go-libp2p-pubsub/topic.go
Marco Munizaga 9e5145fb29
Send IDONTWANT before first publish (#612)
See #610 

We previously send IDONTWANT only when forwarding. This has us send
IDONTWANT on our initial publish as well. Helps in the case that one or
more peers may also publish the same thing at around the same time (see
#610 for a longer explanation) and prevents "boomerang" duplicates where
a peer sends you back the message you sent before you get a chance to
send it to them.

This also serves as a hint to a peer that you are about to send them a
certain message.
2025-05-19 17:02:21 -07:00

533 lines
12 KiB
Go

package pubsub
import (
"context"
"errors"
"fmt"
"sync"
"time"
pb "github.com/libp2p/go-libp2p-pubsub/pb"
"github.com/libp2p/go-libp2p/core/crypto"
"github.com/libp2p/go-libp2p/core/peer"
)
// ErrTopicClosed is returned if a Topic is utilized after it has been closed
var ErrTopicClosed = errors.New("this Topic is closed, try opening a new one")
// ErrNilSignKey is returned if a nil private key was provided
var ErrNilSignKey = errors.New("nil sign key")
// ErrEmptyPeerID is returned if an empty peer ID was provided
var ErrEmptyPeerID = errors.New("empty peer ID")
// Topic is the handle for a pubsub topic
type Topic struct {
p *PubSub
topic string
evtHandlerMux sync.RWMutex
evtHandlers map[*TopicEventHandler]struct{}
mux sync.RWMutex
closed bool
}
// String returns the topic associated with t
func (t *Topic) String() string {
return t.topic
}
// SetScoreParams sets the topic score parameters if the pubsub router supports peer
// scoring
func (t *Topic) SetScoreParams(p *TopicScoreParams) error {
err := p.validate()
if err != nil {
return fmt.Errorf("invalid topic score parameters: %w", err)
}
t.mux.Lock()
defer t.mux.Unlock()
if t.closed {
return ErrTopicClosed
}
result := make(chan error, 1)
update := func() {
gs, ok := t.p.rt.(*GossipSubRouter)
if !ok {
result <- fmt.Errorf("pubsub router is not gossipsub")
return
}
if gs.score == nil {
result <- fmt.Errorf("peer scoring is not enabled in router")
return
}
err := gs.score.SetTopicScoreParams(t.topic, p)
result <- err
}
select {
case t.p.eval <- update:
err = <-result
return err
case <-t.p.ctx.Done():
return t.p.ctx.Err()
}
}
// EventHandler creates a handle for topic specific events
// Multiple event handlers may be created and will operate independently of each other
func (t *Topic) EventHandler(opts ...TopicEventHandlerOpt) (*TopicEventHandler, error) {
t.mux.RLock()
defer t.mux.RUnlock()
if t.closed {
return nil, ErrTopicClosed
}
h := &TopicEventHandler{
topic: t,
err: nil,
evtLog: make(map[peer.ID]EventType),
evtLogCh: make(chan struct{}, 1),
}
for _, opt := range opts {
err := opt(h)
if err != nil {
return nil, err
}
}
done := make(chan struct{}, 1)
select {
case t.p.eval <- func() {
tmap := t.p.topics[t.topic]
for p := range tmap {
h.evtLog[p] = PeerJoin
}
t.evtHandlerMux.Lock()
t.evtHandlers[h] = struct{}{}
t.evtHandlerMux.Unlock()
done <- struct{}{}
}:
case <-t.p.ctx.Done():
return nil, t.p.ctx.Err()
}
<-done
return h, nil
}
func (t *Topic) sendNotification(evt PeerEvent) {
t.evtHandlerMux.RLock()
defer t.evtHandlerMux.RUnlock()
for h := range t.evtHandlers {
h.sendNotification(evt)
}
}
// Subscribe returns a new Subscription for the topic.
// Note that subscription is not an instantaneous operation. It may take some time
// before the subscription is processed by the pubsub main loop and propagated to our peers.
func (t *Topic) Subscribe(opts ...SubOpt) (*Subscription, error) {
t.mux.RLock()
defer t.mux.RUnlock()
if t.closed {
return nil, ErrTopicClosed
}
sub := &Subscription{
topic: t.topic,
ctx: t.p.ctx,
}
for _, opt := range opts {
err := opt(sub)
if err != nil {
return nil, err
}
}
if sub.ch == nil {
// apply the default size
sub.ch = make(chan *Message, 32)
}
out := make(chan *Subscription, 1)
t.p.disc.Discover(sub.topic)
select {
case t.p.addSub <- &addSubReq{
sub: sub,
resp: out,
}:
case <-t.p.ctx.Done():
return nil, t.p.ctx.Err()
}
return <-out, nil
}
// Relay enables message relaying for the topic and returns a reference
// cancel function. Subsequent calls increase the reference counter.
// To completely disable the relay, all references must be cancelled.
func (t *Topic) Relay() (RelayCancelFunc, error) {
t.mux.RLock()
defer t.mux.RUnlock()
if t.closed {
return nil, ErrTopicClosed
}
out := make(chan RelayCancelFunc, 1)
t.p.disc.Discover(t.topic)
select {
case t.p.addRelay <- &addRelayReq{
topic: t.topic,
resp: out,
}:
case <-t.p.ctx.Done():
return nil, t.p.ctx.Err()
}
return <-out, nil
}
// RouterReady is a function that decides if a router is ready to publish
type RouterReady func(rt PubSubRouter, topic string) (bool, error)
// ProvideKey is a function that provides a private key and its associated peer ID when publishing a new message
type ProvideKey func() (crypto.PrivKey, peer.ID)
type PublishOptions struct {
ready RouterReady
customKey ProvideKey
local bool
validatorData any
}
type BatchPublishOptions struct {
Strategy RPCScheduler
}
type PubOpt func(pub *PublishOptions) error
type BatchPubOpt func(pub *BatchPublishOptions) error
func setDefaultBatchPublishOptions(opts *BatchPublishOptions) {
if opts.Strategy == nil {
opts.Strategy = &RoundRobinMessageIDScheduler{}
}
}
// Publish publishes data to topic.
func (t *Topic) Publish(ctx context.Context, data []byte, opts ...PubOpt) error {
msg, err := t.validate(ctx, data, opts...)
if err != nil {
if errors.Is(err, dupeErr{}) {
// If it was a duplicate, we return nil to indicate success.
// Semantically the message was published by us or someone else.
return nil
}
return err
}
return t.p.val.sendMsgBlocking(msg)
}
func (t *Topic) AddToBatch(ctx context.Context, batch *MessageBatch, data []byte, opts ...PubOpt) error {
msg, err := t.validate(ctx, data, opts...)
if err != nil {
if errors.Is(err, dupeErr{}) {
// If it was a duplicate, we return nil to indicate success.
// Semantically the message was published by us or someone else.
// We won't add it to the batch. Since it's already been published.
return nil
}
return err
}
batch.messages = append(batch.messages, msg)
return nil
}
func (t *Topic) validate(ctx context.Context, data []byte, opts ...PubOpt) (*Message, error) {
t.mux.RLock()
defer t.mux.RUnlock()
if t.closed {
return nil, ErrTopicClosed
}
pid := t.p.signID
key := t.p.signKey
pub := &PublishOptions{}
for _, opt := range opts {
err := opt(pub)
if err != nil {
return nil, err
}
}
if pub.customKey != nil && !pub.local {
key, pid = pub.customKey()
if key == nil {
return nil, ErrNilSignKey
}
if len(pid) == 0 {
return nil, ErrEmptyPeerID
}
}
m := &pb.Message{
Data: data,
Topic: &t.topic,
From: nil,
Seqno: nil,
}
if pid != "" {
m.From = []byte(pid)
m.Seqno = t.p.nextSeqno()
}
if key != nil {
m.From = []byte(pid)
err := signMessage(pid, key, m)
if err != nil {
return nil, err
}
}
if pub.ready != nil {
if t.p.disc.discovery != nil {
t.p.disc.Bootstrap(ctx, t.topic, pub.ready)
} else {
// TODO: we could likely do better than polling every 200ms.
// For example, block this goroutine on a channel,
// and check again whenever events tell us that the number of
// peers has increased.
var ticker *time.Ticker
readyLoop:
for {
// Check if ready for publishing.
// Similar to what disc.Bootstrap does.
res := make(chan bool, 1)
select {
case t.p.eval <- func() {
done, _ := pub.ready(t.p.rt, t.topic)
res <- done
}:
if <-res {
break readyLoop
}
case <-t.p.ctx.Done():
return nil, t.p.ctx.Err()
case <-ctx.Done():
return nil, ctx.Err()
}
if ticker == nil {
ticker = time.NewTicker(200 * time.Millisecond)
defer ticker.Stop()
}
select {
case <-ticker.C:
case <-ctx.Done():
return nil, fmt.Errorf("router is not ready: %w", ctx.Err())
}
}
}
}
msg := &Message{m, "", t.p.host.ID(), pub.validatorData, pub.local}
t.p.rt.Preprocess(t.p.host.ID(), []*Message{msg})
err := t.p.val.ValidateLocal(msg)
if err != nil {
return nil, err
}
return msg, nil
}
// WithReadiness returns a publishing option for only publishing when the router is ready.
// This option is not useful unless PubSub is also using WithDiscovery
func WithReadiness(ready RouterReady) PubOpt {
return func(pub *PublishOptions) error {
pub.ready = ready
return nil
}
}
// WithLocalPublication returns a publishing option to notify in-process subscribers only.
// It prevents message publication to mesh peers.
// Useful in edge cases where the msg needs to be only delivered to the in-process subscribers,
// e.g. not to spam the network with outdated msgs.
// Should not be used specifically for in-process pubsubing.
func WithLocalPublication(local bool) PubOpt {
return func(pub *PublishOptions) error {
pub.local = local
return nil
}
}
// WithValidatorData returns a publishing option to set custom validator data for the message.
// This allows users to avoid deserialization of the message data when validating the message locally.
func WithValidatorData(data any) PubOpt {
return func(pub *PublishOptions) error {
pub.validatorData = data
return nil
}
}
// WithSecretKeyAndPeerId returns a publishing option for providing a custom private key and its corresponding peer ID
// This option is useful when we want to send messages from "virtual", never-connectable peers in the network
func WithSecretKeyAndPeerId(key crypto.PrivKey, pid peer.ID) PubOpt {
return func(pub *PublishOptions) error {
pub.customKey = func() (crypto.PrivKey, peer.ID) {
return key, pid
}
return nil
}
}
// Close closes down the topic. Will return an error unless there are no active event handlers or subscriptions.
// Does not error if the topic is already closed.
func (t *Topic) Close() error {
t.mux.Lock()
defer t.mux.Unlock()
if t.closed {
return nil
}
req := &rmTopicReq{t, make(chan error, 1)}
select {
case t.p.rmTopic <- req:
case <-t.p.ctx.Done():
return t.p.ctx.Err()
}
err := <-req.resp
if err == nil {
t.closed = true
}
return err
}
// ListPeers returns a list of peers we are connected to in the given topic.
func (t *Topic) ListPeers() []peer.ID {
t.mux.RLock()
defer t.mux.RUnlock()
if t.closed {
return []peer.ID{}
}
return t.p.ListPeers(t.topic)
}
type EventType int
const (
PeerJoin EventType = iota
PeerLeave
)
// TopicEventHandler is used to manage topic specific events. No Subscription is required to receive events.
type TopicEventHandler struct {
topic *Topic
err error
evtLogMx sync.Mutex
evtLog map[peer.ID]EventType
evtLogCh chan struct{}
}
type TopicEventHandlerOpt func(t *TopicEventHandler) error
type PeerEvent struct {
Type EventType
Peer peer.ID
}
// Cancel closes the topic event handler
func (t *TopicEventHandler) Cancel() {
topic := t.topic
t.err = fmt.Errorf("topic event handler cancelled by calling handler.Cancel()")
topic.evtHandlerMux.Lock()
delete(topic.evtHandlers, t)
t.topic.evtHandlerMux.Unlock()
}
func (t *TopicEventHandler) sendNotification(evt PeerEvent) {
t.evtLogMx.Lock()
t.addToEventLog(evt)
t.evtLogMx.Unlock()
}
// addToEventLog assumes a lock has been taken to protect the event log
func (t *TopicEventHandler) addToEventLog(evt PeerEvent) {
e, ok := t.evtLog[evt.Peer]
if !ok {
t.evtLog[evt.Peer] = evt.Type
// send signal that an event has been added to the event log
select {
case t.evtLogCh <- struct{}{}:
default:
}
} else if e != evt.Type {
delete(t.evtLog, evt.Peer)
}
}
// pullFromEventLog assumes a lock has been taken to protect the event log
func (t *TopicEventHandler) pullFromEventLog() (PeerEvent, bool) {
for k, v := range t.evtLog {
evt := PeerEvent{Peer: k, Type: v}
delete(t.evtLog, k)
return evt, true
}
return PeerEvent{}, false
}
// NextPeerEvent returns the next event regarding subscribed peers
// Guarantees: Peer Join and Peer Leave events for a given peer will fire in order.
// Unless a peer both Joins and Leaves before NextPeerEvent emits either event
// all events will eventually be received from NextPeerEvent.
func (t *TopicEventHandler) NextPeerEvent(ctx context.Context) (PeerEvent, error) {
for {
t.evtLogMx.Lock()
evt, ok := t.pullFromEventLog()
if ok {
// make sure an event log signal is available if there are events in the event log
if len(t.evtLog) > 0 {
select {
case t.evtLogCh <- struct{}{}:
default:
}
}
t.evtLogMx.Unlock()
return evt, nil
}
t.evtLogMx.Unlock()
select {
case <-t.evtLogCh:
continue
case <-ctx.Done():
return PeerEvent{}, ctx.Err()
}
}
}