Delete stale rooms/participants after RECONNECT_TIMEOUT

This commit is contained in:
Antonio Scandurra 2022-12-13 11:32:37 +01:00
parent beea9b68ff
commit 6a6a032f1f
7 changed files with 588 additions and 171 deletions

View file

@ -50,7 +50,7 @@ pub struct Room {
user_store: ModelHandle<UserStore>,
subscriptions: Vec<client::Subscription>,
pending_room_update: Option<Task<()>>,
_maintain_connection: Task<Result<()>>,
maintain_connection: Option<Task<Result<()>>>,
}
impl Entity for Room {
@ -121,7 +121,7 @@ impl Room {
None
};
let _maintain_connection =
let maintain_connection =
cx.spawn_weak(|this, cx| Self::maintain_connection(this, client.clone(), cx));
Self {
@ -138,7 +138,7 @@ impl Room {
pending_room_update: None,
client,
user_store,
_maintain_connection,
maintain_connection: Some(maintain_connection),
}
}
@ -235,6 +235,8 @@ impl Room {
self.participant_user_ids.clear();
self.subscriptions.clear();
self.live_kit.take();
self.pending_room_update.take();
self.maintain_connection.take();
self.client.send(proto::LeaveRoom {})?;
Ok(())
}

View file

@ -3,5 +3,6 @@ ALTER TABLE "room_participants"
CREATE INDEX "index_project_collaborators_on_connection_id" ON "project_collaborators" ("connection_id");
CREATE UNIQUE INDEX "index_project_collaborators_on_project_id_connection_id_and_epoch" ON "project_collaborators" ("project_id", "connection_id", "connection_epoch");
CREATE INDEX "index_room_participants_on_room_id_id" ON "room_participants" ("room_id");
CREATE INDEX "index_room_participants_on_answering_connection_id" ON "room_participants" ("answering_connection_id");
CREATE UNIQUE INDEX "index_room_participants_on_answering_connection_id_and_answering_connection_epoch" ON "room_participants" ("answering_connection_id", "answering_connection_epoch");

View file

@ -131,29 +131,70 @@ impl Database {
.await
}
pub async fn delete_stale_rooms(&self) -> Result<()> {
pub async fn outdated_room_ids(&self) -> Result<Vec<RoomId>> {
self.transaction(|tx| async move {
#[derive(Copy, Clone, Debug, EnumIter, DeriveColumn)]
enum QueryAs {
RoomId,
}
Ok(room_participant::Entity::find()
.select_only()
.column(room_participant::Column::RoomId)
.distinct()
.filter(room_participant::Column::AnsweringConnectionEpoch.ne(self.epoch()))
.into_values::<_, QueryAs>()
.all(&*tx)
.await?)
})
.await
}
pub async fn refresh_room(&self, room_id: RoomId) -> Result<RoomGuard<RefreshedRoom>> {
self.room_transaction(|tx| async move {
let stale_participant_filter = Condition::all()
.add(room_participant::Column::RoomId.eq(room_id))
.add(room_participant::Column::AnsweringConnectionId.is_not_null())
.add(room_participant::Column::AnsweringConnectionEpoch.ne(self.epoch()));
let stale_participant_user_ids = room_participant::Entity::find()
.filter(stale_participant_filter.clone())
.all(&*tx)
.await?
.into_iter()
.map(|participant| participant.user_id)
.collect::<Vec<_>>();
// Delete participants who failed to reconnect.
room_participant::Entity::delete_many()
.filter(
room_participant::Column::AnsweringConnectionEpoch
.ne(self.epoch())
.or(room_participant::Column::CallingConnectionEpoch.ne(self.epoch())),
)
.filter(stale_participant_filter)
.exec(&*tx)
.await?;
room::Entity::delete_many()
.filter(
room::Column::Id.not_in_subquery(
Query::select()
.column(room_participant::Column::RoomId)
.from(room_participant::Entity)
.distinct()
.to_owned(),
),
)
.exec(&*tx)
.await?;
Ok(())
let room = self.get_room(room_id, &tx).await?;
let mut canceled_calls_to_user_ids = Vec::new();
// Delete the room if it becomes empty and cancel pending calls.
if room.participants.is_empty() {
canceled_calls_to_user_ids.extend(
room.pending_participants
.iter()
.map(|pending_participant| UserId::from_proto(pending_participant.user_id)),
);
room_participant::Entity::delete_many()
.filter(room_participant::Column::RoomId.eq(room_id))
.exec(&*tx)
.await?;
room::Entity::delete_by_id(room_id).exec(&*tx).await?;
}
Ok((
room_id,
RefreshedRoom {
room,
stale_participant_user_ids,
canceled_calls_to_user_ids,
},
))
})
.await
}
@ -2575,6 +2616,12 @@ pub struct LeftRoom {
pub canceled_calls_to_user_ids: Vec<UserId>,
}
pub struct RefreshedRoom {
pub room: proto::Room,
pub stale_participant_user_ids: Vec<UserId>,
pub canceled_calls_to_user_ids: Vec<UserId>,
}
pub struct Project {
pub collaborators: Vec<project_collaborator::Model>,
pub worktrees: BTreeMap<u64, Worktree>,

File diff suppressed because it is too large Load diff

View file

@ -2,7 +2,7 @@ pub mod api;
pub mod auth;
pub mod db;
pub mod env;
mod executor;
pub mod executor;
#[cfg(test)]
mod integration_tests;
pub mod rpc;

View file

@ -1,6 +1,6 @@
use anyhow::anyhow;
use axum::{routing::get, Router};
use collab::{db, env, AppState, Config, MigrateConfig, Result};
use collab::{db, env, executor::Executor, AppState, Config, MigrateConfig, Result};
use db::Database;
use std::{
env::args,
@ -56,7 +56,7 @@ async fn main() -> Result<()> {
let listener = TcpListener::bind(&format!("0.0.0.0:{}", state.config.http_port))
.expect("failed to bind TCP listener");
let rpc_server = collab::rpc::Server::new(state.clone());
let rpc_server = collab::rpc::Server::new(state.clone(), Executor::Production);
rpc_server.start().await?;
let app = collab::api::routes(rpc_server.clone(), state.clone())

View file

@ -142,6 +142,7 @@ pub struct Server {
peer: Arc<Peer>,
pub(crate) connection_pool: Arc<Mutex<ConnectionPool>>,
app_state: Arc<AppState>,
executor: Executor,
handlers: HashMap<TypeId, MessageHandler>,
teardown: watch::Sender<()>,
}
@ -168,10 +169,11 @@ where
}
impl Server {
pub fn new(app_state: Arc<AppState>) -> Arc<Self> {
pub fn new(app_state: Arc<AppState>, executor: Executor) -> Arc<Self> {
let mut server = Self {
peer: Peer::new(),
app_state,
executor,
connection_pool: Default::default(),
handlers: Default::default(),
teardown: watch::channel(()).0,
@ -239,8 +241,85 @@ impl Server {
pub async fn start(&self) -> Result<()> {
self.app_state.db.delete_stale_projects().await?;
// TODO: delete stale rooms after timeout.
// self.app_state.db.delete_stale_rooms().await?;
let db = self.app_state.db.clone();
let peer = self.peer.clone();
let timeout = self.executor.sleep(RECONNECT_TIMEOUT);
let pool = self.connection_pool.clone();
let live_kit_client = self.app_state.live_kit_client.clone();
self.executor.spawn_detached(async move {
timeout.await;
if let Some(room_ids) = db.outdated_room_ids().await.trace_err() {
for room_id in room_ids {
let mut contacts_to_update = HashSet::default();
let mut canceled_calls_to_user_ids = Vec::new();
let mut live_kit_room = String::new();
let mut delete_live_kit_room = false;
if let Ok(mut refreshed_room) = db.refresh_room(room_id).await {
room_updated(&refreshed_room.room, &peer);
contacts_to_update
.extend(refreshed_room.stale_participant_user_ids.iter().copied());
contacts_to_update
.extend(refreshed_room.canceled_calls_to_user_ids.iter().copied());
canceled_calls_to_user_ids =
mem::take(&mut refreshed_room.canceled_calls_to_user_ids);
dbg!(&canceled_calls_to_user_ids);
live_kit_room = mem::take(&mut refreshed_room.room.live_kit_room);
delete_live_kit_room = refreshed_room.room.participants.is_empty();
}
{
let pool = pool.lock().await;
for canceled_user_id in canceled_calls_to_user_ids {
for connection_id in pool.user_connection_ids(canceled_user_id) {
peer.send(connection_id, proto::CallCanceled {}).trace_err();
}
}
}
for user_id in contacts_to_update {
if let Some((busy, contacts)) = db
.is_user_busy(user_id)
.await
.trace_err()
.zip(db.get_contacts(user_id).await.trace_err())
{
let pool = pool.lock().await;
let updated_contact = contact_for_user(user_id, false, busy, &pool);
for contact in contacts {
if let db::Contact::Accepted {
user_id: contact_user_id,
..
} = contact
{
for contact_conn_id in pool.user_connection_ids(contact_user_id)
{
peer.send(
contact_conn_id,
proto::UpdateContacts {
contacts: vec![updated_contact.clone()],
remove_contacts: Default::default(),
incoming_requests: Default::default(),
remove_incoming_requests: Default::default(),
outgoing_requests: Default::default(),
remove_outgoing_requests: Default::default(),
},
)
.trace_err();
}
}
}
}
}
if let Some(live_kit) = live_kit_client.as_ref() {
if delete_live_kit_room {
live_kit.delete_room(live_kit_room).await.trace_err();
}
}
}
}
});
Ok(())
}
@ -690,7 +769,7 @@ async fn sign_out(
{
let db = session.db().await;
if let Some(room) = db.decline_call(None, session.user_id).await.trace_err() {
room_updated(&room, &session);
room_updated(&room, &session.peer);
}
}
update_user_contacts(session.user_id, &session).await?;
@ -768,7 +847,7 @@ async fn join_room(
session.connection_id,
)
.await?;
room_updated(&room, &session);
room_updated(&room, &session.peer);
room.clone()
};
@ -843,7 +922,7 @@ async fn call(
initial_project_id,
)
.await?;
room_updated(&room, &session);
room_updated(&room, &session.peer);
mem::take(incoming_call)
};
update_user_contacts(called_user_id, &session).await?;
@ -873,7 +952,7 @@ async fn call(
.await
.call_failed(room_id, called_user_id)
.await?;
room_updated(&room, &session);
room_updated(&room, &session.peer);
}
update_user_contacts(called_user_id, &session).await?;
@ -893,7 +972,7 @@ async fn cancel_call(
.await
.cancel_call(Some(room_id), session.connection_id, called_user_id)
.await?;
room_updated(&room, &session);
room_updated(&room, &session.peer);
}
for connection_id in session
@ -920,7 +999,7 @@ async fn decline_call(message: proto::DeclineCall, session: Session) -> Result<(
.await
.decline_call(Some(room_id), session.user_id)
.await?;
room_updated(&room, &session);
room_updated(&room, &session.peer);
}
for connection_id in session
@ -951,7 +1030,7 @@ async fn update_participant_location(
.await
.update_room_participant_location(room_id, session.connection_id, location)
.await?;
room_updated(&room, &session);
room_updated(&room, &session.peer);
response.send(proto::Ack {})?;
Ok(())
}
@ -973,7 +1052,7 @@ async fn share_project(
response.send(proto::ShareProjectResponse {
project_id: project_id.to_proto(),
})?;
room_updated(&room, &session);
room_updated(&room, &session.peer);
Ok(())
}
@ -992,7 +1071,7 @@ async fn unshare_project(message: proto::UnshareProject, session: Session) -> Re
guest_connection_ids.iter().copied(),
|conn_id| session.peer.send(conn_id, message.clone()),
);
room_updated(&room, &session);
room_updated(&room, &session.peer);
Ok(())
}
@ -1151,7 +1230,7 @@ async fn update_project(
.forward_send(session.connection_id, connection_id, request.clone())
},
);
room_updated(&room, &session);
room_updated(&room, &session.peer);
response.send(proto::Ack {})?;
Ok(())
@ -1798,17 +1877,15 @@ fn contact_for_user(
}
}
fn room_updated(room: &proto::Room, session: &Session) {
fn room_updated(room: &proto::Room, peer: &Peer) {
for participant in &room.participants {
session
.peer
.send(
ConnectionId(participant.peer_id),
proto::RoomUpdated {
room: Some(room.clone()),
},
)
.trace_err();
peer.send(
ConnectionId(participant.peer_id),
proto::RoomUpdated {
room: Some(room.clone()),
},
)
.trace_err();
}
}
@ -1860,7 +1937,7 @@ async fn leave_room_for_session(session: &Session) -> Result<()> {
project_left(project, session);
}
room_updated(&left_room.room, &session);
room_updated(&left_room.room, &session.peer);
canceled_calls_to_user_ids = mem::take(&mut left_room.canceled_calls_to_user_ids);
live_kit_room = mem::take(&mut left_room.room.live_kit_room);
delete_live_kit_room = left_room.room.participants.is_empty();