fix(memcard): preserve watchdog timeout reasons across completion

This commit is contained in:
2026-10-14 10:26:30 +01:00
parent e0e01ed1de
commit bad942fb25
4 changed files with 314 additions and 25 deletions
@@ -65,6 +65,8 @@ struct Request {
control: Arc<PublicationControl>,
receiver: mpsc::Receiver<Completion>,
admitted: Instant,
watchdog_canceled_at: Option<Instant>,
worker_error: Option<Error>,
result: RequestResult,
job: AsyncWorkJobId,
released: bool,
@@ -523,6 +525,8 @@ impl Session {
control,
receiver,
admitted,
watchdog_canceled_at: None,
worker_error: None,
result: RequestResult { id, state: RequestState::Pending, reason: RequestReason::None },
job,
released: !saves_payload,
@@ -664,6 +668,8 @@ impl Session {
control,
receiver,
admitted: now,
watchdog_canceled_at: None,
worker_error: None,
result: RequestResult { id, state: RequestState::Pending, reason: RequestReason::None },
job,
released: false,
@@ -686,27 +692,29 @@ impl Session {
let Some(r) = &mut self.request else { return };
match r.receiver.try_recv() {
Ok(completion) => {
r.worker_error = completion.result.error;
let current =
self.mode.is_some() && r.session == self.context && r.epoch == self.epoch;
if current {
self.slots[r.local].protected = false;
if r.result.reason == RequestReason::None {
if r.result.state == RequestState::Pending {
r.result.state = match completion.result.durable {
DurableResult::Success => RequestState::Success,
DurableResult::Failure => RequestState::Failure,
DurableResult::Indeterminate => RequestState::Indeterminate,
};
r.result.reason = completion
.result
.error
.map_or(RequestReason::None, RequestReason::Storage);
r.result.reason = if r.watchdog_canceled_at.is_some() {
RequestReason::Timeout
} else {
r.worker_error.map_or(RequestReason::None, RequestReason::Storage)
};
if completion.result.durable == DurableResult::Success && r.saves_payload {
let mut s = r.source.write().unwrap();
s.descriptor =
completion.card.current().slots[self.logical[r.local] as usize];
s.dirty = false;
}
} // A timeout result is immutable; returned card requires explicit reconciliation.
} // A restart escalation is immutable; returned card requires explicit reconciliation.
if self.provisional && completion.result.durable != DurableResult::Success {
self.boot_error =
Some(completion.result.error.unwrap_or(Error::InvalidState));
@@ -755,7 +763,10 @@ impl Session {
return;
}
let elapsed = now.saturating_duration_since(r.admitted);
if elapsed >= Duration::from_millis(500) {
if elapsed >= Duration::from_millis(500) && !r.control.canceled() {
// Main-lane cancellation cause is independent of the worker's durable
// outcome. Do not relabel an earlier removal or termination cancel.
r.watchdog_canceled_at = Some(now);
r.control.cancel();
}
if elapsed >= Duration::from_secs(1) {
@@ -788,6 +799,12 @@ impl Session {
.map(|r| r.result)
.ok_or(Error::InvalidRange)
}
/// Original worker error for native diagnostics, independent of guest deadline reason.
/// Pending and RAM-only requests have no worker error. Retention follows query/release.
pub fn request_storage_error(&self, context: u64, id: u64) -> Result<Option<Error>> {
self.query(context, id)?;
Ok(self.request.as_ref().filter(|r| r.id == id).and_then(|r| r.worker_error))
}
pub fn release_request(&mut self, context: u64, id: u64) -> Result<()> {
self.query(context, id)?;
if self.deferred.is_some() {