Make dendritejs work again

This commit is contained in:
Kegan Dougal 2020-07-03 11:15:17 +01:00
parent cd3da772de
commit 31ee8c9201
4 changed files with 104 additions and 145 deletions

View file

@ -5,7 +5,7 @@
# $ docker build -t dendritejs -f DendriteJS.Dockerfile .
# $ docker run --rm -p 8888:80 dendritejs
# Then visit http://localhost:8888
FROM golang:1.13.7-alpine3.11 AS gobuild
FROM golang:1.14-alpine AS gobuild
# Download and build dendrite
WORKDIR /build

View file

@ -67,146 +67,6 @@ func GetPostPublicRooms(
}
}
// GetPostPublicRoomsWithExternal is the same as GetPostPublicRooms but also mixes in public rooms from the provider supplied.
func GetPostPublicRoomsWithExternal(
req *http.Request, rsAPI roomserverAPI.RoomserverInternalAPI, stateAPI currentstateAPI.CurrentStateInternalAPI,
fedClient *gomatrixserverlib.FederationClient, extRoomsProvider api.ExtraPublicRoomsProvider,
) util.JSONResponse {
var request PublicRoomReq
if fillErr := fillPublicRoomsReq(req, &request); fillErr != nil {
return *fillErr
}
response, err := publicRooms(req.Context(), request, rsAPI, stateAPI, extRoomsProvider)
if err != nil {
return jsonerror.InternalServerError()
}
if request.Since != "" {
// TODO: handle pagination tokens sensibly rather than ignoring them.
// ignore paginated requests since we don't handle them yet over federation.
// Only the initial request will contain federated rooms.
return util.JSONResponse{
Code: http.StatusOK,
JSON: response,
}
}
// If we have already hit the limit on the number of rooms, bail.
var limit int
if request.Limit > 0 {
limit = int(request.Limit) - len(response.Chunk)
if limit <= 0 {
return util.JSONResponse{
Code: http.StatusOK,
JSON: response,
}
}
}
// downcasting `limit` is safe as we know it isn't bigger than request.Limit which is int16
//fedRooms := bulkFetchPublicRoomsFromServers(req.Context(), fedClient, extRoomsProvider.Homeservers(), int16(limit))
//response.Chunk = append(response.Chunk, fedRooms...)
// de-duplicate rooms with the same room ID. We can join the room via any of these aliases as we know these servers
// are alive and well, so we arbitrarily pick one (purposefully shuffling them to spread the load a bit)
var publicRooms []gomatrixserverlib.PublicRoom
haveRoomIDs := make(map[string]bool)
rand.Shuffle(len(response.Chunk), func(i, j int) {
response.Chunk[i], response.Chunk[j] = response.Chunk[j], response.Chunk[i]
})
for _, r := range response.Chunk {
if haveRoomIDs[r.RoomID] {
continue
}
haveRoomIDs[r.RoomID] = true
publicRooms = append(publicRooms, r)
}
// sort by member count
sort.SliceStable(publicRooms, func(i, j int) bool {
return publicRooms[i].JoinedMembersCount > publicRooms[j].JoinedMembersCount
})
response.Chunk = publicRooms
return util.JSONResponse{
Code: http.StatusOK,
JSON: response,
}
}
/*
// bulkFetchPublicRoomsFromServers fetches public rooms from the list of homeservers.
// Returns a list of public rooms up to the limit specified.
func bulkFetchPublicRoomsFromServers(
ctx context.Context, fedClient *gomatrixserverlib.FederationClient, homeservers []string, limit int16,
) (publicRooms []gomatrixserverlib.PublicRoom) {
// follow pipeline semantics, see https://blog.golang.org/pipelines for more info.
// goroutines send rooms to this channel
roomCh := make(chan gomatrixserverlib.PublicRoom, int(limit))
// signalling channel to tell goroutines to stop sending rooms and quit
done := make(chan bool)
// signalling to say when we can close the room channel
var wg sync.WaitGroup
wg.Add(len(homeservers))
// concurrently query for public rooms
for _, hs := range homeservers {
go func(homeserverDomain string) {
defer wg.Done()
util.GetLogger(ctx).WithField("hs", homeserverDomain).Info("Querying HS for public rooms")
fres, err := fedClient.GetPublicRooms(ctx, gomatrixserverlib.ServerName(homeserverDomain), int(limit), "", false, "")
if err != nil {
util.GetLogger(ctx).WithError(err).WithField("hs", homeserverDomain).Warn(
"bulkFetchPublicRoomsFromServers: failed to query hs",
)
return
}
for _, room := range fres.Chunk {
// atomically send a room or stop
select {
case roomCh <- room:
case <-done:
util.GetLogger(ctx).WithError(err).WithField("hs", homeserverDomain).Info("Interrupted whilst sending rooms")
return
}
}
}(hs)
}
// Close the room channel when the goroutines have quit so we don't leak, but don't let it stop the in-flight request.
// This also allows the request to fail fast if all HSes experience errors as it will cause the room channel to be
// closed.
go func() {
wg.Wait()
util.GetLogger(ctx).Info("Cleaning up resources")
close(roomCh)
}()
// fan-in results with timeout. We stop when we reach the limit.
FanIn:
for len(publicRooms) < int(limit) || limit == 0 {
// add a room or timeout
select {
case room, ok := <-roomCh:
if !ok {
util.GetLogger(ctx).Info("All homeservers have been queried, returning results.")
break FanIn
}
publicRooms = append(publicRooms, room)
case <-time.After(15 * time.Second): // we've waited long enough, let's tell the client what we got.
util.GetLogger(ctx).Info("Waited 15s for federated public rooms, returning early")
break FanIn
case <-ctx.Done(): // the client hung up on us, let's stop.
util.GetLogger(ctx).Info("Client hung up, returning early")
break FanIn
}
}
// tell goroutines to stop
close(done)
return publicRooms
}
*/
func publicRooms(ctx context.Context, request PublicRoomReq, rsAPI roomserverAPI.RoomserverInternalAPI,
stateAPI currentstateAPI.CurrentStateInternalAPI, extRoomsProvider api.ExtraPublicRoomsProvider,
) (*gomatrixserverlib.RespPublicRooms, error) {
@ -408,6 +268,8 @@ func refreshPublicRoomCache(
publicRoomsCache = []gomatrixserverlib.PublicRoom{}
publicRoomsCache = append(publicRoomsCache, pubRooms...)
publicRoomsCache = append(publicRoomsCache, extraRooms...)
publicRoomsCache = dedupeAndShuffle(publicRoomsCache)
// sort by total joined member count (big to small)
sort.SliceStable(publicRoomsCache, func(i, j int) bool {
return publicRoomsCache[i].JoinedMembersCount > publicRoomsCache[j].JoinedMembersCount
@ -420,3 +282,21 @@ func getPublicRoomsFromCache() []gomatrixserverlib.PublicRoom {
defer cacheMu.Unlock()
return publicRoomsCache
}
func dedupeAndShuffle(in []gomatrixserverlib.PublicRoom) []gomatrixserverlib.PublicRoom {
// de-duplicate rooms with the same room ID. We can join the room via any of these aliases as we know these servers
// are alive and well, so we arbitrarily pick one (purposefully shuffling them to spread the load a bit)
var publicRooms []gomatrixserverlib.PublicRoom
haveRoomIDs := make(map[string]bool)
rand.Shuffle(len(in), func(i, j int) {
in[i], in[j] = in[j], in[i]
})
for _, r := range in {
if haveRoomIDs[r.RoomID] {
continue
}
haveRoomIDs[r.RoomID] = true
publicRooms = append(publicRooms, r)
}
return publicRooms
}

View file

@ -211,7 +211,7 @@ func main() {
)
fedSenderAPI := federationsender.NewInternalAPI(base, federation, rsAPI, &keyRing)
rsAPI.SetFederationSenderAPI(fedSenderAPI)
p2pPublicRoomProvider := NewLibP2PPublicRoomsProvider(node, fedSenderAPI)
p2pPublicRoomProvider := NewLibP2PPublicRoomsProvider(node, fedSenderAPI, federation)
stateAPI := currentstateserver.NewInternalAPI(base.Cfg, base.KafkaConsumer)

View file

@ -18,22 +18,29 @@ package main
import (
"context"
"sync"
"time"
"github.com/matrix-org/dendrite/federationsender/api"
go_http_js_libp2p "github.com/matrix-org/go-http-js-libp2p"
"github.com/matrix-org/gomatrixserverlib"
"github.com/matrix-org/util"
)
type libp2pPublicRoomsProvider struct {
node *go_http_js_libp2p.P2pLocalNode
providers []go_http_js_libp2p.PeerInfo
fedSender api.FederationSenderInternalAPI
fedClient *gomatrixserverlib.FederationClient
}
func NewLibP2PPublicRoomsProvider(node *go_http_js_libp2p.P2pLocalNode, fedSender api.FederationSenderInternalAPI) *libp2pPublicRoomsProvider {
func NewLibP2PPublicRoomsProvider(
node *go_http_js_libp2p.P2pLocalNode, fedSender api.FederationSenderInternalAPI, fedClient *gomatrixserverlib.FederationClient,
) *libp2pPublicRoomsProvider {
p := &libp2pPublicRoomsProvider{
node: node,
fedSender: fedSender,
fedClient: fedClient,
}
node.RegisterFoundProviders(p.foundProviders)
return p
@ -63,13 +70,85 @@ func (p *libp2pPublicRoomsProvider) foundProviders(peerInfos []go_http_js_libp2p
}
func (p *libp2pPublicRoomsProvider) Rooms() []gomatrixserverlib.PublicRoom {
return nil
return bulkFetchPublicRoomsFromServers(context.Background(), p.fedClient, p.homeservers())
}
func (p *libp2pPublicRoomsProvider) Homeservers() []string {
func (p *libp2pPublicRoomsProvider) homeservers() []string {
result := make([]string, len(p.providers))
for i := range p.providers {
result[i] = p.providers[i].Id
}
return result
}
// bulkFetchPublicRoomsFromServers fetches public rooms from the list of homeservers.
// Returns a list of public rooms.
func bulkFetchPublicRoomsFromServers(
ctx context.Context, fedClient *gomatrixserverlib.FederationClient, homeservers []string,
) (publicRooms []gomatrixserverlib.PublicRoom) {
limit := 200
// follow pipeline semantics, see https://blog.golang.org/pipelines for more info.
// goroutines send rooms to this channel
roomCh := make(chan gomatrixserverlib.PublicRoom, int(limit))
// signalling channel to tell goroutines to stop sending rooms and quit
done := make(chan bool)
// signalling to say when we can close the room channel
var wg sync.WaitGroup
wg.Add(len(homeservers))
// concurrently query for public rooms
for _, hs := range homeservers {
go func(homeserverDomain string) {
defer wg.Done()
util.GetLogger(ctx).WithField("hs", homeserverDomain).Info("Querying HS for public rooms")
fres, err := fedClient.GetPublicRooms(ctx, gomatrixserverlib.ServerName(homeserverDomain), int(limit), "", false, "")
if err != nil {
util.GetLogger(ctx).WithError(err).WithField("hs", homeserverDomain).Warn(
"bulkFetchPublicRoomsFromServers: failed to query hs",
)
return
}
for _, room := range fres.Chunk {
// atomically send a room or stop
select {
case roomCh <- room:
case <-done:
util.GetLogger(ctx).WithError(err).WithField("hs", homeserverDomain).Info("Interrupted whilst sending rooms")
return
}
}
}(hs)
}
// Close the room channel when the goroutines have quit so we don't leak, but don't let it stop the in-flight request.
// This also allows the request to fail fast if all HSes experience errors as it will cause the room channel to be
// closed.
go func() {
wg.Wait()
util.GetLogger(ctx).Info("Cleaning up resources")
close(roomCh)
}()
// fan-in results with timeout. We stop when we reach the limit.
FanIn:
for len(publicRooms) < int(limit) || limit == 0 {
// add a room or timeout
select {
case room, ok := <-roomCh:
if !ok {
util.GetLogger(ctx).Info("All homeservers have been queried, returning results.")
break FanIn
}
publicRooms = append(publicRooms, room)
case <-time.After(15 * time.Second): // we've waited long enough, let's tell the client what we got.
util.GetLogger(ctx).Info("Waited 15s for federated public rooms, returning early")
break FanIn
case <-ctx.Done(): // the client hung up on us, let's stop.
util.GetLogger(ctx).Info("Client hung up, returning early")
break FanIn
}
}
// tell goroutines to stop
close(done)
return publicRooms
}