sliding-sync/sync3/dispatcher.go

300 lines
9.0 KiB
Go
Raw Permalink Normal View History

package sync3
import (
2023-01-18 11:41:20 +00:00
"context"
"encoding/json"
"os"
"sync"
"github.com/matrix-org/sliding-sync/internal"
"github.com/matrix-org/sliding-sync/sync3/caches"
"github.com/rs/zerolog"
"github.com/tidwall/gjson"
)
var logger = zerolog.New(os.Stdout).With().Timestamp().Logger().Output(zerolog.ConsoleWriter{
Out: os.Stderr,
TimeFormat: "15:04:05",
})
const DispatcherAllUsers = "-"
// Receiver represents the callbacks that a Dispatcher may fire.
type Receiver interface {
2023-01-18 11:41:20 +00:00
OnNewEvent(ctx context.Context, event *caches.EventData)
OnReceipt(ctx context.Context, receipt internal.Receipt)
2023-01-18 11:41:20 +00:00
OnEphemeralEvent(ctx context.Context, roomID string, ephEvent json.RawMessage)
// OnRegistered is called after a successful call to Dispatcher.Register
OnRegistered(ctx context.Context) error
}
2022-03-23 14:13:59 +00:00
// Dispatches live events to caches
type Dispatcher struct {
jrt *JoinedRoomsTracker
userToReceiver map[string]Receiver
userToReceiverMu *sync.RWMutex
}
func NewDispatcher() *Dispatcher {
return &Dispatcher{
jrt: NewJoinedRoomsTracker(),
userToReceiver: make(map[string]Receiver),
userToReceiverMu: &sync.RWMutex{},
}
}
func (d *Dispatcher) IsUserJoined(userID, roomID string) bool {
return d.jrt.IsUserJoined(userID, roomID)
}
// Load joined members into the dispatcher.
// MUST BE CALLED BEFORE V2 POLL LOOPS START.
func (d *Dispatcher) Startup(roomToJoinedUsers map[string][]string) error {
// populate joined rooms tracker
add extensions for typing and receipts; bugfixes and additional perf improvements Features: - Add `typing` extension. - Add `receipts` extension. - Add comprehensive prometheus `/metrics` activated via `SYNCV3_PROM`. - Add `SYNCV3_PPROF` support. - Add `by_notification_level` sort order. - Add `include_old_rooms` support. - Add support for `$ME` and `$LAZY`. - Add correct filtering when `*,*` is used as `required_state`. - Add `num_live` to each room response to indicate how many timeline entries are live. Bug fixes: - Use a stricter comparison function on ranges: fixes an issue whereby UTs fail on go1.19 due to change in sorting algorithm. - Send back an `errcode` on HTTP errors (e.g expired sessions). - Remove `unsigned.txn_id` on insertion into the DB. Otherwise other users would see other users txn IDs :( - Improve range delta algorithm: previously it didn't handle cases like `[0,20] -> [20,30]` and would panic. - Send HTTP 400 for invalid range requests. - Don't publish no-op unread counts which just adds extra noise. - Fix leaking DB connections which could eventually consume all available connections. - Ensure we always unblock WaitUntilInitialSync even on invalid access tokens. Other code relies on WaitUntilInitialSync() actually returning at _some_ point e.g on startup we have N workers which bound the number of concurrent pollers made at any one time, we need to not just hog a worker forever. Improvements: - Greatly improve startup times of sync3 handlers by improving `JoinedRoomsTracker`: a modest amount of data would take ~28s to create the handler, now it takes 4s. - Massively improve initial initial v3 sync times, by refactoring `JoinedRoomsTracker`, from ~47s to <1s. - Add `SlidingSyncUntil...` in tests to reduce races. - Tweak the API shape of JoinedUsersForRoom to reduce state block processing time for large rooms from 63s to 39s. - Add trace task for initial syncs. - Include the proxy version in UA strings. - HTTP errors now wait 1s before returning to stop clients tight-looping on error. - Pending event buffer is now 2000. - Index the room ID first to cull the most events when returning timeline entries. Speeds up `SelectLatestEventsBetween` by a factor of 8. - Remove cancelled `m.room_key_requests` from the to-device inbox. Cuts down the amount of events in the inbox by ~94% for very large (20k+) inboxes, ~50% for moderate sized (200 events) inboxes. Adds book-keeping to remember the unacked to-device position for each client.
2022-12-14 18:53:55 +00:00
d.jrt.Startup(roomToJoinedUsers)
return nil
}
2023-11-07 18:43:20 +00:00
func (d *Dispatcher) Unregister(userID string) {
d.userToReceiverMu.Lock()
defer d.userToReceiverMu.Unlock()
delete(d.userToReceiver, userID)
}
// UnregisterBulk accepts a slice of user IDs to unregister. The given users need not
// already be registered (in which case unregistering them is a no-op). Returns the
// list of users that were unregistered.
func (d *Dispatcher) UnregisterBulk(userIDs []string) []string {
d.userToReceiverMu.Lock()
defer d.userToReceiverMu.Unlock()
2023-11-07 18:43:20 +00:00
unregistered := make([]string, 0)
for _, userID := range userIDs {
_, exists := d.userToReceiver[userID]
if exists {
delete(d.userToReceiver, userID)
unregistered = append(unregistered, userID)
}
}
return unregistered
}
2023-04-05 15:36:05 +01:00
func (d *Dispatcher) Register(ctx context.Context, userID string, r Receiver) error {
d.userToReceiverMu.Lock()
defer d.userToReceiverMu.Unlock()
if _, ok := d.userToReceiver[userID]; ok {
logger.Warn().Str("user", userID).Msg("Dispatcher.Register: receiver already registered")
}
d.userToReceiver[userID] = r
return r.OnRegistered(ctx)
}
func (d *Dispatcher) ReceiverForUser(userID string) Receiver {
d.userToReceiverMu.RLock()
defer d.userToReceiverMu.RUnlock()
return d.userToReceiver[userID]
}
add extensions for typing and receipts; bugfixes and additional perf improvements Features: - Add `typing` extension. - Add `receipts` extension. - Add comprehensive prometheus `/metrics` activated via `SYNCV3_PROM`. - Add `SYNCV3_PPROF` support. - Add `by_notification_level` sort order. - Add `include_old_rooms` support. - Add support for `$ME` and `$LAZY`. - Add correct filtering when `*,*` is used as `required_state`. - Add `num_live` to each room response to indicate how many timeline entries are live. Bug fixes: - Use a stricter comparison function on ranges: fixes an issue whereby UTs fail on go1.19 due to change in sorting algorithm. - Send back an `errcode` on HTTP errors (e.g expired sessions). - Remove `unsigned.txn_id` on insertion into the DB. Otherwise other users would see other users txn IDs :( - Improve range delta algorithm: previously it didn't handle cases like `[0,20] -> [20,30]` and would panic. - Send HTTP 400 for invalid range requests. - Don't publish no-op unread counts which just adds extra noise. - Fix leaking DB connections which could eventually consume all available connections. - Ensure we always unblock WaitUntilInitialSync even on invalid access tokens. Other code relies on WaitUntilInitialSync() actually returning at _some_ point e.g on startup we have N workers which bound the number of concurrent pollers made at any one time, we need to not just hog a worker forever. Improvements: - Greatly improve startup times of sync3 handlers by improving `JoinedRoomsTracker`: a modest amount of data would take ~28s to create the handler, now it takes 4s. - Massively improve initial initial v3 sync times, by refactoring `JoinedRoomsTracker`, from ~47s to <1s. - Add `SlidingSyncUntil...` in tests to reduce races. - Tweak the API shape of JoinedUsersForRoom to reduce state block processing time for large rooms from 63s to 39s. - Add trace task for initial syncs. - Include the proxy version in UA strings. - HTTP errors now wait 1s before returning to stop clients tight-looping on error. - Pending event buffer is now 2000. - Index the room ID first to cull the most events when returning timeline entries. Speeds up `SelectLatestEventsBetween` by a factor of 8. - Remove cancelled `m.room_key_requests` from the to-device inbox. Cuts down the amount of events in the inbox by ~94% for very large (20k+) inboxes, ~50% for moderate sized (200 events) inboxes. Adds book-keeping to remember the unacked to-device position for each client.
2022-12-14 18:53:55 +00:00
func (d *Dispatcher) newEventData(event json.RawMessage, roomID string, latestPos int64) *caches.EventData {
// parse the event to pull out fields we care about
var stateKey *string
ev := gjson.ParseBytes(event)
if sk := ev.Get("state_key"); sk.Exists() {
stateKey = &sk.Str
}
eventType := ev.Get("type").Str
return &caches.EventData{
Event: event,
RoomID: roomID,
EventType: eventType,
StateKey: stateKey,
Content: ev.Get("content"),
NID: latestPos,
Timestamp: ev.Get("origin_server_ts").Uint(),
Sender: ev.Get("sender").Str,
TransactionID: ev.Get("unsigned.transaction_id").Str,
add extensions for typing and receipts; bugfixes and additional perf improvements Features: - Add `typing` extension. - Add `receipts` extension. - Add comprehensive prometheus `/metrics` activated via `SYNCV3_PROM`. - Add `SYNCV3_PPROF` support. - Add `by_notification_level` sort order. - Add `include_old_rooms` support. - Add support for `$ME` and `$LAZY`. - Add correct filtering when `*,*` is used as `required_state`. - Add `num_live` to each room response to indicate how many timeline entries are live. Bug fixes: - Use a stricter comparison function on ranges: fixes an issue whereby UTs fail on go1.19 due to change in sorting algorithm. - Send back an `errcode` on HTTP errors (e.g expired sessions). - Remove `unsigned.txn_id` on insertion into the DB. Otherwise other users would see other users txn IDs :( - Improve range delta algorithm: previously it didn't handle cases like `[0,20] -> [20,30]` and would panic. - Send HTTP 400 for invalid range requests. - Don't publish no-op unread counts which just adds extra noise. - Fix leaking DB connections which could eventually consume all available connections. - Ensure we always unblock WaitUntilInitialSync even on invalid access tokens. Other code relies on WaitUntilInitialSync() actually returning at _some_ point e.g on startup we have N workers which bound the number of concurrent pollers made at any one time, we need to not just hog a worker forever. Improvements: - Greatly improve startup times of sync3 handlers by improving `JoinedRoomsTracker`: a modest amount of data would take ~28s to create the handler, now it takes 4s. - Massively improve initial initial v3 sync times, by refactoring `JoinedRoomsTracker`, from ~47s to <1s. - Add `SlidingSyncUntil...` in tests to reduce races. - Tweak the API shape of JoinedUsersForRoom to reduce state block processing time for large rooms from 63s to 39s. - Add trace task for initial syncs. - Include the proxy version in UA strings. - HTTP errors now wait 1s before returning to stop clients tight-looping on error. - Pending event buffer is now 2000. - Index the room ID first to cull the most events when returning timeline entries. Speeds up `SelectLatestEventsBetween` by a factor of 8. - Remove cancelled `m.room_key_requests` from the to-device inbox. Cuts down the amount of events in the inbox by ~94% for very large (20k+) inboxes, ~50% for moderate sized (200 events) inboxes. Adds book-keeping to remember the unacked to-device position for each client.
2022-12-14 18:53:55 +00:00
}
}
// Called by v2 pollers when we receive an initial state block. Very similar to OnNewEvents but
// done in bulk for speed.
2023-01-18 11:41:20 +00:00
func (d *Dispatcher) OnNewInitialRoomState(ctx context.Context, roomID string, state []json.RawMessage) {
add extensions for typing and receipts; bugfixes and additional perf improvements Features: - Add `typing` extension. - Add `receipts` extension. - Add comprehensive prometheus `/metrics` activated via `SYNCV3_PROM`. - Add `SYNCV3_PPROF` support. - Add `by_notification_level` sort order. - Add `include_old_rooms` support. - Add support for `$ME` and `$LAZY`. - Add correct filtering when `*,*` is used as `required_state`. - Add `num_live` to each room response to indicate how many timeline entries are live. Bug fixes: - Use a stricter comparison function on ranges: fixes an issue whereby UTs fail on go1.19 due to change in sorting algorithm. - Send back an `errcode` on HTTP errors (e.g expired sessions). - Remove `unsigned.txn_id` on insertion into the DB. Otherwise other users would see other users txn IDs :( - Improve range delta algorithm: previously it didn't handle cases like `[0,20] -> [20,30]` and would panic. - Send HTTP 400 for invalid range requests. - Don't publish no-op unread counts which just adds extra noise. - Fix leaking DB connections which could eventually consume all available connections. - Ensure we always unblock WaitUntilInitialSync even on invalid access tokens. Other code relies on WaitUntilInitialSync() actually returning at _some_ point e.g on startup we have N workers which bound the number of concurrent pollers made at any one time, we need to not just hog a worker forever. Improvements: - Greatly improve startup times of sync3 handlers by improving `JoinedRoomsTracker`: a modest amount of data would take ~28s to create the handler, now it takes 4s. - Massively improve initial initial v3 sync times, by refactoring `JoinedRoomsTracker`, from ~47s to <1s. - Add `SlidingSyncUntil...` in tests to reduce races. - Tweak the API shape of JoinedUsersForRoom to reduce state block processing time for large rooms from 63s to 39s. - Add trace task for initial syncs. - Include the proxy version in UA strings. - HTTP errors now wait 1s before returning to stop clients tight-looping on error. - Pending event buffer is now 2000. - Index the room ID first to cull the most events when returning timeline entries. Speeds up `SelectLatestEventsBetween` by a factor of 8. - Remove cancelled `m.room_key_requests` from the to-device inbox. Cuts down the amount of events in the inbox by ~94% for very large (20k+) inboxes, ~50% for moderate sized (200 events) inboxes. Adds book-keeping to remember the unacked to-device position for each client.
2022-12-14 18:53:55 +00:00
// sanity check
if _, jc := d.jrt.JoinedUsersForRoom(roomID, nil); jc > 0 {
logger.Warn().Int("join_count", jc).Str("room", roomID).Int("num_state", len(state)).Msg(
"OnNewInitialRoomState but have entries in JoinedRoomsTracker already, this should be impossible. Degrading to live events",
)
for _, s := range state {
d.OnNewEvent(ctx, roomID, s, 0)
}
add extensions for typing and receipts; bugfixes and additional perf improvements Features: - Add `typing` extension. - Add `receipts` extension. - Add comprehensive prometheus `/metrics` activated via `SYNCV3_PROM`. - Add `SYNCV3_PPROF` support. - Add `by_notification_level` sort order. - Add `include_old_rooms` support. - Add support for `$ME` and `$LAZY`. - Add correct filtering when `*,*` is used as `required_state`. - Add `num_live` to each room response to indicate how many timeline entries are live. Bug fixes: - Use a stricter comparison function on ranges: fixes an issue whereby UTs fail on go1.19 due to change in sorting algorithm. - Send back an `errcode` on HTTP errors (e.g expired sessions). - Remove `unsigned.txn_id` on insertion into the DB. Otherwise other users would see other users txn IDs :( - Improve range delta algorithm: previously it didn't handle cases like `[0,20] -> [20,30]` and would panic. - Send HTTP 400 for invalid range requests. - Don't publish no-op unread counts which just adds extra noise. - Fix leaking DB connections which could eventually consume all available connections. - Ensure we always unblock WaitUntilInitialSync even on invalid access tokens. Other code relies on WaitUntilInitialSync() actually returning at _some_ point e.g on startup we have N workers which bound the number of concurrent pollers made at any one time, we need to not just hog a worker forever. Improvements: - Greatly improve startup times of sync3 handlers by improving `JoinedRoomsTracker`: a modest amount of data would take ~28s to create the handler, now it takes 4s. - Massively improve initial initial v3 sync times, by refactoring `JoinedRoomsTracker`, from ~47s to <1s. - Add `SlidingSyncUntil...` in tests to reduce races. - Tweak the API shape of JoinedUsersForRoom to reduce state block processing time for large rooms from 63s to 39s. - Add trace task for initial syncs. - Include the proxy version in UA strings. - HTTP errors now wait 1s before returning to stop clients tight-looping on error. - Pending event buffer is now 2000. - Index the room ID first to cull the most events when returning timeline entries. Speeds up `SelectLatestEventsBetween` by a factor of 8. - Remove cancelled `m.room_key_requests` from the to-device inbox. Cuts down the amount of events in the inbox by ~94% for very large (20k+) inboxes, ~50% for moderate sized (200 events) inboxes. Adds book-keeping to remember the unacked to-device position for each client.
2022-12-14 18:53:55 +00:00
return
}
// create event datas for state
eventDatas := make([]*caches.EventData, len(state))
var joined, invited []string
for i, event := range state {
ed := d.newEventData(event, roomID, 0)
eventDatas[i] = ed
if ed.EventType == "m.room.member" && ed.StateKey != nil {
membership := ed.Content.Get("membership").Str
switch membership {
case "invite":
invited = append(invited, *ed.StateKey)
case "join":
joined = append(joined, *ed.StateKey)
}
}
}
// bulk update joined room tracker
forceInitial := d.jrt.UsersJoinedRoom(joined, roomID)
d.jrt.UsersInvitedToRoom(invited, roomID)
inviteCount := d.jrt.NumInvitedUsersForRoom(roomID)
// work out who to notify
userIDs, joinCount := d.jrt.JoinedUsersForRoom(roomID, func(userID string) bool {
if userID == DispatcherAllUsers {
return false // safety guard to prevent dupe global callbacks
}
return d.ReceiverForUser(userID) != nil
add extensions for typing and receipts; bugfixes and additional perf improvements Features: - Add `typing` extension. - Add `receipts` extension. - Add comprehensive prometheus `/metrics` activated via `SYNCV3_PROM`. - Add `SYNCV3_PPROF` support. - Add `by_notification_level` sort order. - Add `include_old_rooms` support. - Add support for `$ME` and `$LAZY`. - Add correct filtering when `*,*` is used as `required_state`. - Add `num_live` to each room response to indicate how many timeline entries are live. Bug fixes: - Use a stricter comparison function on ranges: fixes an issue whereby UTs fail on go1.19 due to change in sorting algorithm. - Send back an `errcode` on HTTP errors (e.g expired sessions). - Remove `unsigned.txn_id` on insertion into the DB. Otherwise other users would see other users txn IDs :( - Improve range delta algorithm: previously it didn't handle cases like `[0,20] -> [20,30]` and would panic. - Send HTTP 400 for invalid range requests. - Don't publish no-op unread counts which just adds extra noise. - Fix leaking DB connections which could eventually consume all available connections. - Ensure we always unblock WaitUntilInitialSync even on invalid access tokens. Other code relies on WaitUntilInitialSync() actually returning at _some_ point e.g on startup we have N workers which bound the number of concurrent pollers made at any one time, we need to not just hog a worker forever. Improvements: - Greatly improve startup times of sync3 handlers by improving `JoinedRoomsTracker`: a modest amount of data would take ~28s to create the handler, now it takes 4s. - Massively improve initial initial v3 sync times, by refactoring `JoinedRoomsTracker`, from ~47s to <1s. - Add `SlidingSyncUntil...` in tests to reduce races. - Tweak the API shape of JoinedUsersForRoom to reduce state block processing time for large rooms from 63s to 39s. - Add trace task for initial syncs. - Include the proxy version in UA strings. - HTTP errors now wait 1s before returning to stop clients tight-looping on error. - Pending event buffer is now 2000. - Index the room ID first to cull the most events when returning timeline entries. Speeds up `SelectLatestEventsBetween` by a factor of 8. - Remove cancelled `m.room_key_requests` from the to-device inbox. Cuts down the amount of events in the inbox by ~94% for very large (20k+) inboxes, ~50% for moderate sized (200 events) inboxes. Adds book-keeping to remember the unacked to-device position for each client.
2022-12-14 18:53:55 +00:00
})
// notify listeners
for _, ed := range eventDatas {
ed.InviteCount = inviteCount
ed.JoinCount = joinCount
2023-01-18 11:41:20 +00:00
d.notifyListeners(ctx, ed, userIDs, "", forceInitial, "")
add extensions for typing and receipts; bugfixes and additional perf improvements Features: - Add `typing` extension. - Add `receipts` extension. - Add comprehensive prometheus `/metrics` activated via `SYNCV3_PROM`. - Add `SYNCV3_PPROF` support. - Add `by_notification_level` sort order. - Add `include_old_rooms` support. - Add support for `$ME` and `$LAZY`. - Add correct filtering when `*,*` is used as `required_state`. - Add `num_live` to each room response to indicate how many timeline entries are live. Bug fixes: - Use a stricter comparison function on ranges: fixes an issue whereby UTs fail on go1.19 due to change in sorting algorithm. - Send back an `errcode` on HTTP errors (e.g expired sessions). - Remove `unsigned.txn_id` on insertion into the DB. Otherwise other users would see other users txn IDs :( - Improve range delta algorithm: previously it didn't handle cases like `[0,20] -> [20,30]` and would panic. - Send HTTP 400 for invalid range requests. - Don't publish no-op unread counts which just adds extra noise. - Fix leaking DB connections which could eventually consume all available connections. - Ensure we always unblock WaitUntilInitialSync even on invalid access tokens. Other code relies on WaitUntilInitialSync() actually returning at _some_ point e.g on startup we have N workers which bound the number of concurrent pollers made at any one time, we need to not just hog a worker forever. Improvements: - Greatly improve startup times of sync3 handlers by improving `JoinedRoomsTracker`: a modest amount of data would take ~28s to create the handler, now it takes 4s. - Massively improve initial initial v3 sync times, by refactoring `JoinedRoomsTracker`, from ~47s to <1s. - Add `SlidingSyncUntil...` in tests to reduce races. - Tweak the API shape of JoinedUsersForRoom to reduce state block processing time for large rooms from 63s to 39s. - Add trace task for initial syncs. - Include the proxy version in UA strings. - HTTP errors now wait 1s before returning to stop clients tight-looping on error. - Pending event buffer is now 2000. - Index the room ID first to cull the most events when returning timeline entries. Speeds up `SelectLatestEventsBetween` by a factor of 8. - Remove cancelled `m.room_key_requests` from the to-device inbox. Cuts down the amount of events in the inbox by ~94% for very large (20k+) inboxes, ~50% for moderate sized (200 events) inboxes. Adds book-keeping to remember the unacked to-device position for each client.
2022-12-14 18:53:55 +00:00
}
}
func (d *Dispatcher) OnNewEvent(
ctx context.Context, roomID string, event json.RawMessage, nid int64,
) {
ed := d.newEventData(event, roomID, nid)
// update the tracker
targetUser := ""
membership := ""
shouldForceInitial := false
leaveAfterJoinOrInvite := false
if ed.EventType == "m.room.member" && ed.StateKey != nil {
targetUser = *ed.StateKey
membership = ed.Content.Get("membership").Str
switch membership {
case "invite":
// we only do this to track invite counts correctly.
add extensions for typing and receipts; bugfixes and additional perf improvements Features: - Add `typing` extension. - Add `receipts` extension. - Add comprehensive prometheus `/metrics` activated via `SYNCV3_PROM`. - Add `SYNCV3_PPROF` support. - Add `by_notification_level` sort order. - Add `include_old_rooms` support. - Add support for `$ME` and `$LAZY`. - Add correct filtering when `*,*` is used as `required_state`. - Add `num_live` to each room response to indicate how many timeline entries are live. Bug fixes: - Use a stricter comparison function on ranges: fixes an issue whereby UTs fail on go1.19 due to change in sorting algorithm. - Send back an `errcode` on HTTP errors (e.g expired sessions). - Remove `unsigned.txn_id` on insertion into the DB. Otherwise other users would see other users txn IDs :( - Improve range delta algorithm: previously it didn't handle cases like `[0,20] -> [20,30]` and would panic. - Send HTTP 400 for invalid range requests. - Don't publish no-op unread counts which just adds extra noise. - Fix leaking DB connections which could eventually consume all available connections. - Ensure we always unblock WaitUntilInitialSync even on invalid access tokens. Other code relies on WaitUntilInitialSync() actually returning at _some_ point e.g on startup we have N workers which bound the number of concurrent pollers made at any one time, we need to not just hog a worker forever. Improvements: - Greatly improve startup times of sync3 handlers by improving `JoinedRoomsTracker`: a modest amount of data would take ~28s to create the handler, now it takes 4s. - Massively improve initial initial v3 sync times, by refactoring `JoinedRoomsTracker`, from ~47s to <1s. - Add `SlidingSyncUntil...` in tests to reduce races. - Tweak the API shape of JoinedUsersForRoom to reduce state block processing time for large rooms from 63s to 39s. - Add trace task for initial syncs. - Include the proxy version in UA strings. - HTTP errors now wait 1s before returning to stop clients tight-looping on error. - Pending event buffer is now 2000. - Index the room ID first to cull the most events when returning timeline entries. Speeds up `SelectLatestEventsBetween` by a factor of 8. - Remove cancelled `m.room_key_requests` from the to-device inbox. Cuts down the amount of events in the inbox by ~94% for very large (20k+) inboxes, ~50% for moderate sized (200 events) inboxes. Adds book-keeping to remember the unacked to-device position for each client.
2022-12-14 18:53:55 +00:00
d.jrt.UsersInvitedToRoom([]string{targetUser}, ed.RoomID)
case "join":
if d.jrt.UserJoinedRoom(targetUser, ed.RoomID) {
shouldForceInitial = true
}
case "ban":
fallthrough
case "leave":
leaveAfterJoinOrInvite = d.jrt.UserLeftRoom(targetUser, ed.RoomID)
}
ed.InviteCount = d.jrt.NumInvitedUsersForRoom(ed.RoomID)
}
// notify all people in this room
add extensions for typing and receipts; bugfixes and additional perf improvements Features: - Add `typing` extension. - Add `receipts` extension. - Add comprehensive prometheus `/metrics` activated via `SYNCV3_PROM`. - Add `SYNCV3_PPROF` support. - Add `by_notification_level` sort order. - Add `include_old_rooms` support. - Add support for `$ME` and `$LAZY`. - Add correct filtering when `*,*` is used as `required_state`. - Add `num_live` to each room response to indicate how many timeline entries are live. Bug fixes: - Use a stricter comparison function on ranges: fixes an issue whereby UTs fail on go1.19 due to change in sorting algorithm. - Send back an `errcode` on HTTP errors (e.g expired sessions). - Remove `unsigned.txn_id` on insertion into the DB. Otherwise other users would see other users txn IDs :( - Improve range delta algorithm: previously it didn't handle cases like `[0,20] -> [20,30]` and would panic. - Send HTTP 400 for invalid range requests. - Don't publish no-op unread counts which just adds extra noise. - Fix leaking DB connections which could eventually consume all available connections. - Ensure we always unblock WaitUntilInitialSync even on invalid access tokens. Other code relies on WaitUntilInitialSync() actually returning at _some_ point e.g on startup we have N workers which bound the number of concurrent pollers made at any one time, we need to not just hog a worker forever. Improvements: - Greatly improve startup times of sync3 handlers by improving `JoinedRoomsTracker`: a modest amount of data would take ~28s to create the handler, now it takes 4s. - Massively improve initial initial v3 sync times, by refactoring `JoinedRoomsTracker`, from ~47s to <1s. - Add `SlidingSyncUntil...` in tests to reduce races. - Tweak the API shape of JoinedUsersForRoom to reduce state block processing time for large rooms from 63s to 39s. - Add trace task for initial syncs. - Include the proxy version in UA strings. - HTTP errors now wait 1s before returning to stop clients tight-looping on error. - Pending event buffer is now 2000. - Index the room ID first to cull the most events when returning timeline entries. Speeds up `SelectLatestEventsBetween` by a factor of 8. - Remove cancelled `m.room_key_requests` from the to-device inbox. Cuts down the amount of events in the inbox by ~94% for very large (20k+) inboxes, ~50% for moderate sized (200 events) inboxes. Adds book-keeping to remember the unacked to-device position for each client.
2022-12-14 18:53:55 +00:00
userIDs, joinCount := d.jrt.JoinedUsersForRoom(ed.RoomID, func(userID string) bool {
if userID == DispatcherAllUsers {
return false // safety guard to prevent dupe global callbacks
}
return d.ReceiverForUser(userID) != nil
add extensions for typing and receipts; bugfixes and additional perf improvements Features: - Add `typing` extension. - Add `receipts` extension. - Add comprehensive prometheus `/metrics` activated via `SYNCV3_PROM`. - Add `SYNCV3_PPROF` support. - Add `by_notification_level` sort order. - Add `include_old_rooms` support. - Add support for `$ME` and `$LAZY`. - Add correct filtering when `*,*` is used as `required_state`. - Add `num_live` to each room response to indicate how many timeline entries are live. Bug fixes: - Use a stricter comparison function on ranges: fixes an issue whereby UTs fail on go1.19 due to change in sorting algorithm. - Send back an `errcode` on HTTP errors (e.g expired sessions). - Remove `unsigned.txn_id` on insertion into the DB. Otherwise other users would see other users txn IDs :( - Improve range delta algorithm: previously it didn't handle cases like `[0,20] -> [20,30]` and would panic. - Send HTTP 400 for invalid range requests. - Don't publish no-op unread counts which just adds extra noise. - Fix leaking DB connections which could eventually consume all available connections. - Ensure we always unblock WaitUntilInitialSync even on invalid access tokens. Other code relies on WaitUntilInitialSync() actually returning at _some_ point e.g on startup we have N workers which bound the number of concurrent pollers made at any one time, we need to not just hog a worker forever. Improvements: - Greatly improve startup times of sync3 handlers by improving `JoinedRoomsTracker`: a modest amount of data would take ~28s to create the handler, now it takes 4s. - Massively improve initial initial v3 sync times, by refactoring `JoinedRoomsTracker`, from ~47s to <1s. - Add `SlidingSyncUntil...` in tests to reduce races. - Tweak the API shape of JoinedUsersForRoom to reduce state block processing time for large rooms from 63s to 39s. - Add trace task for initial syncs. - Include the proxy version in UA strings. - HTTP errors now wait 1s before returning to stop clients tight-looping on error. - Pending event buffer is now 2000. - Index the room ID first to cull the most events when returning timeline entries. Speeds up `SelectLatestEventsBetween` by a factor of 8. - Remove cancelled `m.room_key_requests` from the to-device inbox. Cuts down the amount of events in the inbox by ~94% for very large (20k+) inboxes, ~50% for moderate sized (200 events) inboxes. Adds book-keeping to remember the unacked to-device position for each client.
2022-12-14 18:53:55 +00:00
})
ed.JoinCount = joinCount
if leaveAfterJoinOrInvite {
// Only tell the target user about a leave if they were previously aware of the
// room. This prevents us from leaking pre-emptive bans.
userIDs = append(userIDs, targetUser)
}
2023-01-18 11:41:20 +00:00
d.notifyListeners(ctx, ed, userIDs, targetUser, shouldForceInitial, membership)
add extensions for typing and receipts; bugfixes and additional perf improvements Features: - Add `typing` extension. - Add `receipts` extension. - Add comprehensive prometheus `/metrics` activated via `SYNCV3_PROM`. - Add `SYNCV3_PPROF` support. - Add `by_notification_level` sort order. - Add `include_old_rooms` support. - Add support for `$ME` and `$LAZY`. - Add correct filtering when `*,*` is used as `required_state`. - Add `num_live` to each room response to indicate how many timeline entries are live. Bug fixes: - Use a stricter comparison function on ranges: fixes an issue whereby UTs fail on go1.19 due to change in sorting algorithm. - Send back an `errcode` on HTTP errors (e.g expired sessions). - Remove `unsigned.txn_id` on insertion into the DB. Otherwise other users would see other users txn IDs :( - Improve range delta algorithm: previously it didn't handle cases like `[0,20] -> [20,30]` and would panic. - Send HTTP 400 for invalid range requests. - Don't publish no-op unread counts which just adds extra noise. - Fix leaking DB connections which could eventually consume all available connections. - Ensure we always unblock WaitUntilInitialSync even on invalid access tokens. Other code relies on WaitUntilInitialSync() actually returning at _some_ point e.g on startup we have N workers which bound the number of concurrent pollers made at any one time, we need to not just hog a worker forever. Improvements: - Greatly improve startup times of sync3 handlers by improving `JoinedRoomsTracker`: a modest amount of data would take ~28s to create the handler, now it takes 4s. - Massively improve initial initial v3 sync times, by refactoring `JoinedRoomsTracker`, from ~47s to <1s. - Add `SlidingSyncUntil...` in tests to reduce races. - Tweak the API shape of JoinedUsersForRoom to reduce state block processing time for large rooms from 63s to 39s. - Add trace task for initial syncs. - Include the proxy version in UA strings. - HTTP errors now wait 1s before returning to stop clients tight-looping on error. - Pending event buffer is now 2000. - Index the room ID first to cull the most events when returning timeline entries. Speeds up `SelectLatestEventsBetween` by a factor of 8. - Remove cancelled `m.room_key_requests` from the to-device inbox. Cuts down the amount of events in the inbox by ~94% for very large (20k+) inboxes, ~50% for moderate sized (200 events) inboxes. Adds book-keeping to remember the unacked to-device position for each client.
2022-12-14 18:53:55 +00:00
}
2023-01-18 11:41:20 +00:00
func (d *Dispatcher) OnEphemeralEvent(ctx context.Context, roomID string, ephEvent json.RawMessage) {
add extensions for typing and receipts; bugfixes and additional perf improvements Features: - Add `typing` extension. - Add `receipts` extension. - Add comprehensive prometheus `/metrics` activated via `SYNCV3_PROM`. - Add `SYNCV3_PPROF` support. - Add `by_notification_level` sort order. - Add `include_old_rooms` support. - Add support for `$ME` and `$LAZY`. - Add correct filtering when `*,*` is used as `required_state`. - Add `num_live` to each room response to indicate how many timeline entries are live. Bug fixes: - Use a stricter comparison function on ranges: fixes an issue whereby UTs fail on go1.19 due to change in sorting algorithm. - Send back an `errcode` on HTTP errors (e.g expired sessions). - Remove `unsigned.txn_id` on insertion into the DB. Otherwise other users would see other users txn IDs :( - Improve range delta algorithm: previously it didn't handle cases like `[0,20] -> [20,30]` and would panic. - Send HTTP 400 for invalid range requests. - Don't publish no-op unread counts which just adds extra noise. - Fix leaking DB connections which could eventually consume all available connections. - Ensure we always unblock WaitUntilInitialSync even on invalid access tokens. Other code relies on WaitUntilInitialSync() actually returning at _some_ point e.g on startup we have N workers which bound the number of concurrent pollers made at any one time, we need to not just hog a worker forever. Improvements: - Greatly improve startup times of sync3 handlers by improving `JoinedRoomsTracker`: a modest amount of data would take ~28s to create the handler, now it takes 4s. - Massively improve initial initial v3 sync times, by refactoring `JoinedRoomsTracker`, from ~47s to <1s. - Add `SlidingSyncUntil...` in tests to reduce races. - Tweak the API shape of JoinedUsersForRoom to reduce state block processing time for large rooms from 63s to 39s. - Add trace task for initial syncs. - Include the proxy version in UA strings. - HTTP errors now wait 1s before returning to stop clients tight-looping on error. - Pending event buffer is now 2000. - Index the room ID first to cull the most events when returning timeline entries. Speeds up `SelectLatestEventsBetween` by a factor of 8. - Remove cancelled `m.room_key_requests` from the to-device inbox. Cuts down the amount of events in the inbox by ~94% for very large (20k+) inboxes, ~50% for moderate sized (200 events) inboxes. Adds book-keeping to remember the unacked to-device position for each client.
2022-12-14 18:53:55 +00:00
notifyUserIDs, _ := d.jrt.JoinedUsersForRoom(roomID, func(userID string) bool {
if userID == DispatcherAllUsers {
return false // safety guard to prevent dupe global callbacks
}
return d.ReceiverForUser(userID) != nil
add extensions for typing and receipts; bugfixes and additional perf improvements Features: - Add `typing` extension. - Add `receipts` extension. - Add comprehensive prometheus `/metrics` activated via `SYNCV3_PROM`. - Add `SYNCV3_PPROF` support. - Add `by_notification_level` sort order. - Add `include_old_rooms` support. - Add support for `$ME` and `$LAZY`. - Add correct filtering when `*,*` is used as `required_state`. - Add `num_live` to each room response to indicate how many timeline entries are live. Bug fixes: - Use a stricter comparison function on ranges: fixes an issue whereby UTs fail on go1.19 due to change in sorting algorithm. - Send back an `errcode` on HTTP errors (e.g expired sessions). - Remove `unsigned.txn_id` on insertion into the DB. Otherwise other users would see other users txn IDs :( - Improve range delta algorithm: previously it didn't handle cases like `[0,20] -> [20,30]` and would panic. - Send HTTP 400 for invalid range requests. - Don't publish no-op unread counts which just adds extra noise. - Fix leaking DB connections which could eventually consume all available connections. - Ensure we always unblock WaitUntilInitialSync even on invalid access tokens. Other code relies on WaitUntilInitialSync() actually returning at _some_ point e.g on startup we have N workers which bound the number of concurrent pollers made at any one time, we need to not just hog a worker forever. Improvements: - Greatly improve startup times of sync3 handlers by improving `JoinedRoomsTracker`: a modest amount of data would take ~28s to create the handler, now it takes 4s. - Massively improve initial initial v3 sync times, by refactoring `JoinedRoomsTracker`, from ~47s to <1s. - Add `SlidingSyncUntil...` in tests to reduce races. - Tweak the API shape of JoinedUsersForRoom to reduce state block processing time for large rooms from 63s to 39s. - Add trace task for initial syncs. - Include the proxy version in UA strings. - HTTP errors now wait 1s before returning to stop clients tight-looping on error. - Pending event buffer is now 2000. - Index the room ID first to cull the most events when returning timeline entries. Speeds up `SelectLatestEventsBetween` by a factor of 8. - Remove cancelled `m.room_key_requests` from the to-device inbox. Cuts down the amount of events in the inbox by ~94% for very large (20k+) inboxes, ~50% for moderate sized (200 events) inboxes. Adds book-keeping to remember the unacked to-device position for each client.
2022-12-14 18:53:55 +00:00
})
d.userToReceiverMu.RLock()
defer d.userToReceiverMu.RUnlock()
// global listeners (invoke before per-user listeners so caches can update)
listener := d.userToReceiver[DispatcherAllUsers]
if listener != nil {
2023-01-18 11:41:20 +00:00
listener.OnEphemeralEvent(ctx, roomID, ephEvent)
add extensions for typing and receipts; bugfixes and additional perf improvements Features: - Add `typing` extension. - Add `receipts` extension. - Add comprehensive prometheus `/metrics` activated via `SYNCV3_PROM`. - Add `SYNCV3_PPROF` support. - Add `by_notification_level` sort order. - Add `include_old_rooms` support. - Add support for `$ME` and `$LAZY`. - Add correct filtering when `*,*` is used as `required_state`. - Add `num_live` to each room response to indicate how many timeline entries are live. Bug fixes: - Use a stricter comparison function on ranges: fixes an issue whereby UTs fail on go1.19 due to change in sorting algorithm. - Send back an `errcode` on HTTP errors (e.g expired sessions). - Remove `unsigned.txn_id` on insertion into the DB. Otherwise other users would see other users txn IDs :( - Improve range delta algorithm: previously it didn't handle cases like `[0,20] -> [20,30]` and would panic. - Send HTTP 400 for invalid range requests. - Don't publish no-op unread counts which just adds extra noise. - Fix leaking DB connections which could eventually consume all available connections. - Ensure we always unblock WaitUntilInitialSync even on invalid access tokens. Other code relies on WaitUntilInitialSync() actually returning at _some_ point e.g on startup we have N workers which bound the number of concurrent pollers made at any one time, we need to not just hog a worker forever. Improvements: - Greatly improve startup times of sync3 handlers by improving `JoinedRoomsTracker`: a modest amount of data would take ~28s to create the handler, now it takes 4s. - Massively improve initial initial v3 sync times, by refactoring `JoinedRoomsTracker`, from ~47s to <1s. - Add `SlidingSyncUntil...` in tests to reduce races. - Tweak the API shape of JoinedUsersForRoom to reduce state block processing time for large rooms from 63s to 39s. - Add trace task for initial syncs. - Include the proxy version in UA strings. - HTTP errors now wait 1s before returning to stop clients tight-looping on error. - Pending event buffer is now 2000. - Index the room ID first to cull the most events when returning timeline entries. Speeds up `SelectLatestEventsBetween` by a factor of 8. - Remove cancelled `m.room_key_requests` from the to-device inbox. Cuts down the amount of events in the inbox by ~94% for very large (20k+) inboxes, ~50% for moderate sized (200 events) inboxes. Adds book-keeping to remember the unacked to-device position for each client.
2022-12-14 18:53:55 +00:00
}
// poke user caches OnEphemeralEvent which then pokes ConnState
for _, userID := range notifyUserIDs {
l := d.userToReceiver[userID]
if l == nil {
continue
}
2023-01-18 11:41:20 +00:00
l.OnEphemeralEvent(ctx, roomID, ephEvent)
add extensions for typing and receipts; bugfixes and additional perf improvements Features: - Add `typing` extension. - Add `receipts` extension. - Add comprehensive prometheus `/metrics` activated via `SYNCV3_PROM`. - Add `SYNCV3_PPROF` support. - Add `by_notification_level` sort order. - Add `include_old_rooms` support. - Add support for `$ME` and `$LAZY`. - Add correct filtering when `*,*` is used as `required_state`. - Add `num_live` to each room response to indicate how many timeline entries are live. Bug fixes: - Use a stricter comparison function on ranges: fixes an issue whereby UTs fail on go1.19 due to change in sorting algorithm. - Send back an `errcode` on HTTP errors (e.g expired sessions). - Remove `unsigned.txn_id` on insertion into the DB. Otherwise other users would see other users txn IDs :( - Improve range delta algorithm: previously it didn't handle cases like `[0,20] -> [20,30]` and would panic. - Send HTTP 400 for invalid range requests. - Don't publish no-op unread counts which just adds extra noise. - Fix leaking DB connections which could eventually consume all available connections. - Ensure we always unblock WaitUntilInitialSync even on invalid access tokens. Other code relies on WaitUntilInitialSync() actually returning at _some_ point e.g on startup we have N workers which bound the number of concurrent pollers made at any one time, we need to not just hog a worker forever. Improvements: - Greatly improve startup times of sync3 handlers by improving `JoinedRoomsTracker`: a modest amount of data would take ~28s to create the handler, now it takes 4s. - Massively improve initial initial v3 sync times, by refactoring `JoinedRoomsTracker`, from ~47s to <1s. - Add `SlidingSyncUntil...` in tests to reduce races. - Tweak the API shape of JoinedUsersForRoom to reduce state block processing time for large rooms from 63s to 39s. - Add trace task for initial syncs. - Include the proxy version in UA strings. - HTTP errors now wait 1s before returning to stop clients tight-looping on error. - Pending event buffer is now 2000. - Index the room ID first to cull the most events when returning timeline entries. Speeds up `SelectLatestEventsBetween` by a factor of 8. - Remove cancelled `m.room_key_requests` from the to-device inbox. Cuts down the amount of events in the inbox by ~94% for very large (20k+) inboxes, ~50% for moderate sized (200 events) inboxes. Adds book-keeping to remember the unacked to-device position for each client.
2022-12-14 18:53:55 +00:00
}
}
func (d *Dispatcher) OnReceipt(ctx context.Context, receipt internal.Receipt) {
notifyUserIDs, _ := d.jrt.JoinedUsersForRoom(receipt.RoomID, func(userID string) bool {
if userID == DispatcherAllUsers {
return false // safety guard to prevent dupe global callbacks
}
return d.ReceiverForUser(userID) != nil
})
d.userToReceiverMu.RLock()
defer d.userToReceiverMu.RUnlock()
// global listeners (invoke before per-user listeners so caches can update)
listener := d.userToReceiver[DispatcherAllUsers]
if listener != nil {
listener.OnReceipt(ctx, receipt) // FIXME: redundant, it doesn't care about receipts
}
// poke user caches OnReceipt which then pokes ConnState
for _, userID := range notifyUserIDs {
l := d.userToReceiver[userID]
if l == nil {
continue
}
l.OnReceipt(ctx, receipt)
}
}
2023-01-18 11:41:20 +00:00
func (d *Dispatcher) notifyListeners(ctx context.Context, ed *caches.EventData, userIDs []string, targetUser string, shouldForceInitial bool, membership string) {
2023-06-06 11:48:48 +01:00
internal.Logf(ctx, "dispatcher", "%s: notify %d users (nid=%d,join_count=%d)", ed.RoomID, len(userIDs), ed.NID, ed.JoinCount)
// invoke listeners
d.userToReceiverMu.RLock()
defer d.userToReceiverMu.RUnlock()
// global listeners (invoke before per-user listeners so caches can update)
listener := d.userToReceiver[DispatcherAllUsers]
if listener != nil {
2023-01-18 11:41:20 +00:00
listener.OnNewEvent(ctx, ed)
}
// per-user listeners
for _, userID := range userIDs {
l := d.userToReceiver[userID]
if l != nil {
edd := *ed
if targetUser == userID {
if shouldForceInitial {
edd.ForceInitial = true
}
}
2023-01-18 11:41:20 +00:00
l.OnNewEvent(ctx, &edd)
}
}
}
func (d *Dispatcher) OnInvalidateRoom(roomID string, joins, invites []string) {
// Reset the joined room tracker.
d.jrt.ReloadMembershipsForRoom(roomID, joins, invites)
}