dendrite/federationapi/consumers/roomserver.go
Neil Alexander a763cbb0e1
Roomserver/federation input refactor (#2104)
* Put federation client functions into their own file

* Look for missing auth events in RS input

* Remove retrieveMissingAuthEvents from federation API

* Logging

* Sorta transplanted the code over

* Use event origin failing all else

* Don't get stuck on mutexes:

* Add verifier

* Don't mark state events with zero snapshot NID as not existing

* Check missing state if not an outlier before storing the event

* Reject instead of soft-fail, don't copy roominfo so much

* Use synchronous contexts, limit time to fetch missing events

* Clean up some commented out bits

* Simplify `/send` endpoint significantly

* Submit async

* Report errors on sending to RS input

* Set max payload in NATS to 16MB

* Tweak metrics

* Add `workerForRoom` for tidiness

* Try skipping unmarshalling errors for RespMissingEvents

* Track missing prev events separately to avoid calculating state when not possible

* Tweak logic around checking missing state

* Care about state when checking missing prev events

* Don't check missing state for create events

* Try that again

* Handle create events better

* Send create room events as new

* Use given event kind when sending auth/state events

* Revert "Use given event kind when sending auth/state events"

This reverts commit 089d64d271b5fca8c104e1554711187420dbebca.

* Only search for missing prev events or state for new events

* Tweaks

* We only have missing prev if we don't supply state

* Room version tweaks

* Allow async inputs again

* Apply backpressure to consumers/synchronous requests to hopefully stop things being overwhelmed

* Set timeouts on roomserver input tasks (need to decide what timeout makes sense)

* Use work queue policy, deliver all on restart

* Reduce chance of duplicates being sent by NATS

* Limit the number of servers we attempt to reduce backpressure

* Some review comment fixes

* Tidy up a couple things

* Don't limit servers, randomise order using map

* Some context refactoring

* Update gmsl

* Don't resend create events

* Set stateIDs length correctly or else the roomserver thinks there are missing events when there aren't

* Exclude our own servername

* Try backing off servers

* Make excluding self behaviour optional

* Exclude self from g_m_e

* Update sytest-whitelist

* Update consumers for the roomserver output stream

* Remember to send outliers for state returned from /gme

* Make full HTTP tests less upsetti

* Remove 'If a device list update goes missing, the server resyncs on the next one' from the sytest blacklist

* Remove debugging test

* Fix blacklist again, remove unnecessary duplicate context

* Clearer contexts, don't use background in case there's something happening there

* Don't queue up events more than once in memory

* Correctly identify create events when checking for state

* Fill in gaps again in /gme code

* Remove `AuthEventIDs` from `InputRoomEvent`

* Remove stray field

Co-authored-by: Kegan Dougal <kegan@matrix.org>
2022-01-27 14:29:14 +00:00

412 lines
13 KiB
Go

// Copyright 2017 Vector Creations Ltd
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
package consumers
import (
"context"
"encoding/json"
"fmt"
"github.com/matrix-org/dendrite/federationapi/queue"
"github.com/matrix-org/dendrite/federationapi/storage"
"github.com/matrix-org/dendrite/federationapi/types"
"github.com/matrix-org/dendrite/roomserver/api"
"github.com/matrix-org/dendrite/setup/config"
"github.com/matrix-org/dendrite/setup/jetstream"
"github.com/matrix-org/dendrite/setup/process"
"github.com/matrix-org/gomatrixserverlib"
"github.com/nats-io/nats.go"
log "github.com/sirupsen/logrus"
)
// OutputRoomEventConsumer consumes events that originated in the room server.
type OutputRoomEventConsumer struct {
ctx context.Context
cfg *config.FederationAPI
rsAPI api.RoomserverInternalAPI
jetstream nats.JetStreamContext
durable nats.SubOpt
db storage.Database
queues *queue.OutgoingQueues
topic string
}
// NewOutputRoomEventConsumer creates a new OutputRoomEventConsumer. Call Start() to begin consuming from room servers.
func NewOutputRoomEventConsumer(
process *process.ProcessContext,
cfg *config.FederationAPI,
js nats.JetStreamContext,
queues *queue.OutgoingQueues,
store storage.Database,
rsAPI api.RoomserverInternalAPI,
) *OutputRoomEventConsumer {
return &OutputRoomEventConsumer{
ctx: process.Context(),
cfg: cfg,
jetstream: js,
db: store,
queues: queues,
rsAPI: rsAPI,
durable: cfg.Matrix.JetStream.Durable("FederationAPIRoomServerConsumer"),
topic: cfg.Matrix.JetStream.TopicFor(jetstream.OutputRoomEvent),
}
}
// 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 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
}
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")
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)
// causing the federationapi to start sending messages to the peeking server
func (s *OutputRoomEventConsumer) processInboundPeek(orp api.OutputNewInboundPeek) error {
// FIXME: there's a race here - we should start /sending new peeked events
// atomically after the orp.LatestEventID to ensure there are no gaps between
// the peek beginning and the send stream beginning.
//
// We probably need to track orp.LatestEventID on the inbound peek, but it's
// unclear how we then use that to prevent the race when we start the send
// stream.
//
// This is making the tests flakey.
return s.db.AddInboundPeek(s.ctx, orp.ServerName, orp.RoomID, orp.PeekID, orp.RenewalInterval)
}
// processMessage updates the list of currently joined hosts in the room
// and then sends the event to the hosts that were joined before the event.
func (s *OutputRoomEventConsumer) processMessage(ore api.OutputNewRoomEvent) error {
addsJoinedHosts, err := joinedHostsFromEvents(gomatrixserverlib.UnwrapEventHeaders(ore.AddsState()))
if err != nil {
return err
}
// Update our copy of the current state.
// We keep a copy of the current state because the state at each event is
// expressed as a delta against the current state.
// TODO(#290): handle EventIDMismatchError and recover the current state by
// talking to the roomserver
oldJoinedHosts, err := s.db.UpdateRoom(
s.ctx,
ore.Event.RoomID(),
ore.LastSentEventID,
ore.Event.EventID(),
addsJoinedHosts,
ore.RemovesStateEventIDs,
)
if err != nil {
return err
}
if oldJoinedHosts == nil {
// This means that there is nothing to update as this is a duplicate
// message.
// This can happen if dendrite crashed between reading the message and
// persisting the stream position.
return nil
}
if ore.SendAsServer == api.DoNotSendToOtherServers {
// Ignore event that we don't need to send anywhere.
return nil
}
// Work out which hosts were joined at the event itself.
joinedHostsAtEvent, err := s.joinedHostsAtEvent(ore, oldJoinedHosts)
if err != nil {
return err
}
// TODO: do housekeeping to evict unrenewed peeking hosts
// TODO: implement query to let the fedapi check whether a given peek is live or not
// Send the event.
return s.queues.SendEvent(
ore.Event, gomatrixserverlib.ServerName(ore.SendAsServer), joinedHostsAtEvent,
)
}
// joinedHostsAtEvent works out a list of matrix servers that were joined to
// the room at the event (including peeking ones)
// It is important to use the state at the event for sending messages because:
// 1) We shouldn't send messages to servers that weren't in the room.
// 2) If a server is kicked from the rooms it should still be told about the
// kick event,
// Usually the list can be calculated locally, but sometimes it will need fetch
// events from the room server.
// Returns an error if there was a problem talking to the room server.
func (s *OutputRoomEventConsumer) joinedHostsAtEvent(
ore api.OutputNewRoomEvent, oldJoinedHosts []types.JoinedHost,
) ([]gomatrixserverlib.ServerName, error) {
// Combine the delta into a single delta so that the adds and removes can
// cancel each other out. This should reduce the number of times we need
// to fetch a state event from the room server.
combinedAdds, combinedRemoves := combineDeltas(
ore.AddsStateEventIDs, ore.RemovesStateEventIDs,
ore.StateBeforeAddsEventIDs, ore.StateBeforeRemovesEventIDs,
)
combinedAddsEvents, err := s.lookupStateEvents(combinedAdds, ore.Event.Event)
if err != nil {
return nil, err
}
combinedAddsJoinedHosts, err := joinedHostsFromEvents(combinedAddsEvents)
if err != nil {
return nil, err
}
removed := map[string]bool{}
for _, eventID := range combinedRemoves {
removed[eventID] = true
}
joined := map[gomatrixserverlib.ServerName]bool{}
for _, joinedHost := range oldJoinedHosts {
if removed[joinedHost.MemberEventID] {
// This m.room.member event is part of the current state of the
// room, but not part of the state at the event we are processing
// Therefore we can't use it to tell whether the server was in
// the room at the event.
continue
}
joined[joinedHost.ServerName] = true
}
for _, joinedHost := range combinedAddsJoinedHosts {
// This m.room.member event was part of the state of the room at the
// event, but isn't part of the current state of the room now.
joined[joinedHost.ServerName] = true
}
// handle peeking hosts
inboundPeeks, err := s.db.GetInboundPeeks(s.ctx, ore.Event.Event.RoomID())
if err != nil {
return nil, err
}
for _, inboundPeek := range inboundPeeks {
joined[inboundPeek.ServerName] = true
}
var result []gomatrixserverlib.ServerName
for serverName, include := range joined {
if include {
result = append(result, serverName)
}
}
return result, nil
}
// joinedHostsFromEvents turns a list of state events into a list of joined hosts.
// This errors if one of the events was invalid.
// It should be impossible for an invalid event to get this far in the pipeline.
func joinedHostsFromEvents(evs []*gomatrixserverlib.Event) ([]types.JoinedHost, error) {
var joinedHosts []types.JoinedHost
for _, ev := range evs {
if ev.Type() != "m.room.member" || ev.StateKey() == nil {
continue
}
membership, err := ev.Membership()
if err != nil {
return nil, err
}
if membership != gomatrixserverlib.Join {
continue
}
_, serverName, err := gomatrixserverlib.SplitID('@', *ev.StateKey())
if err != nil {
return nil, err
}
joinedHosts = append(joinedHosts, types.JoinedHost{
MemberEventID: ev.EventID(), ServerName: serverName,
})
}
return joinedHosts, nil
}
// combineDeltas combines two deltas into a single delta.
// Assumes that the order of operations is add(1), remove(1), add(2), remove(2).
// Removes duplicate entries and redundant operations from each delta.
func combineDeltas(adds1, removes1, adds2, removes2 []string) (adds, removes []string) {
addSet := map[string]bool{}
removeSet := map[string]bool{}
// combine processes each unique value in a list.
// If the value is in the removeFrom set then it is removed from that set.
// Otherwise it is added to the addTo set.
combine := func(values []string, removeFrom, addTo map[string]bool) {
processed := map[string]bool{}
for _, value := range values {
if processed[value] {
continue
}
processed[value] = true
if removeFrom[value] {
delete(removeFrom, value)
} else {
addTo[value] = true
}
}
}
combine(adds1, nil, addSet)
combine(removes1, addSet, removeSet)
combine(adds2, removeSet, addSet)
combine(removes2, addSet, removeSet)
for value := range addSet {
adds = append(adds, value)
}
for value := range removeSet {
removes = append(removes, value)
}
return
}
// lookupStateEvents looks up the state events that are added by a new event.
func (s *OutputRoomEventConsumer) lookupStateEvents(
addsStateEventIDs []string, event *gomatrixserverlib.Event,
) ([]*gomatrixserverlib.Event, error) {
// Fast path if there aren't any new state events.
if len(addsStateEventIDs) == 0 {
return nil, nil
}
// Fast path if the only state event added is the event itself.
if len(addsStateEventIDs) == 1 && addsStateEventIDs[0] == event.EventID() {
return []*gomatrixserverlib.Event{event}, nil
}
missing := addsStateEventIDs
var result []*gomatrixserverlib.Event
// Check if event itself is being added.
for _, eventID := range missing {
if eventID == event.EventID() {
result = append(result, event)
break
}
}
missing = missingEventsFrom(result, addsStateEventIDs)
if len(missing) == 0 {
return result, nil
}
// At this point the missing events are neither the event itself nor are
// they present in our local database. Our only option is to fetch them
// from the roomserver using the query API.
eventReq := api.QueryEventsByIDRequest{EventIDs: missing}
var eventResp api.QueryEventsByIDResponse
if err := s.rsAPI.QueryEventsByID(s.ctx, &eventReq, &eventResp); err != nil {
return nil, err
}
for _, headeredEvent := range eventResp.Events {
result = append(result, headeredEvent.Event)
}
missing = missingEventsFrom(result, addsStateEventIDs)
if len(missing) != 0 {
return nil, fmt.Errorf(
"missing %d state events IDs at event %q", len(missing), event.EventID(),
)
}
return result, nil
}
func missingEventsFrom(events []*gomatrixserverlib.Event, required []string) []string {
have := map[string]bool{}
for _, event := range events {
have[event.EventID()] = true
}
var missing []string
for _, eventID := range required {
if !have[eventID] {
missing = append(missing, eventID)
}
}
return missing
}