fix: Throttle client-triggered remote event fetches

changelog.d/2090.bugfix.md

@@ -0,0 +1 @@
+Fixed events being repeatedly re-requested over federation when they were fetched successfully but could not be resolved, and throttled remote event fetches made on behalf of clients. Contributed by @eleboucher.

src/api/client/room/event.rs

@@ -1,6 +1,5 @@
 use axum::extract::State;
 use conduwuit::{Err, Event, Result, debug_warn, err};
-use futures::{FutureExt, TryFutureExt, future::try_join};
 use ruma::api::client::room::get_room_event;
 
 use crate::{Ruma, client::is_ignored_pdu};
@@ -16,19 +15,39 @@ pub(crate) async fn get_room_event_route(
 	let event_id = &body.event_id;
 	let room_id = &body.room_id;
 
-	let event = services
-		.rooms
-		.timeline
-		.get_remote_pdu(room_id, event_id)
-		.map_err(|_| err!(Request(NotFound("Event {} not found.", event_id))));
+	let mut event = match services.rooms.timeline.get_pdu(event_id).await {
+		| Ok(event) => event,
+		| Err(_) => {
+			// Only fetch over federation for users who could see the event.
+			if !services
+				.rooms
+				.state_cache
+				.is_joined(sender_user, room_id)
+				.await && !services
+				.rooms
+				.state_accessor
+				.is_world_readable(room_id)
+				.await
+			{
+				return Err!(Request(NotFound("Event {} not found.", event_id)));
+			}
 
+			services
+				.rooms
+				.timeline
+				.get_remote_pdu(room_id, event_id)
+				.await
+				.map_err(|_| err!(Request(NotFound("Event {} not found.", event_id))))?
+		},
+	};
+
+	// NOTE: checked after fetching, as visibility of an unknown event cannot
+	// be determined (user_can_see_event returns true for those).
 	let visible = services
 		.rooms
 		.state_accessor
 		.user_can_see_event(sender_user, room_id, event_id)
-		.map(Ok);
-
-	let (mut event, visible) = try_join(event, visible).await?;
+		.await;
 
 	if !visible || is_ignored_pdu(services, &event, sender_user).await? {
 		return Err!(Request(Forbidden("You don't have permission to view this event.")));

src/service/rooms/timeline/backfill.rs

@@ -120,6 +120,13 @@ impl super::Service {
 			return Err!(Request(NotFound("No one can backfill this PDU, room is empty.")));
 		}
 
+		self.services
+			.event_handler
+			.ensure_can_pull_event(event_id)
+			.inspect_err(|e| {
+				debug_warn!(%event_id, error=?e, "Not attempting to fetch remote PDU");
+			})?;
+
 		let servers = self.candidate_backfill_servers(room_id).await;
 
 		for backfill_server in servers {
@@ -145,7 +152,12 @@ impl super::Service {
 						.event_handler
 						.handle_incoming_pdu(&backfill_server, room_id, event_id, value, true)
 						.boxed()
-						.await?;
+						.await
+						.inspect_err(|_| {
+							self.services
+								.event_handler
+								.hit_failed_pdu_pull(event_id.to_owned());
+						})?;
 					debug!("Successfully backfilled {event_id} from {backfill_server}");
 					Some(self.get_pdu(event_id).await)
 				},
@@ -156,10 +168,14 @@ impl super::Service {
 			};
 			if let Some(pdu) = pdu {
 				debug!("Fetched {event_id} from {backfill_server}");
+				self.services.event_handler.clear_failed_pdu(event_id);
 				return pdu;
 			}
 		}
 
+		self.services
+			.event_handler
+			.hit_failed_pdu_pull(event_id.to_owned());
 		Err!("No servers could be used to fetch {} in {}.", room_id, event_id)
 	}