Use pull consumers (#2140)

* Pull consumers

* Pull consumers

* Only nuke consumers if they are push consumers

* Clean up old consumers

* Better error handling

* Update comments
This commit is contained in:
Neil Alexander 2022-02-02 13:32:48 +00:00 committed by GitHub
parent 2dee706f9e
commit c773b038bb
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
11 changed files with 521 additions and 459 deletions

View file

@ -34,7 +34,7 @@ import (
type OutputEDUConsumer struct {
ctx context.Context
jetstream nats.JetStreamContext
durable nats.SubOpt
durable string
db storage.Database
queues *queue.OutgoingQueues
ServerName gomatrixserverlib.ServerName
@ -66,13 +66,22 @@ func NewOutputEDUConsumer(
// Start consuming from EDU servers
func (t *OutputEDUConsumer) Start() error {
if _, err := t.jetstream.Subscribe(t.typingTopic, t.onTypingEvent, t.durable); err != nil {
if err := jetstream.JetStreamConsumer(
t.ctx, t.jetstream, t.typingTopic, t.durable, t.onTypingEvent,
nats.DeliverAll(), nats.ManualAck(),
); err != nil {
return err
}
if _, err := t.jetstream.Subscribe(t.sendToDeviceTopic, t.onSendToDeviceEvent, t.durable); err != nil {
if err := jetstream.JetStreamConsumer(
t.ctx, t.jetstream, t.sendToDeviceTopic, t.durable, t.onSendToDeviceEvent,
nats.DeliverAll(), nats.ManualAck(),
); err != nil {
return err
}
if _, err := t.jetstream.Subscribe(t.receiptTopic, t.onReceiptEvent, t.durable); err != nil {
if err := jetstream.JetStreamConsumer(
t.ctx, t.jetstream, t.receiptTopic, t.durable, t.onReceiptEvent,
nats.DeliverAll(), nats.ManualAck(),
); err != nil {
return err
}
return nil
@ -80,175 +89,169 @@ func (t *OutputEDUConsumer) Start() error {
// onSendToDeviceEvent is called in response to a message received on the
// send-to-device events topic from the EDU server.
func (t *OutputEDUConsumer) onSendToDeviceEvent(msg *nats.Msg) {
func (t *OutputEDUConsumer) onSendToDeviceEvent(ctx context.Context, msg *nats.Msg) bool {
// Extract the send-to-device event from msg.
jetstream.WithJetStreamMessage(msg, func(msg *nats.Msg) bool {
var ote api.OutputSendToDeviceEvent
if err := json.Unmarshal(msg.Data, &ote); err != nil {
log.WithError(err).Errorf("eduserver output log: message parse failed (expected send-to-device)")
return true
}
// only send send-to-device events which originated from us
_, originServerName, err := gomatrixserverlib.SplitID('@', ote.Sender)
if err != nil {
log.WithError(err).WithField("user_id", ote.Sender).Error("Failed to extract domain from send-to-device sender")
return true
}
if originServerName != t.ServerName {
log.WithField("other_server", originServerName).Info("Suppressing send-to-device: originated elsewhere")
return true
}
_, destServerName, err := gomatrixserverlib.SplitID('@', ote.UserID)
if err != nil {
log.WithError(err).WithField("user_id", ote.UserID).Error("Failed to extract domain from send-to-device destination")
return true
}
// Pack the EDU and marshal it
edu := &gomatrixserverlib.EDU{
Type: gomatrixserverlib.MDirectToDevice,
Origin: string(t.ServerName),
}
tdm := gomatrixserverlib.ToDeviceMessage{
Sender: ote.Sender,
Type: ote.Type,
MessageID: util.RandomString(32),
Messages: map[string]map[string]json.RawMessage{
ote.UserID: {
ote.DeviceID: ote.Content,
},
},
}
if edu.Content, err = json.Marshal(tdm); err != nil {
log.WithError(err).Error("failed to marshal EDU JSON")
return true
}
log.Infof("Sending send-to-device message into %q destination queue", destServerName)
if err := t.queues.SendEDU(edu, t.ServerName, []gomatrixserverlib.ServerName{destServerName}); err != nil {
log.WithError(err).Error("failed to send EDU")
return false
}
var ote api.OutputSendToDeviceEvent
if err := json.Unmarshal(msg.Data, &ote); err != nil {
log.WithError(err).Errorf("eduserver output log: message parse failed (expected send-to-device)")
return true
})
}
// only send send-to-device events which originated from us
_, originServerName, err := gomatrixserverlib.SplitID('@', ote.Sender)
if err != nil {
log.WithError(err).WithField("user_id", ote.Sender).Error("Failed to extract domain from send-to-device sender")
return true
}
if originServerName != t.ServerName {
log.WithField("other_server", originServerName).Info("Suppressing send-to-device: originated elsewhere")
return true
}
_, destServerName, err := gomatrixserverlib.SplitID('@', ote.UserID)
if err != nil {
log.WithError(err).WithField("user_id", ote.UserID).Error("Failed to extract domain from send-to-device destination")
return true
}
// Pack the EDU and marshal it
edu := &gomatrixserverlib.EDU{
Type: gomatrixserverlib.MDirectToDevice,
Origin: string(t.ServerName),
}
tdm := gomatrixserverlib.ToDeviceMessage{
Sender: ote.Sender,
Type: ote.Type,
MessageID: util.RandomString(32),
Messages: map[string]map[string]json.RawMessage{
ote.UserID: {
ote.DeviceID: ote.Content,
},
},
}
if edu.Content, err = json.Marshal(tdm); err != nil {
log.WithError(err).Error("failed to marshal EDU JSON")
return true
}
log.Infof("Sending send-to-device message into %q destination queue", destServerName)
if err := t.queues.SendEDU(edu, t.ServerName, []gomatrixserverlib.ServerName{destServerName}); err != nil {
log.WithError(err).Error("failed to send EDU")
return false
}
return true
}
// onTypingEvent is called in response to a message received on the typing
// events topic from the EDU server.
func (t *OutputEDUConsumer) onTypingEvent(msg *nats.Msg) {
jetstream.WithJetStreamMessage(msg, func(msg *nats.Msg) bool {
// Extract the typing event from msg.
var ote api.OutputTypingEvent
if err := json.Unmarshal(msg.Data, &ote); err != nil {
// Skip this msg but continue processing messages.
log.WithError(err).Errorf("eduserver output log: message parse failed (expected typing)")
_ = msg.Ack()
return true
}
// only send typing events which originated from us
_, typingServerName, err := gomatrixserverlib.SplitID('@', ote.Event.UserID)
if err != nil {
log.WithError(err).WithField("user_id", ote.Event.UserID).Error("Failed to extract domain from typing sender")
_ = msg.Ack()
return true
}
if typingServerName != t.ServerName {
return true
}
joined, err := t.db.GetJoinedHosts(t.ctx, ote.Event.RoomID)
if err != nil {
log.WithError(err).WithField("room_id", ote.Event.RoomID).Error("failed to get joined hosts for room")
return false
}
names := make([]gomatrixserverlib.ServerName, len(joined))
for i := range joined {
names[i] = joined[i].ServerName
}
edu := &gomatrixserverlib.EDU{Type: ote.Event.Type}
if edu.Content, err = json.Marshal(map[string]interface{}{
"room_id": ote.Event.RoomID,
"user_id": ote.Event.UserID,
"typing": ote.Event.Typing,
}); err != nil {
log.WithError(err).Error("failed to marshal EDU JSON")
return true
}
if err := t.queues.SendEDU(edu, t.ServerName, names); err != nil {
log.WithError(err).Error("failed to send EDU")
return false
}
func (t *OutputEDUConsumer) onTypingEvent(ctx context.Context, msg *nats.Msg) bool {
// Extract the typing event from msg.
var ote api.OutputTypingEvent
if err := json.Unmarshal(msg.Data, &ote); err != nil {
// Skip this msg but continue processing messages.
log.WithError(err).Errorf("eduserver output log: message parse failed (expected typing)")
_ = msg.Ack()
return true
})
}
// only send typing events which originated from us
_, typingServerName, err := gomatrixserverlib.SplitID('@', ote.Event.UserID)
if err != nil {
log.WithError(err).WithField("user_id", ote.Event.UserID).Error("Failed to extract domain from typing sender")
_ = msg.Ack()
return true
}
if typingServerName != t.ServerName {
return true
}
joined, err := t.db.GetJoinedHosts(ctx, ote.Event.RoomID)
if err != nil {
log.WithError(err).WithField("room_id", ote.Event.RoomID).Error("failed to get joined hosts for room")
return false
}
names := make([]gomatrixserverlib.ServerName, len(joined))
for i := range joined {
names[i] = joined[i].ServerName
}
edu := &gomatrixserverlib.EDU{Type: ote.Event.Type}
if edu.Content, err = json.Marshal(map[string]interface{}{
"room_id": ote.Event.RoomID,
"user_id": ote.Event.UserID,
"typing": ote.Event.Typing,
}); err != nil {
log.WithError(err).Error("failed to marshal EDU JSON")
return true
}
if err := t.queues.SendEDU(edu, t.ServerName, names); err != nil {
log.WithError(err).Error("failed to send EDU")
return false
}
return true
}
// onReceiptEvent is called in response to a message received on the receipt
// events topic from the EDU server.
func (t *OutputEDUConsumer) onReceiptEvent(msg *nats.Msg) {
jetstream.WithJetStreamMessage(msg, func(msg *nats.Msg) bool {
// Extract the typing event from msg.
var receipt api.OutputReceiptEvent
if err := json.Unmarshal(msg.Data, &receipt); err != nil {
// Skip this msg but continue processing messages.
log.WithError(err).Errorf("eduserver output log: message parse failed (expected receipt)")
return true
}
// only send receipt events which originated from us
_, receiptServerName, err := gomatrixserverlib.SplitID('@', receipt.UserID)
if err != nil {
log.WithError(err).WithField("user_id", receipt.UserID).Error("failed to extract domain from receipt sender")
return true
}
if receiptServerName != t.ServerName {
return true
}
joined, err := t.db.GetJoinedHosts(t.ctx, receipt.RoomID)
if err != nil {
log.WithError(err).WithField("room_id", receipt.RoomID).Error("failed to get joined hosts for room")
return false
}
names := make([]gomatrixserverlib.ServerName, len(joined))
for i := range joined {
names[i] = joined[i].ServerName
}
content := map[string]api.FederationReceiptMRead{}
content[receipt.RoomID] = api.FederationReceiptMRead{
User: map[string]api.FederationReceiptData{
receipt.UserID: {
Data: api.ReceiptTS{
TS: receipt.Timestamp,
},
EventIDs: []string{receipt.EventID},
},
},
}
edu := &gomatrixserverlib.EDU{
Type: gomatrixserverlib.MReceipt,
Origin: string(t.ServerName),
}
if edu.Content, err = json.Marshal(content); err != nil {
log.WithError(err).Error("failed to marshal EDU JSON")
return true
}
if err := t.queues.SendEDU(edu, t.ServerName, names); err != nil {
log.WithError(err).Error("failed to send EDU")
return false
}
func (t *OutputEDUConsumer) onReceiptEvent(ctx context.Context, msg *nats.Msg) bool {
// Extract the typing event from msg.
var receipt api.OutputReceiptEvent
if err := json.Unmarshal(msg.Data, &receipt); err != nil {
// Skip this msg but continue processing messages.
log.WithError(err).Errorf("eduserver output log: message parse failed (expected receipt)")
return true
})
}
// only send receipt events which originated from us
_, receiptServerName, err := gomatrixserverlib.SplitID('@', receipt.UserID)
if err != nil {
log.WithError(err).WithField("user_id", receipt.UserID).Error("failed to extract domain from receipt sender")
return true
}
if receiptServerName != t.ServerName {
return true
}
joined, err := t.db.GetJoinedHosts(ctx, receipt.RoomID)
if err != nil {
log.WithError(err).WithField("room_id", receipt.RoomID).Error("failed to get joined hosts for room")
return false
}
names := make([]gomatrixserverlib.ServerName, len(joined))
for i := range joined {
names[i] = joined[i].ServerName
}
content := map[string]api.FederationReceiptMRead{}
content[receipt.RoomID] = api.FederationReceiptMRead{
User: map[string]api.FederationReceiptData{
receipt.UserID: {
Data: api.ReceiptTS{
TS: receipt.Timestamp,
},
EventIDs: []string{receipt.EventID},
},
},
}
edu := &gomatrixserverlib.EDU{
Type: gomatrixserverlib.MReceipt,
Origin: string(t.ServerName),
}
if edu.Content, err = json.Marshal(content); err != nil {
log.WithError(err).Error("failed to marshal EDU JSON")
return true
}
if err := t.queues.SendEDU(edu, t.ServerName, names); err != nil {
log.WithError(err).Error("failed to send EDU")
return false
}
return true
}

View file

@ -37,7 +37,7 @@ type OutputRoomEventConsumer struct {
cfg *config.FederationAPI
rsAPI api.RoomserverInternalAPI
jetstream nats.JetStreamContext
durable nats.SubOpt
durable string
db storage.Database
queues *queue.OutgoingQueues
topic string
@ -66,74 +66,70 @@ func NewOutputRoomEventConsumer(
// Start consuming from room servers
func (s *OutputRoomEventConsumer) Start() error {
_, err := s.jetstream.Subscribe(
s.topic, s.onMessage, s.durable,
nats.DeliverAll(),
nats.ManualAck(),
return jetstream.JetStreamConsumer(
s.ctx, s.jetstream, s.topic, s.durable, s.onMessage,
nats.DeliverAll(), nats.ManualAck(),
)
return err
}
// onMessage is called when the federation server receives a new event from the room server output log.
// It is unsafe to call this with messages for the same room in multiple gorountines
// because updates it will likely fail with a types.EventIDMismatchError when it
// realises that it cannot update the room state using the deltas.
func (s *OutputRoomEventConsumer) onMessage(msg *nats.Msg) {
jetstream.WithJetStreamMessage(msg, func(msg *nats.Msg) bool {
// Parse out the event JSON
var output api.OutputEvent
if err := json.Unmarshal(msg.Data, &output); err != nil {
// If the message was invalid, log it and move on to the next message in the stream
log.WithError(err).Errorf("roomserver output log: message parse failure")
return true
}
func (s *OutputRoomEventConsumer) onMessage(ctx context.Context, msg *nats.Msg) bool {
// Parse out the event JSON
var output api.OutputEvent
if err := json.Unmarshal(msg.Data, &output); err != nil {
// If the message was invalid, log it and move on to the next message in the stream
log.WithError(err).Errorf("roomserver output log: message parse failure")
return true
}
switch output.Type {
case api.OutputTypeNewRoomEvent:
ev := output.NewRoomEvent.Event
switch output.Type {
case api.OutputTypeNewRoomEvent:
ev := output.NewRoomEvent.Event
if output.NewRoomEvent.RewritesState {
if err := s.db.PurgeRoomState(s.ctx, ev.RoomID()); err != nil {
log.WithError(err).Errorf("roomserver output log: purge room state failure")
return false
}
}
if err := s.processMessage(*output.NewRoomEvent); err != nil {
switch err.(type) {
case *queue.ErrorFederationDisabled:
log.WithField("error", output.Type).Info(
err.Error(),
)
default:
// panic rather than continue with an inconsistent database
log.WithFields(log.Fields{
"event_id": ev.EventID(),
"event": string(ev.JSON()),
"add": output.NewRoomEvent.AddsStateEventIDs,
"del": output.NewRoomEvent.RemovesStateEventIDs,
log.ErrorKey: err,
}).Panicf("roomserver output log: write room event failure")
}
}
case api.OutputTypeNewInboundPeek:
if err := s.processInboundPeek(*output.NewInboundPeek); err != nil {
log.WithFields(log.Fields{
"event": output.NewInboundPeek,
log.ErrorKey: err,
}).Panicf("roomserver output log: remote peek event failure")
if output.NewRoomEvent.RewritesState {
if err := s.db.PurgeRoomState(s.ctx, ev.RoomID()); err != nil {
log.WithError(err).Errorf("roomserver output log: purge room state failure")
return false
}
default:
log.WithField("type", output.Type).Debug(
"roomserver output log: ignoring unknown output type",
)
}
return true
})
if err := s.processMessage(*output.NewRoomEvent); err != nil {
switch err.(type) {
case *queue.ErrorFederationDisabled:
log.WithField("error", output.Type).Info(
err.Error(),
)
default:
// panic rather than continue with an inconsistent database
log.WithFields(log.Fields{
"event_id": ev.EventID(),
"event": string(ev.JSON()),
"add": output.NewRoomEvent.AddsStateEventIDs,
"del": output.NewRoomEvent.RemovesStateEventIDs,
log.ErrorKey: err,
}).Panicf("roomserver output log: write room event failure")
}
}
case api.OutputTypeNewInboundPeek:
if err := s.processInboundPeek(*output.NewInboundPeek); err != nil {
log.WithFields(log.Fields{
"event": output.NewInboundPeek,
log.ErrorKey: err,
}).Panicf("roomserver output log: remote peek event failure")
return false
}
default:
log.WithField("type", output.Type).Debug(
"roomserver output log: ignoring unknown output type",
)
}
return true
}
// processInboundPeek starts tracking a new federated inbound peek (replacing the existing one if any)