Skip to content

Commit

Permalink
feat(gateway)!: swallow WebSocket errors
Browse files Browse the repository at this point in the history
Reading received messages on shutdown requires verbosely matching on WebSocket errors to prevent the shard from automatically reconnecting. While potentially useful for debugging, my experience is that WebSocket errors are generally just noise, and as such they are replaced with close messages with a [status code of 1006](https://datatracker.ietf.org/doc/html/rfc6455#section-7.4.1) indicating abnormal closure. This has the secondary benefit of removing confusion about transport errors being somehow actionable (any internet connection may naturally error at any time for reasons outside of your control).
  • Loading branch information
vilgotf committed Jul 6, 2024
1 parent 6a44028 commit 470e221
Show file tree
Hide file tree
Showing 4 changed files with 31 additions and 48 deletions.
6 changes: 0 additions & 6 deletions twilight-gateway/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -87,12 +87,6 @@ async fn runner(mut shard: Shard) {
let event = match item {
Ok(Event::GatewayClose(_)) if SHUTDOWN.load(Ordering::Relaxed) => break,
Ok(event) => event,
Err(source)
if SHUTDOWN.load(Ordering::Relaxed)
&& matches!(source.kind(), ReceiveMessageErrorType::WebSocket) =>
{
break
}
Err(source) => {
tracing::warn!(?source, "error receiving event");
Expand Down
14 changes: 1 addition & 13 deletions twilight-gateway/src/error.rs
Original file line number Diff line number Diff line change
Expand Up @@ -175,14 +175,6 @@ impl ReceiveMessageError {
source: Some(Box::new(source)),
}
}

/// Shortcut to create a new error for a websocket error.
pub(crate) fn from_websocket(source: tokio_websockets::Error) -> Self {
Self {
kind: ReceiveMessageErrorType::WebSocket,
source: Some(Box::new(source)),
}
}
}

impl Display for ReceiveMessageError {
Expand All @@ -197,7 +189,6 @@ impl Display for ReceiveMessageError {
f.write_str(event)
}
ReceiveMessageErrorType::Reconnect => f.write_str("failed to reconnect to the gateway"),
ReceiveMessageErrorType::WebSocket => f.write_str("websocket connection error"),
}
}
}
Expand Down Expand Up @@ -228,8 +219,6 @@ pub enum ReceiveMessageErrorType {
},
/// Shard failed to reconnect to the gateway.
Reconnect,
/// WebSocket connection error.
WebSocket,
}

#[cfg(test)]
Expand All @@ -243,7 +232,7 @@ mod tests {

#[test]
fn receive_message_error_display() {
let messages: [(ReceiveMessageErrorType, &str); 4] = [
let messages: [(ReceiveMessageErrorType, &str); 3] = [
(
ReceiveMessageErrorType::Compression,
"binary message could not be decompressed",
Expand All @@ -258,7 +247,6 @@ mod tests {
ReceiveMessageErrorType::Reconnect,
"failed to reconnect to the gateway",
),
(ReceiveMessageErrorType::WebSocket, "websocket connection error"),
];

for (kind, message) in messages {
Expand Down
3 changes: 3 additions & 0 deletions twilight-gateway/src/message.rs
Original file line number Diff line number Diff line change
Expand Up @@ -24,6 +24,9 @@ pub enum Message {
}

impl Message {
/// Close message indicating the connection was closed abnormally.
pub(crate) const ABNORMAL_CLOSE: Self = Self::Close(Some(CloseFrame::new(1006, "")));

/// Whether the message is a close message.
pub const fn is_close(&self) -> bool {
matches!(self, Self::Close(_))
Expand Down
56 changes: 27 additions & 29 deletions twilight-gateway/src/shard.rs
Original file line number Diff line number Diff line change
Expand Up @@ -437,8 +437,7 @@ impl<Q> Shard<Q> {
/// continue showing the bot as online until its presence times out.
///
/// To read all remaining messages, continue calling [`poll_next`] until it
/// returns [`Message::Close`] or a [`ReceiveMessageErrorType::WebSocket`]
/// error type.
/// returns [`Message::Close`].
///
/// # Example
///
Expand All @@ -457,7 +456,6 @@ impl<Q> Shard<Q> {
/// match item {
/// Ok(Message::Close(_)) => break,
/// Ok(Message::Text(_)) => unimplemented!(),
/// Err(source) if matches!(source.kind(), ReceiveMessageErrorType::WebSocket) => break,
/// Err(source) => unimplemented!(),
/// }
/// }
Expand Down Expand Up @@ -550,19 +548,16 @@ impl<Q> Shard<Q> {
}

/// Send and flush the pending message.
fn poll_flush_pending(
&mut self,
cx: &mut Context<'_>,
) -> Poll<Result<(), ReceiveMessageError>> {
fn poll_flush_pending(&mut self, cx: &mut Context<'_>) -> Poll<Result<(), WebsocketError>> {
if self.pending.is_none() {
return Poll::Ready(Ok(()));
}

ready!(Pin::new(self.connection.as_mut().unwrap()).poll_ready(cx)).map_err(|source| {
if let Err(e) = ready!(Pin::new(self.connection.as_mut().unwrap()).poll_ready(cx)) {
self.disconnect(CloseInitiator::Transport);
self.connection = None;
ReceiveMessageError::from_websocket(source)
})?;
return Poll::Ready(Err(e));
}

let pending = self.pending.as_mut().unwrap();

Expand All @@ -574,18 +569,17 @@ impl<Q> Shard<Q> {
}

let ws_message = pending.gateway_event.take().unwrap().into_websocket_msg();
if let Err(source) = Pin::new(self.connection.as_mut().unwrap()).start_send(ws_message)
{
if let Err(e) = Pin::new(self.connection.as_mut().unwrap()).start_send(ws_message) {
self.disconnect(CloseInitiator::Transport);
self.connection = None;
return Poll::Ready(Err(ReceiveMessageError::from_websocket(source)));
return Poll::Ready(Err(e));
}
}

if let Err(source) = ready!(Pin::new(self.connection.as_mut().unwrap()).poll_flush(cx)) {
if let Err(e) = ready!(Pin::new(self.connection.as_mut().unwrap()).poll_flush(cx)) {
self.disconnect(CloseInitiator::Transport);
self.connection = None;
return Poll::Ready(Err(ReceiveMessageError::from_websocket(source)));
return Poll::Ready(Err(e));
}

if pending.is_heartbeat {
Expand Down Expand Up @@ -798,7 +792,9 @@ impl<Q: Queue + Unpin> Stream for Shard<Q> {
_ => {}
}

ready!(self.poll_flush_pending(cx))?;
if ready!(self.poll_flush_pending(cx)).is_err() {
return Poll::Ready(Some(Ok(Message::ABNORMAL_CLOSE)));
}

if !self.state.is_disconnected() {
if let Poll::Ready(frame) = self.user_channel.close_rx.poll_recv(cx) {
Expand All @@ -807,7 +803,9 @@ impl<Q: Queue + Unpin> Stream for Shard<Q> {
tracing::debug!("sending close frame from user channel");
self.disconnect(CloseInitiator::Shard(frame));

ready!(self.poll_flush_pending(cx))?;
if ready!(self.poll_flush_pending(cx)).is_err() {
return Poll::Ready(Some(Ok(Message::ABNORMAL_CLOSE)));
}
}
}

Expand All @@ -834,7 +832,9 @@ impl<Q: Queue + Unpin> Stream for Shard<Q> {
self.heartbeat_interval_event = false;
}

ready!(self.poll_flush_pending(cx))?;
if ready!(self.poll_flush_pending(cx)).is_err() {
return Poll::Ready(Some(Ok(Message::ABNORMAL_CLOSE)));
}
}

let not_ratelimited = self
Expand Down Expand Up @@ -873,7 +873,9 @@ impl<Q: Queue + Unpin> Stream for Shard<Q> {
);
self.identify_rx = None;

ready!(self.poll_flush_pending(cx))?;
if ready!(self.poll_flush_pending(cx)).is_err() {
return Poll::Ready(Some(Ok(Message::ABNORMAL_CLOSE)));
}
}
}

Expand All @@ -884,7 +886,9 @@ impl<Q: Queue + Unpin> Stream for Shard<Q> {
tracing::debug!("sending command from user channel");
self.pending = Pending::text(command, false);

ready!(self.poll_flush_pending(cx))?;
if ready!(self.poll_flush_pending(cx)).is_err() {
return Poll::Ready(Some(Ok(Message::ABNORMAL_CLOSE)));
}
}
}

Expand Down Expand Up @@ -921,24 +925,18 @@ impl<Q: Queue + Unpin> Stream for Shard<Q> {
{
continue
}
Some(Err(source)) => {
Some(Err(_)) => {
self.disconnect(CloseInitiator::Transport);

return Poll::Ready(Some(Err(ReceiveMessageError {
kind: ReceiveMessageErrorType::WebSocket,
source: Some(Box::new(source)),
})));
return Poll::Ready(Some(Ok(Message::ABNORMAL_CLOSE)));
}
None => {
let res = ready!(Pin::new(self.connection.as_mut().unwrap()).poll_close(cx));
_ = ready!(Pin::new(self.connection.as_mut().unwrap()).poll_close(cx));
tracing::debug!("gateway WebSocket connection closed");
// Unclean closure.
if !self.state.is_disconnected() {
self.disconnect(CloseInitiator::Transport);
}
self.connection = None;

res.map_err(ReceiveMessageError::from_websocket)?;
}
}
};
Expand Down

0 comments on commit 470e221

Please sign in to comment.