| 1 | use std::collections::{HashMap, VecDeque}; |
| 2 | |
| 3 | use matrix_sdk::{Room, RoomState}; |
| 4 | use matrix_sdk_ui::timeline::{RoomExt, Timeline, TimelineFocus, TimelineReadReceiptTracking}; |
| 5 | use ruma::OwnedRoomId; |
| 6 | use tokio::sync::mpsc; |
| 7 | |
| 8 | use super::executor::{AbortHandle, JoinError, TaskSet, is_panic, resume_unwind}; |
| 9 | use super::repair::{InstallRequest, RepairController}; |
| 10 | use super::timeline::{SessionContext, TaskGuard, TimelineSession}; |
| 11 | use crate::{ |
| 12 | Error, ErrorKind, Event, HistoryRepairCommand, TimelineCommand, TimelineId, TypingState, |
| 13 | transport::Responder, |
| 14 | }; |
| 15 | |
| 16 | pub(super) enum ActorRequest { |
| 17 | /// opening an open room answers with a fresh reset |
| 18 | Open(Responder), |
| 19 | Execute(TimelineCommand, Responder), |
| 20 | Repair(HistoryRepairCommand, Responder), |
| 21 | Close(Responder), |
| 22 | } |
| 23 | |
| 24 | impl ActorRequest { |
| 25 | pub(super) fn into_reply(self) -> Responder { |
| 26 | match self { |
| 27 | Self::Open(reply) |
| 28 | | Self::Execute(_, reply) |
| 29 | | Self::Repair(_, reply) |
| 30 | | Self::Close(reply) => reply, |
| 31 | } |
| 32 | } |
| 33 | } |
| 34 | |
| 35 | pub(super) enum ActorNote { |
| 36 | /// stopping on its own, after a leave or a kick. anything sent its way now comes back in `Exited` |
| 37 | Leaving { |
| 38 | room_id: OwnedRoomId, |
| 39 | generation: u64, |
| 40 | }, |
| 41 | Exited { |
| 42 | room_id: OwnedRoomId, |
| 43 | generation: u64, |
| 44 | /// requests that arrived too late to be served, for the engine to answer or replay |
| 45 | leftovers: Vec<ActorRequest>, |
| 46 | }, |
| 47 | } |
| 48 | |
| 49 | pub(super) struct RoomActor { |
| 50 | pub(super) room_id: OwnedRoomId, |
| 51 | pub(super) generation: u64, |
| 52 | pub(super) room: Room, |
| 53 | pub(super) requests: mpsc::Receiver<ActorRequest>, |
| 54 | pub(super) notes: mpsc::UnboundedSender<ActorNote>, |
| 55 | pub(super) context: SessionContext, |
| 56 | pub(super) panics: mpsc::UnboundedReceiver<&'static str>, |
| 57 | pub(super) opening: Vec<Responder>, |
| 58 | } |
| 59 | |
| 60 | impl RoomActor { |
| 61 | pub(super) async fn run(mut self) { |
| 62 | self.serve().await; |
| 63 | self.requests.close(); |
| 64 | let mut leftovers = Vec::new(); |
| 65 | while let Ok(request) = self.requests.try_recv() { |
| 66 | leftovers.push(request); |
| 67 | } |
| 68 | let _ = self.notes.send(ActorNote::Exited { |
| 69 | room_id: self.room_id.clone(), |
| 70 | generation: self.generation, |
| 71 | leftovers, |
| 72 | }); |
| 73 | } |
| 74 | |
| 75 | async fn serve(&mut self) { |
| 76 | let timeline = match build_timeline( |
| 77 | &self.room, |
| 78 | TimelineFocus::Live { |
| 79 | hide_threaded_events: true, |
| 80 | }, |
| 81 | ) |
| 82 | .await |
| 83 | { |
| 84 | Ok(timeline) => timeline, |
| 85 | Err(error) => return self.refuse_everything(error).await, |
| 86 | }; |
| 87 | let Ok(room_timeline) = TimelineSession::start( |
| 88 | TimelineId::Room(self.room_id.clone()), |
| 89 | timeline, |
| 90 | self.context.clone(), |
| 91 | ) |
| 92 | .await |
| 93 | else { |
| 94 | return; |
| 95 | }; |
| 96 | for reply in self.opening.drain(..) { |
| 97 | reply.done(Ok(())); |
| 98 | } |
| 99 | let (typing_guard, typing) = self.room.subscribe_to_typing_notifications(); |
| 100 | let mut typing_watcher = TaskGuard(self.context.supervisor.spawn( |
| 101 | "typing watcher", |
| 102 | watch_typing(self.room_id.clone(), typing, self.context.events.clone()), |
| 103 | )); |
| 104 | let mut timelines = Timelines { |
| 105 | room: room_timeline, |
| 106 | threads: Threads::new(self.requests.max_capacity()), |
| 107 | work: TaskSet::new(), |
| 108 | }; |
| 109 | let mut repair = RepairController::new(self.room.clone(), self.context.clone()); |
| 110 | let close = self.serve_open(&mut timelines, &mut repair).await; |
| 111 | repair.shutdown().await; |
| 112 | if close.is_none() { |
| 113 | self.requests.close(); |
| 114 | let _ = self.notes.send(ActorNote::Leaving { |
| 115 | room_id: self.room_id.clone(), |
| 116 | generation: self.generation, |
| 117 | }); |
| 118 | } |
| 119 | |
| 120 | // let redactions and receipts that already left finish rather than lie about them |
| 121 | while timelines.work.join_next().await.is_some() {} |
| 122 | drop(typing_guard); |
| 123 | typing_watcher.0.abort(); |
| 124 | let _ = (&mut typing_watcher.0).await; |
| 125 | let _ = self |
| 126 | .context |
| 127 | .events |
| 128 | .send(Event::TypingChanged(TypingState { |
| 129 | room_id: self.room_id.clone(), |
| 130 | users: Vec::new(), |
| 131 | })) |
| 132 | .await; |
| 133 | timelines.threads.close().await; |
| 134 | timelines.room.close().await; |
| 135 | if let Some(reply) = close { |
| 136 | reply.done(Ok(())); |
| 137 | } |
| 138 | } |
| 139 | |
| 140 | /// returns the close request's responder, if a close is why we stopped |
| 141 | async fn serve_open( |
| 142 | &mut self, |
| 143 | timelines: &mut Timelines, |
| 144 | repair: &mut RepairController, |
| 145 | ) -> Option<Responder> { |
| 146 | let mut room_info = self.room.subscribe_info(); |
| 147 | // a leave that landed while the timeline was building won't come round again |
| 148 | if matches!(self.room.state(), RoomState::Left | RoomState::Banned) { |
| 149 | return None; |
| 150 | } |
| 151 | loop { |
| 152 | tokio::select! { |
| 153 | Some(info) = room_info.next() => { |
| 154 | if matches!(info.state(), RoomState::Left | RoomState::Banned) { |
| 155 | tracing::debug!(room_id = %self.room_id, "room actor stopped because we left the room"); |
| 156 | return None; |
| 157 | } |
| 158 | } |
| 159 | request = self.requests.recv() => match request { |
| 160 | None => return None, |
| 161 | Some(ActorRequest::Close(reply)) => return Some(reply), |
| 162 | Some(ActorRequest::Open(reply)) => reply.done(timelines.room.resync().await), |
| 163 | Some(ActorRequest::Execute(command, reply)) => { |
| 164 | timelines.execute(command, reply, &self.room).await; |
| 165 | } |
| 166 | Some(ActorRequest::Repair(command, reply)) => { |
| 167 | repair.execute(command, reply).await; |
| 168 | } |
| 169 | }, |
| 170 | install = repair.next_install(), if repair.is_running() => { |
| 171 | if let Some(install) = install { |
| 172 | timelines.install_history(install, &self.room, &self.context).await; |
| 173 | } |
| 174 | } |
| 175 | Some(built) = timelines.threads.builds.join_next(), if !timelines.threads.builds.is_empty() => { |
| 176 | timelines.thread_built(built, &self.room_id, &self.context).await; |
| 177 | } |
| 178 | Some(label) = self.panics.recv() => panic!("{label} panicked in room {}", self.room_id), |
| 179 | Some(done) = timelines.work.join_next(), if !timelines.work.is_empty() => { |
| 180 | if let Err(error) = done && is_panic(&error) { |
| 181 | tracing::warn!(room_id = %self.room_id, %error, "timeline operation panicked"); |
| 182 | } |
| 183 | } |
| 184 | } |
| 185 | } |
| 186 | } |
| 187 | |
| 188 | async fn refuse_everything(&mut self, error: matrix_sdk_ui::timeline::Error) { |
| 189 | tracing::warn!(%error, room_id = %self.room_id, "room actor could not build its timeline"); |
| 190 | let error = Error::with_source(ErrorKind::Timeline, "room timeline build failed", error); |
| 191 | for reply in self.opening.drain(..) { |
| 192 | reply.done(Err(error.clone())); |
| 193 | } |
| 194 | self.requests.close(); |
| 195 | while let Some(request) = self.requests.recv().await { |
| 196 | request.into_reply().done(Err(error.clone())); |
| 197 | } |
| 198 | } |
| 199 | } |
| 200 | |
| 201 | async fn build_timeline( |
| 202 | room: &Room, |
| 203 | focus: TimelineFocus, |
| 204 | ) -> Result<Timeline, matrix_sdk_ui::timeline::Error> { |
| 205 | room.timeline_builder() |
| 206 | .with_focus(focus) |
| 207 | .track_read_marker_and_receipts(TimelineReadReceiptTracking::AllEvents) |
| 208 | .build() |
| 209 | .await |
| 210 | } |
| 211 | |
| 212 | async fn watch_typing( |
| 213 | room_id: OwnedRoomId, |
| 214 | mut typing: tokio::sync::broadcast::Receiver<Vec<ruma::OwnedUserId>>, |
| 215 | events: crate::transport::EventSender, |
| 216 | ) { |
| 217 | loop { |
| 218 | match typing.recv().await { |
| 219 | Ok(users) => { |
| 220 | let state = TypingState { |
| 221 | room_id: room_id.clone(), |
| 222 | users, |
| 223 | }; |
| 224 | if events.send(Event::TypingChanged(state)).await.is_err() { |
| 225 | return; |
| 226 | } |
| 227 | } |
| 228 | Err(tokio::sync::broadcast::error::RecvError::Lagged(skipped)) => { |
| 229 | tracing::warn!(%room_id, skipped, "typing watcher lagged"); |
| 230 | } |
| 231 | Err(tokio::sync::broadcast::error::RecvError::Closed) => return, |
| 232 | } |
| 233 | } |
| 234 | } |
| 235 | |
| 236 | struct Timelines { |
| 237 | room: TimelineSession, |
| 238 | threads: Threads, |
| 239 | work: TaskSet<()>, |
| 240 | } |
| 241 | |
| 242 | impl Timelines { |
| 243 | async fn install_history( |
| 244 | &mut self, |
| 245 | install: InstallRequest, |
| 246 | room: &Room, |
| 247 | context: &SessionContext, |
| 248 | ) { |
| 249 | let InstallRequest { |
| 250 | events, |
| 251 | prev_batch, |
| 252 | reply, |
| 253 | } = install; |
| 254 | let result = self |
| 255 | .install_history_inner(events, prev_batch, room, context) |
| 256 | .await; |
| 257 | let _ = reply.send(result); |
| 258 | } |
| 259 | |
| 260 | async fn install_history_inner( |
| 261 | &mut self, |
| 262 | events: Vec<matrix_sdk::deserialized_responses::TimelineEvent>, |
| 263 | prev_batch: Option<String>, |
| 264 | room: &Room, |
| 265 | context: &SessionContext, |
| 266 | ) -> Result<(), Error> { |
| 267 | let threads_capacity = self.threads.capacity; |
| 268 | self.threads.close().await; |
| 269 | self.threads = Threads::new(threads_capacity); |
| 270 | self.room.pause_for_reseed().await; |
| 271 | |
| 272 | let reseed = async { |
| 273 | let (cache, _drop_handles) = room.event_cache().await.map_err(|source| { |
| 274 | Error::with_source(ErrorKind::Timeline, "history event cache failed", source) |
| 275 | })?; |
| 276 | cache.reseed(events, prev_batch).await.map_err(|source| { |
| 277 | Error::with_source(ErrorKind::Timeline, "history cache reseed failed", source) |
| 278 | }) |
| 279 | } |
| 280 | .await; |
| 281 | if let Err(error) = reseed { |
| 282 | return match self.room.resume_after_failed_reseed().await { |
| 283 | Ok(()) => Err(error), |
| 284 | Err(restart_error) => Err(restart_error), |
| 285 | }; |
| 286 | } |
| 287 | |
| 288 | let timeline = match build_timeline( |
| 289 | room, |
| 290 | TimelineFocus::Live { |
| 291 | hide_threaded_events: true, |
| 292 | }, |
| 293 | ) |
| 294 | .await |
| 295 | { |
| 296 | Ok(timeline) => timeline, |
| 297 | Err(source) => { |
| 298 | let error = |
| 299 | Error::with_source(ErrorKind::Timeline, "room timeline rebuild failed", source); |
| 300 | return match self.room.resume_after_failed_reseed().await { |
| 301 | Ok(()) => Err(error), |
| 302 | Err(restart_error) => Err(restart_error), |
| 303 | }; |
| 304 | } |
| 305 | }; |
| 306 | let timeline_id = self.room.id.clone(); |
| 307 | let mut context = context.clone(); |
| 308 | context.initial_history_limit = 0; |
| 309 | let timeline = match TimelineSession::start(timeline_id, timeline, context).await { |
| 310 | Ok(timeline) => timeline, |
| 311 | Err(source) => { |
| 312 | let error = |
| 313 | Error::with_source(ErrorKind::Timeline, "room timeline reset failed", source); |
| 314 | return match self.room.resume_after_failed_reseed().await { |
| 315 | Ok(()) => Err(error), |
| 316 | Err(restart_error) => Err(restart_error), |
| 317 | }; |
| 318 | } |
| 319 | }; |
| 320 | let previous = std::mem::replace(&mut self.room, timeline); |
| 321 | drop(previous); |
| 322 | Ok(()) |
| 323 | } |
| 324 | |
| 325 | async fn execute(&mut self, command: TimelineCommand, reply: Responder, room: &Room) { |
| 326 | match command { |
| 327 | TimelineCommand::OpenThread { |
| 328 | room_id, |
| 329 | root_event_id, |
| 330 | } => self |
| 331 | .threads |
| 332 | .open(TimelineId::thread(room_id, root_event_id), reply, room), |
| 333 | TimelineCommand::CloseThread { |
| 334 | room_id, |
| 335 | root_event_id, |
| 336 | } => { |
| 337 | let id = TimelineId::thread(room_id, root_event_id); |
| 338 | reply.done(self.threads.close_one(&id).await); |
| 339 | } |
| 340 | TimelineCommand::OpenPinned(room_id) => { |
| 341 | self.threads.open(TimelineId::Pinned(room_id), reply, room); |
| 342 | } |
| 343 | TimelineCommand::ClosePinned(room_id) => { |
| 344 | reply.done(self.threads.close_one(&TimelineId::Pinned(room_id)).await); |
| 345 | } |
| 346 | TimelineCommand::Resync { |
| 347 | timeline_id: TimelineId::Room(_), |
| 348 | } => reply.done(self.room.resync().await), |
| 349 | command => { |
| 350 | let Some(id) = command |
| 351 | .timeline_id() |
| 352 | .filter(|id| !matches!(id, TimelineId::Room(_))) |
| 353 | .cloned() |
| 354 | else { |
| 355 | return self.room.execute(command, reply, &mut self.work).await; |
| 356 | }; |
| 357 | if matches!(id, TimelineId::Pinned(_)) |
| 358 | && matches!( |
| 359 | command, |
| 360 | TimelineCommand::Send { .. } | TimelineCommand::Paginate { .. } |
| 361 | ) |
| 362 | { |
| 363 | return reply.done(Err(Error::new( |
| 364 | ErrorKind::Invalid, |
| 365 | "the pinned timeline can't send or paginate", |
| 366 | ))); |
| 367 | } |
| 368 | if let TimelineCommand::MarkRead { |
| 369 | timeline_id, |
| 370 | event_id, |
| 371 | } = &command |
| 372 | && !self.threads.open.contains_key(&id) |
| 373 | { |
| 374 | let client = room.client(); |
| 375 | let (timeline_id, event_id) = (timeline_id.clone(), event_id.clone()); |
| 376 | self.work.spawn(async move { |
| 377 | reply.done( |
| 378 | super::commands::mark_read(&client, &timeline_id, event_id).await, |
| 379 | ); |
| 380 | }); |
| 381 | return; |
| 382 | } |
| 383 | match self.threads.route(id, command, reply) { |
| 384 | Routed::Open(session, TimelineCommand::Resync { .. }, reply) => { |
| 385 | reply.done(session.resync().await); |
| 386 | } |
| 387 | Routed::Open(session, command, reply) => { |
| 388 | session.execute(command, reply, &mut self.work).await; |
| 389 | } |
| 390 | Routed::Queued => {} |
| 391 | } |
| 392 | } |
| 393 | } |
| 394 | } |
| 395 | |
| 396 | async fn thread_built( |
| 397 | &mut self, |
| 398 | built: Result<Built, JoinError>, |
| 399 | room_id: &OwnedRoomId, |
| 400 | context: &SessionContext, |
| 401 | ) { |
| 402 | let Some((id, queued, timeline)) = self.threads.built(built) else { |
| 403 | return; |
| 404 | }; |
| 405 | let timeline = match timeline { |
| 406 | Ok(timeline) => timeline, |
| 407 | Err(error) => { |
| 408 | tracing::warn!(%error, %room_id, ?id, "timeline build failed"); |
| 409 | let error = Error::with_source(ErrorKind::Timeline, "timeline build failed", error); |
| 410 | for (_, reply) in queued { |
| 411 | reply.done(Err(error.clone())); |
| 412 | } |
| 413 | return; |
| 414 | } |
| 415 | }; |
| 416 | let Ok(session) = TimelineSession::start(id.clone(), timeline, context.clone()).await |
| 417 | else { |
| 418 | return; |
| 419 | }; |
| 420 | let session = self.threads.open.entry(id).insert_entry(session).into_mut(); |
| 421 | for (command, reply) in queued { |
| 422 | match command { |
| 423 | None => reply.done(Ok(())), |
| 424 | Some(TimelineCommand::Resync { .. }) => reply.done(session.resync().await), |
| 425 | Some(command) => session.execute(command, reply, &mut self.work).await, |
| 426 | } |
| 427 | } |
| 428 | } |
| 429 | } |
| 430 | |
| 431 | type Built = ( |
| 432 | TimelineId, |
| 433 | u64, |
| 434 | Result<Timeline, matrix_sdk_ui::timeline::Error>, |
| 435 | ); |
| 436 | |
| 437 | /// `None` is the open itself, everything else waits behind it |
| 438 | type Queued = VecDeque<(Option<TimelineCommand>, Responder)>; |
| 439 | |
| 440 | enum Routed<'a> { |
| 441 | Open(&'a mut TimelineSession, TimelineCommand, Responder), |
| 442 | Queued, |
| 443 | } |
| 444 | |
| 445 | struct OpeningThread { |
| 446 | generation: u64, |
| 447 | build: AbortHandle, |
| 448 | queued: Queued, |
| 449 | } |
| 450 | |
| 451 | /// threads and the pinned timeline, which build off the actor and queue what arrives meanwhile |
| 452 | struct Threads { |
| 453 | open: HashMap<TimelineId, TimelineSession>, |
| 454 | opening: HashMap<TimelineId, OpeningThread>, |
| 455 | builds: TaskSet<Built>, |
| 456 | next_generation: u64, |
| 457 | queued: usize, |
| 458 | capacity: usize, |
| 459 | } |
| 460 | |
| 461 | impl Threads { |
| 462 | fn new(capacity: usize) -> Self { |
| 463 | Self { |
| 464 | open: HashMap::new(), |
| 465 | opening: HashMap::new(), |
| 466 | builds: TaskSet::new(), |
| 467 | next_generation: 0, |
| 468 | queued: 0, |
| 469 | capacity, |
| 470 | } |
| 471 | } |
| 472 | |
| 473 | fn enqueue(&mut self, id: &TimelineId, command: Option<TimelineCommand>, reply: Responder) { |
| 474 | if self.queued >= self.capacity { |
| 475 | return reply.done(Err(Error::new( |
| 476 | ErrorKind::QueueFull, |
| 477 | "thread command queue is full", |
| 478 | ))); |
| 479 | } |
| 480 | let Some(opening) = self.opening.get_mut(id) else { |
| 481 | return reply.done(Err(Error::new( |
| 482 | ErrorKind::TimelineNotOpen, |
| 483 | "timeline is not open", |
| 484 | ))); |
| 485 | }; |
| 486 | opening.queued.push_back((command, reply)); |
| 487 | self.queued += 1; |
| 488 | } |
| 489 | |
| 490 | fn open(&mut self, id: TimelineId, reply: Responder, room: &Room) { |
| 491 | if self.open.contains_key(&id) { |
| 492 | return reply.done(Ok(())); |
| 493 | } |
| 494 | if !self.opening.contains_key(&id) { |
| 495 | let generation = self.next_generation; |
| 496 | self.next_generation = self.next_generation.wrapping_add(1); |
| 497 | let room = room.clone(); |
| 498 | let build_id = id.clone(); |
| 499 | let build = self.builds.spawn(async move { |
| 500 | let focus = match &build_id { |
| 501 | TimelineId::Thread { root_event_id, .. } => TimelineFocus::Thread { |
| 502 | root_event_id: root_event_id.clone(), |
| 503 | }, |
| 504 | _ => TimelineFocus::PinnedEvents, |
| 505 | }; |
| 506 | (build_id, generation, build_timeline(&room, focus).await) |
| 507 | }); |
| 508 | self.opening.insert( |
| 509 | id.clone(), |
| 510 | OpeningThread { |
| 511 | generation, |
| 512 | build, |
| 513 | queued: VecDeque::new(), |
| 514 | }, |
| 515 | ); |
| 516 | } |
| 517 | self.enqueue(&id, None, reply); |
| 518 | } |
| 519 | |
| 520 | fn route(&mut self, id: TimelineId, command: TimelineCommand, reply: Responder) -> Routed<'_> { |
| 521 | if !self.open.contains_key(&id) { |
| 522 | self.enqueue(&id, Some(command), reply); |
| 523 | return Routed::Queued; |
| 524 | } |
| 525 | let session = self.open.get_mut(&id).expect("timeline was just checked"); |
| 526 | Routed::Open(session, command, reply) |
| 527 | } |
| 528 | |
| 529 | /// `None` when the build belongs to a thread that was closed while it ran |
| 530 | fn built( |
| 531 | &mut self, |
| 532 | built: Result<Built, JoinError>, |
| 533 | ) -> Option<( |
| 534 | TimelineId, |
| 535 | Queued, |
| 536 | Result<Timeline, matrix_sdk_ui::timeline::Error>, |
| 537 | )> { |
| 538 | let (id, generation, timeline) = match built { |
| 539 | Ok(built) => built, |
| 540 | Err(error) if error.is_cancelled() => return None, |
| 541 | Err(error) => resume_unwind(error), |
| 542 | }; |
| 543 | if self |
| 544 | .opening |
| 545 | .get(&id) |
| 546 | .is_none_or(|opening| opening.generation != generation) |
| 547 | { |
| 548 | return None; |
| 549 | } |
| 550 | let opening = self.opening.remove(&id)?; |
| 551 | self.queued -= opening.queued.len(); |
| 552 | Some((id, opening.queued, timeline)) |
| 553 | } |
| 554 | |
| 555 | async fn close_one(&mut self, id: &TimelineId) -> Result<(), Error> { |
| 556 | if let Some(session) = self.open.remove(id) { |
| 557 | session.close().await; |
| 558 | return Ok(()); |
| 559 | } |
| 560 | let Some(opening) = self.opening.remove(id) else { |
| 561 | return Err(Error::new( |
| 562 | ErrorKind::TimelineNotOpen, |
| 563 | "timeline is not open", |
| 564 | )); |
| 565 | }; |
| 566 | opening.build.abort(); |
| 567 | self.queued -= opening.queued.len(); |
| 568 | Ok(()) |
| 569 | } |
| 570 | |
| 571 | async fn close(&mut self) { |
| 572 | self.builds.shutdown().await; |
| 573 | self.opening.clear(); |
| 574 | for (_, session) in self.open.drain() { |
| 575 | session.close().await; |
| 576 | } |
| 577 | } |
| 578 | } |