fix(websocket): lossless EOF signal + pending sweep (WS-02, CON-02)

Replace the axum/tungstenite pump paths' Notify-based read-EOF signal
with a retained tokio watch channel so a late subscriber observes EOF
regardless of when it fired. Extend the from_wss drop monitor to sweep
the pending map (1 s interval) once EOF is observed, so calls
registered after the initial fail_all also resolve retryable instead
of hanging.

cargo test; cargo clippy --all-targets -- -D warnings (default +
all-features); cargo fmt --check
This commit is contained in:
2026-08-29 08:48:10 +00:00
parent a943d142c4
commit 5ff88756eb
3 changed files with 85 additions and 28 deletions
+43 -10
View File
@@ -23,10 +23,18 @@
//! Reconnect policy (OQ-03 disposition): v1 = none. A connection drop //! Reconnect policy (OQ-03 disposition): v1 = none. A connection drop
//! fails in-flight calls retryable: alkcall's client-side read pump only //! fails in-flight calls retryable: alkcall's client-side read pump only
//! routes envelopes and does not observe EOF, so the adapter owns drop //! routes envelopes and does not observe EOF, so the adapter owns drop
//! semantics — the [`WssSession`] monitor awaits the WS read pump and //! semantics — the [`WssSession`] monitor awaits the WS read pump's EOF
//! fails all pending calls with retryable `CONNECTION_CLOSED`. Subsequent //! signal and fails all pending calls with retryable `CONNECTION_CLOSED`.
//! handler calls fail on write; reconnect policy is the assembly layer's //! The EOF signal is lossless (a retained watch value, WS-02): EOF is
//! job. //! observed even if it fires before the monitor starts, or after the
//! session was forgotten (fire-and-forget import). The monitor also
//! sweeps the pending map once a second while the session lives: calls
//! registered *after* the initial `fail_all` (still possible on the
//! forgotten-session path, where imports may race the drop) are failed
//! on the next sweep tick once EOF has been observed. No hang: pendings
//! resolve retryable regardless of registration-vs-EOF ordering
//! (CON-02). Subsequent handler calls fail on write; reconnect policy is
//! the assembly layer's job.
//! //!
//! [ADR-070]: https://docs.rs/alkhttp (docs/architecture/decisions) //! [ADR-070]: https://docs.rs/alkhttp (docs/architecture/decisions)
@@ -43,6 +51,11 @@ use tokio_tungstenite::tungstenite::client::IntoClientRequest;
use crate::websocket::split_tungstenite_to_bytes; use crate::websocket::split_tungstenite_to_bytes;
/// How often the drop monitor sweeps the pending map for calls
/// registered after the last `fail_all` (only effective once EOF has
/// been observed — see the module doc).
const PENDING_SWEEP_INTERVAL: std::time::Duration = std::time::Duration::from_secs(1);
pub struct FromWss { pub struct FromWss {
endpoint: String, endpoint: String,
auth_token: Option<String>, auth_token: Option<String>,
@@ -151,18 +164,38 @@ impl WssSession {
})?; })?;
let call_connection = Arc::new(call_connection); let call_connection = Arc::new(call_connection);
let (close_tx, close_rx) = tokio::sync::oneshot::channel(); let (close_tx, mut close_rx) = tokio::sync::oneshot::channel();
let pending = Arc::clone(call_connection.pending()); let pending = Arc::clone(call_connection.pending());
let mut eof_rx = pumps.read_eof();
tokio::spawn(async move { tokio::spawn(async move {
let mut sweep = tokio::time::interval(PENDING_SWEEP_INTERVAL);
sweep.set_missed_tick_behavior(tokio::time::MissedTickBehavior::Delay);
loop {
tokio::select! { tokio::select! {
_ = pumps.read_eof() => {}, _ = crate::websocket::wait_for_eof(&mut eof_rx) => break,
_ = close_rx => {}, _ = &mut close_rx => break,
} _ = sweep.tick() => {
pending.lock().fail_all(CallError::new( if *eof_rx.borrow() {
pending.lock().fail_all(
CallError::new(
"CONNECTION_CLOSED", "CONNECTION_CLOSED",
"from_wss connection dropped", "from_wss connection dropped",
true, true,
)); ),
);
}
}
}
}
let failure = CallError::new("CONNECTION_CLOSED", "from_wss connection dropped", true);
let mut guard = pending.lock();
guard.fail_all(failure.clone());
if *eof_rx.borrow() {
let swept = guard.fail_all(failure);
if !swept.is_empty() {
tracing::debug!("from_wss: sweep failed {} late pendings", swept.len());
}
}
}); });
Ok(Self { Ok(Self {
+37 -15
View File
@@ -36,9 +36,6 @@ use std::{
task::{Context, Poll}, task::{Context, Poll},
}; };
#[cfg(any(test, feature = "wss"))]
use std::sync::Arc;
use axum::extract::ws::{CloseFrame, Message as AxumMessage, WebSocket}; use axum::extract::ws::{CloseFrame, Message as AxumMessage, WebSocket};
use futures::channel::mpsc as futures_mpsc; use futures::channel::mpsc as futures_mpsc;
use futures::{SinkExt, StreamExt}; use futures::{SinkExt, StreamExt};
@@ -84,7 +81,7 @@ pub struct WsPumps {
read_task: tokio::task::JoinHandle<()>, read_task: tokio::task::JoinHandle<()>,
write_task: tokio::task::JoinHandle<()>, write_task: tokio::task::JoinHandle<()>,
#[cfg(feature = "wss")] #[cfg(feature = "wss")]
read_eof: Arc<tokio::sync::Notify>, read_eof: tokio::sync::watch::Sender<bool>,
} }
impl WsPumps { impl WsPumps {
@@ -93,12 +90,35 @@ impl WsPumps {
self.write_task.abort(); self.write_task.abort();
} }
/// Fires when the WS read side reaches EOF (socket close from either /// A lossless receiver for the WS read-EOF signal (socket close from
/// side) — used by `from_wss`'s connection-drop monitor to await /// either side): the watch channel retains the latest value, so an
/// socket EOF (ADR-070). /// EOF signaled at any point — including before the receiver is
/// taken or the observer starts awaiting — is still observed, and
/// may be observed repeatedly. Used by `from_wss`'s
/// connection-drop monitor (ADR-070); await it with
/// [`wait_for_eof`].
#[cfg(feature = "wss")] #[cfg(feature = "wss")]
pub(crate) async fn read_eof(&self) { pub(crate) fn read_eof(&self) -> tokio::sync::watch::Receiver<bool> {
self.read_eof.notified().await; self.read_eof.subscribe()
}
}
/// Resolves once the given WS read-EOF watch receiver observes `true`.
///
/// Lossless by construction: a `watch::Receiver` retains the latest
/// value, so an EOF signaled before this call (or before the receiver
/// existed) is still observed — the property a one-shot `Notify` lacked
/// (WS-02). Returns early (treated as EOF) if the sender half is
/// dropped, e.g. the read pump was aborted.
#[cfg(feature = "wss")]
pub(crate) async fn wait_for_eof(rx: &mut tokio::sync::watch::Receiver<bool>) {
loop {
if *rx.borrow_and_update() {
return;
}
if rx.changed().await.is_err() {
return;
}
} }
} }
@@ -111,11 +131,11 @@ pub fn split_ws_to_bytes(socket: WebSocket) -> (WsByteStream, WsPumps) {
let (write_tx, mut write_rx) = futures_mpsc::channel::<WriteMsg>(WRITE_SLOTS); let (write_tx, mut write_rx) = futures_mpsc::channel::<WriteMsg>(WRITE_SLOTS);
#[cfg(feature = "wss")] #[cfg(feature = "wss")]
let read_eof = Arc::new(tokio::sync::Notify::new()); let read_eof = tokio::sync::watch::channel(false).0;
let write_tx_for_read = write_tx.clone(); let write_tx_for_read = write_tx.clone();
#[cfg(feature = "wss")] #[cfg(feature = "wss")]
let read_eof_for_task = Arc::clone(&read_eof); let read_eof_for_task = read_eof.clone();
let read_task = tokio::spawn(async move { let read_task = tokio::spawn(async move {
while let Some(msg) = ws_stream.next().await { while let Some(msg) = ws_stream.next().await {
match msg { match msg {
@@ -136,7 +156,9 @@ pub fn split_ws_to_bytes(socket: WebSocket) -> (WsByteStream, WsPumps) {
} }
} }
#[cfg(feature = "wss")] #[cfg(feature = "wss")]
read_eof_for_task.notify_waiters(); {
let _ = read_eof_for_task.send(true);
}
}); });
let write_task = tokio::spawn(async move { let write_task = tokio::spawn(async move {
@@ -293,10 +315,10 @@ where
let (read_tx, read_rx) = mpsc::channel::<Vec<u8>>(READ_SLOTS); let (read_tx, read_rx) = mpsc::channel::<Vec<u8>>(READ_SLOTS);
let (write_tx, mut write_rx) = futures_mpsc::channel::<WriteMsg>(WRITE_SLOTS); let (write_tx, mut write_rx) = futures_mpsc::channel::<WriteMsg>(WRITE_SLOTS);
let read_eof = Arc::new(tokio::sync::Notify::new()); let read_eof = tokio::sync::watch::channel(false).0;
let write_tx_for_read = write_tx.clone(); let write_tx_for_read = write_tx.clone();
let read_eof_for_task = Arc::clone(&read_eof); let read_eof_for_task = read_eof.clone();
let read_task = tokio::spawn(async move { let read_task = tokio::spawn(async move {
while let Some(msg) = ws_stream.next().await { while let Some(msg) = ws_stream.next().await {
match msg { match msg {
@@ -316,7 +338,7 @@ where
Ok(_) => {} Ok(_) => {}
} }
} }
read_eof_for_task.notify_waiters(); let _ = read_eof_for_task.send(true);
}); });
let write_task = tokio::spawn(async move { let write_task = tokio::spawn(async move {
+2
View File
@@ -18,6 +18,8 @@ pub use byte_adapter::{
#[cfg(any(test, feature = "wss"))] #[cfg(any(test, feature = "wss"))]
pub use byte_adapter::split_tungstenite_to_bytes; pub use byte_adapter::split_tungstenite_to_bytes;
#[cfg(feature = "wss")]
pub(crate) use byte_adapter::wait_for_eof;
pub use upgrade::{run_channels_session, ws_bearer_auth, ws_upgrade_handler}; pub use upgrade::{run_channels_session, ws_bearer_auth, ws_upgrade_handler};
#[cfg(any(test, feature = "test-support"))] #[cfg(any(test, feature = "test-support"))]