fix(mm): call mm-lobby-cleanup from mm-gc even for preemptive lobbies without sql row

svc/pkg/mm/ops/lobby-find-fail/src/lib.rs

@@ -30,7 +30,9 @@ async fn handle(
 			futs.push(fail_query(ctx.clone(), redis, query_id, ctx.error_code).boxed());
 		}
 	}
-	futures_util::future::try_join_all(futs).await?;
+	if !futs.is_empty() {
+		futures_util::future::try_join_all(futs).await?;
+	}
 
 	Ok(mm::lobby_find_fail::Response {})
 }

svc/pkg/mm/ops/lobby-find-lobby-query-list/src/lib.rs

@@ -2,6 +2,8 @@ use proto::backend::pkg::*;
 use redis::AsyncCommands;
 use rivet_operation::prelude::*;
 
+const MAX_COUNT: isize = 16;
+
 #[operation(name = "mm-lobby-find-lobby-query-list")]
 async fn handle(
 	ctx: OperationContext<mm::lobby_find_lobby_query_list::Request>,
@@ -11,7 +13,7 @@ async fn handle(
 	let query_ids = ctx
 		.redis_mm()
 		.await?
-		.zrange::<_, Vec<String>>(util_mm::key::lobby_find_queries(lobby_id), 0, -1)
+		.zrange::<_, Vec<String>>(util_mm::key::lobby_find_queries(lobby_id), 0, MAX_COUNT - 1)
 		.await?
 		.iter()
 		.map(String::as_str)
@@ -20,5 +22,10 @@ async fn handle(
 		.map(common::Uuid::from)
 		.collect::<Vec<common::Uuid>>();
 
+	if query_ids.len() as isize == MAX_COUNT {
+		tracing::warn!("too many find queries, short circuiting to prevent bad things from happening");
+		return Ok(mm::lobby_find_lobby_query_list::Response { query_ids: Vec::new() })
+	}
+
 	Ok(mm::lobby_find_lobby_query_list::Response { query_ids })
 }

svc/pkg/mm/worker/src/workers/lobby_cleanup.rs

@@ -57,6 +57,8 @@ async fn worker(ctx: &OperationContext<mm::msg::lobby_cleanup::Message>) -> Glob
 		// This is idempotent, don't raise error
 		tracing::info!("lobby not present in redis");
 
+		remove_from_redis_without_config(&mut redis_mm, lobby_id).await?;
+
 		false
 	};
 
@@ -256,3 +258,18 @@ async fn remove_from_redis(
 
 	Ok(())
 }
+
+async fn remove_from_redis_without_config(
+	redis_mm: &mut RedisPool,
+	lobby_id: Uuid,
+) -> GlobalResult<()> {
+	let mut pipe = redis::pipe();
+	pipe.atomic()
+		.unlink(util_mm::key::lobby_config(lobby_id))
+		.unlink(util_mm::key::lobby_tags(lobby_id))
+		.zrem(util_mm::key::lobby_unready(), lobby_id.to_string())
+		.query_async(redis_mm)
+		.await?;
+
+	Ok(())
+}

svc/pkg/mm/worker/src/workers/lobby_stop.rs

@@ -11,6 +11,11 @@ struct LobbyRow {
 async fn worker(ctx: &OperationContext<mm::msg::lobby_stop::Message>) -> GlobalResult<()> {
 	let lobby_id = unwrap_ref!(ctx.lobby_id).as_uuid();
 
+	if ctx.req_dt() > util::duration::minutes(5) {
+		tracing::error!("discarding stale message");
+		return Ok(());
+	}
+
 	// Fetch the lobby.
 	//
 	// This also ensures that mm-lobby-find or mm-lobby-create
@@ -38,24 +43,24 @@ async fn worker(ctx: &OperationContext<mm::msg::lobby_stop::Message>) -> GlobalR
 	.await?;
 	tracing::info!(?lobby_row, "lobby row");
 
-	let Some(lobby_row) = lobby_row else {
-		if ctx.req_dt() > util::duration::minutes(5) {
-			tracing::error!("discarding stale message");
-			return Ok(());
-		} else {
-			retry_bail!("lobby not found, may be race condition with insertion");
-		}
-	};
-
-	// conflicting locks on the lobby row
 	// Cleanup the lobby ASAP.
 	//
+	// Conflicting locks on the lobby row, so dont cleanup after the SQL query but before the retry_bail in
+	// case the lobby does not exist in the db. lobby_cleanup will remove it from Redis
+	// appropriately.
+	//
 	// This will also be called in `job-run-cleanup`, but this is idempotent.
 	msg!([ctx] mm::msg::lobby_cleanup(lobby_id) {
 		lobby_id: Some(lobby_id.into()),
 	})
 	.await?;
 
+	let Some(lobby_row) = lobby_row else {
+		// Don't use `retry_bail` because this will retry frequently, and we need to call
+		// `mm::msg::lobby_cleanup` first
+		bail!("lobby not found, may be race condition with insertion");
+	};
+
 	// Stop the job. This will call cleanup and delete the lobby row.
 	if let Some(run_id) = lobby_row.run_id {
 		msg!([ctx] job_run::msg::stop(run_id) {