
"adding peer to a server", "peer", node.ID.String(),
"connected", len(t.connectedPeers), "max", t.maxCachedPeers)
// This can happen when the monotonic clock is not precise enough and
// multiple peers gets added at the same clock time, resulting in all
// of them having the same discoveredTime.
// At which point a random peer will be removed, sometimes being the
// peer we just added.
// We could make sure that the latest added peer is not removed,
// but this is simpler, and peers will be fresh enough as resolution
// should be quite high (ms at least).
// This has been reported on windows builds
// only https://github.com/status-im/nim-status-client/issues/522
if t.pendingPeers[nodeID] == nil {
log.Debug("peer added has just been removed", "peer", nodeID)
return nil
}
// the upper limit is not reached, so let's add this peer
if len(t.connectedPeers) < t.maxCachedPeers {
t.addServerPeer(server, t.pendingPeers[nodeID].peerInfo)
} else {
t.addToQueue(t.pendingPeers[nodeID].peerInfo)
}
return nil
}
func (t *TopicPool) addServerPeer(server *p2p.Server, info *peerInfo) {
info.added = true
n := enode.NewV4(info.publicKey, info.node.IP, int(info.node.TCP), int(info.node.UDP))
server.AddPeer(n)
}
func (t *TopicPool) removeServerPeer(server *p2p.Server, info *peerInfo) {
info.added = false
n := enode.NewV4(info.publicKey, info.node.IP, int(info.node.TCP), int(info.node.UDP))
server.RemovePeer(n)
}
func (t *TopicPool) isStopped() bool {
t.mu.Lock()
defer t.mu.Unlock()
return t.currentMode == 0
}
// StopSearch stops the closes stop
func (t *TopicPool) StopSearch(server *p2p.Server) {
if !atomic.CompareAndSwapInt32(&t.running, 1, 0) {
return
}
if t.quit == nil {
return
}
select {
case <-t.quit:
return
default:
}
log.Debug("stoping search", "topic", t.topic)
close(t.quit)
t.mu.Lock()
if t.fastModeTimeoutCancel != nil {
close(t.fastModeTimeoutCancel)
t.fastModeTimeoutCancel = nil
}
t.currentMode = 0
t.mu.Unlock()
// wait for poolWG to exit because it writes to period channel
t.poolWG.Wait()
close(t.period)
t.discWG.Wait()
}
// Topic exposes the internal discovery topic.
func (t *TopicPool) Topic() discv5.Topic {
return t.topic
}
// SetLimits set the limits for the current TopicPool.
func (t *TopicPool) SetLimits(limits params.Limits) {
t.mu.Lock()
defer t.mu.Unlock()
t.limits = limits
}