Add configuration for make_join and send_join attempt counts.

Signed-off-by: Jason Volk <jason@zemos.net>

src/api/client/membership/join.rs

@@ -5,7 +5,7 @@ use ruma::{
 	RoomId, RoomOrAliasId,
 	api::client::membership::{join_room_by_id, join_room_by_id_or_alias},
 };
-use tuwunel_core::Result;
+use tuwunel_core::{Result, warn};
 
 use super::banned_room_check;
 use crate::{Ruma, client::membership::get_join_params};
@@ -36,7 +36,8 @@ pub(crate) async fn join_room_by_id_route(
 
 	let state_lock = services.state.mutex.lock(&room_id).await;
 
-	services
+	let mut errors = 0_usize;
+	while let Err(e) = services
 		.membership
 		.join(
 			sender_user,
@@ -47,7 +48,17 @@ pub(crate) async fn join_room_by_id_route(
 			&state_lock,
 		)
 		.boxed()
-		.await?;
+		.await
+	{
+		errors = errors.saturating_add(1);
+		if errors >= services.config.max_join_attempts_per_join_request {
+			warn!(
+				"Several servers failed. Giving up for this request. Try again for different \
+				 server selection."
+			);
+			return Err(e);
+		}
+	}
 
 	drop(state_lock);
 
@@ -80,7 +91,8 @@ pub(crate) async fn join_room_by_id_or_alias_route(
 
 	let state_lock = services.state.mutex.lock(&room_id).await;
 
-	services
+	let mut errors = 0_usize;
+	while let Err(e) = services
 		.membership
 		.join(
 			sender_user,
@@ -91,7 +103,17 @@ pub(crate) async fn join_room_by_id_or_alias_route(
 			&state_lock,
 		)
 		.boxed()
-		.await?;
+		.await
+	{
+		errors = errors.saturating_add(1);
+		if errors >= services.config.max_join_attempts_per_join_request {
+			warn!(
+				"Several servers failed. Giving up for this request. Try again for different \
+				 server selection."
+			);
+			return Err(e);
+		}
+	}
 
 	drop(state_lock);
 

src/core/config/mod.rs

@@ -1757,6 +1757,26 @@ pub struct Config {
 	)]
 	pub deprioritize_joins_through_servers: RegexSet,
 
+	/// Maximum make_join requests to attempt within each join attempt. Each
+	/// attempt tries a different server, as each server is only tried once;
+	/// though retries can occur when the join request as a whole is retried.
+	///
+	/// default: 48
+	#[serde(default = "default_max_make_join_attempts_per_join_attempt")]
+	pub max_make_join_attempts_per_join_attempt: usize,
+
+	/// Maximum join attempts to conduct per client join request. Each join
+	/// attempt consists of one or more make_join requests limited above, and a
+	/// single send_join request. This value allows for additional servers to
+	/// act as the join-server prior to reporting the last error back to the
+	/// client, which can be frustrating for users. Therefor the default value
+	/// is greater than one, but less than excessively exceeding the client's
+	/// request timeout, though that may not be avoidable in some cases.
+	///
+	/// default: 3
+	#[serde(default = "default_max_join_attempts_per_join_request")]
+	pub max_join_attempts_per_join_request: usize,
+
 	/// Retry failed and incomplete messages to remote servers immediately upon
 	/// startup. This is called bursting. If this is disabled, said messages may
 	/// not be delivered until more messages are queued for that server. Do not
@@ -2997,3 +3017,7 @@ fn default_deprioritize_joins_through_servers() -> RegexSet {
 }
 
 fn default_one_time_key_limit() -> usize { 256 }
+
+fn default_max_make_join_attempts_per_join_attempt() -> usize { 48 }
+
+fn default_max_join_attempts_per_join_request() -> usize { 3 }

src/service/membership/join.rs

@@ -806,10 +806,15 @@ async fn make_join_request(
 				return make_join_response_and_server;
 			}
 
-			if make_join_counter > 40 {
+			let max_attempts = self
+				.services
+				.config
+				.max_make_join_attempts_per_join_attempt;
+			if make_join_counter >= max_attempts {
+				warn!(?remote_server, "last make_join failure reason: {e}");
 				warn!(
-					"40 servers failed to provide valid make_join response, assuming no server \
-					 can assist in joining."
+					"{max_attempts} servers failed to provide valid make_join response, \
+					 assuming no server can assist in joining."
 				);
 				make_join_response_and_server =
 					Err!(BadServerResponse("No server available to assist in joining."));

tuwunel-example.toml

@@ -1516,6 +1516,22 @@
 #
 #deprioritize_joins_through_servers = ["matrix\.org"]
 
+# Maximum make_join requests to attempt within each join attempt. Each
+# attempt tries a different server, as each server is only tried once;
+# though retries can occur when the join request as a whole is retried.
+#
+#max_make_join_attempts_per_join_attempt = 48
+
+# Maximum join attempts to conduct per client join request. Each join
+# attempt consists of one or more make_join requests limited above, and a
+# single send_join request. This value allows for additional servers to
+# act as the join-server prior to reporting the last error back to the
+# client, which can be frustrating for users. Therefor the default value
+# is greater than one, but less than excessively exceeding the client's
+# request timeout, though that may not be avoidable in some cases.
+#
+#max_join_attempts_per_join_request = 3
+
 # Retry failed and incomplete messages to remote servers immediately upon
 # startup. This is called bursting. If this is disabled, said messages may
 # not be delivered until more messages are queued for that server. Do not