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
+46 -13
View File
@@ -23,10 +23,18 @@
//! Reconnect policy (OQ-03 disposition): v1 = none. A connection drop
//! fails in-flight calls retryable: alkcall's client-side read pump only
//! routes envelopes and does not observe EOF, so the adapter owns drop
//! semantics — the [`WssSession`] monitor awaits the WS read pump and
//! fails all pending calls with retryable `CONNECTION_CLOSED`. Subsequent
//! handler calls fail on write; reconnect policy is the assembly layer's
//! job.
//! semantics — the [`WssSession`] monitor awaits the WS read pump's EOF
//! signal and fails all pending calls with retryable `CONNECTION_CLOSED`.
//! The EOF signal is lossless (a retained watch value, WS-02): EOF is
//! 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)
@@ -43,6 +51,11 @@ use tokio_tungstenite::tungstenite::client::IntoClientRequest;
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 {
endpoint: String,
auth_token: Option<String>,
@@ -151,18 +164,38 @@ impl WssSession {
})?;
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 mut eof_rx = pumps.read_eof();
tokio::spawn(async move {
tokio::select! {
_ = pumps.read_eof() => {},
_ = close_rx => {},
let mut sweep = tokio::time::interval(PENDING_SWEEP_INTERVAL);
sweep.set_missed_tick_behavior(tokio::time::MissedTickBehavior::Delay);
loop {
tokio::select! {
_ = crate::websocket::wait_for_eof(&mut eof_rx) => break,
_ = &mut close_rx => break,
_ = sweep.tick() => {
if *eof_rx.borrow() {
pending.lock().fail_all(
CallError::new(
"CONNECTION_CLOSED",
"from_wss connection dropped",
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());
}
}
pending.lock().fail_all(CallError::new(
"CONNECTION_CLOSED",
"from_wss connection dropped",
true,
));
});
Ok(Self {