fix: make hosted downloads controllable and failures diagnosable
Sync source to CNB / Sync Git ref (push) Skipped
Axolotl desktop CI / guardrails (push) Canceled after 0s
Axolotl desktop CI / desktop (macos-latest) (push) Canceled after 0s
Axolotl desktop CI / desktop (ubuntu-latest) (push) Canceled after 0s
Axolotl desktop CI / desktop (windows-latest) (push) Canceled after 0s
Axolotl desktop CI / website (push) Canceled after 0s
Repository checks / typos (push) Canceled after 0s
Repository checks / tombi (push) Canceled after 0s
Rust checks / shear (push) Canceled after 0s
Sync source to CNB / Sync Git ref (push) Skipped
Axolotl desktop CI / guardrails (push) Canceled after 0s
Axolotl desktop CI / desktop (macos-latest) (push) Canceled after 0s
Axolotl desktop CI / desktop (ubuntu-latest) (push) Canceled after 0s
Axolotl desktop CI / desktop (windows-latest) (push) Canceled after 0s
Axolotl desktop CI / website (push) Canceled after 0s
Repository checks / typos (push) Canceled after 0s
Repository checks / tombi (push) Canceled after 0s
Rust checks / shear (push) Canceled after 0s
This commit is contained in:
1 parent
1838afce56
commit
bf95d715a4
17 files changed
+848
-72
No files matched your search
@@ -0,0 +1,329 @@
|
||||
use futures::{Stream, StreamExt};
|
||||
use serde::{Deserialize, Serialize};
|
||||
use std::{
|
||||
future::Future,
|
||||
sync::{Arc, LazyLock, Mutex},
|
||||
};
|
||||
use tokio::sync::watch;
|
||||
|
||||
#[derive(
|
||||
Clone, Copy, Debug, Default, Deserialize, Serialize, Hash, PartialEq, Eq,
|
||||
)]
|
||||
#[serde(rename_all = "snake_case")]
|
||||
pub enum HostedTaskState {
|
||||
#[default]
|
||||
Running,
|
||||
Paused,
|
||||
Canceling,
|
||||
Canceled,
|
||||
Failed,
|
||||
Finalizing,
|
||||
}
|
||||
|
||||
static CONTROLS: LazyLock<dashmap::DashMap<String, Arc<Control>>> =
|
||||
LazyLock::new(dashmap::DashMap::new);
|
||||
|
||||
pub(super) struct Control {
|
||||
instance_id: String,
|
||||
state: Mutex<HostedTaskState>,
|
||||
changes: watch::Sender<HostedTaskState>,
|
||||
}
|
||||
|
||||
fn stopped() -> crate::Error {
|
||||
super::invalid("已停止整合包下载;已下载的缓存会保留,可重试继续")
|
||||
}
|
||||
|
||||
impl Control {
|
||||
fn new(instance_id: &str) -> Self {
|
||||
Self {
|
||||
instance_id: instance_id.to_owned(),
|
||||
state: Mutex::new(HostedTaskState::Running),
|
||||
changes: watch::channel(HostedTaskState::Running).0,
|
||||
}
|
||||
}
|
||||
|
||||
fn state(&self) -> HostedTaskState {
|
||||
*self.state.lock().unwrap_or_else(|error| error.into_inner())
|
||||
}
|
||||
|
||||
fn transition(
|
||||
&self,
|
||||
allowed: &[HostedTaskState],
|
||||
next: HostedTaskState,
|
||||
) -> crate::Result<()> {
|
||||
let mut state =
|
||||
self.state.lock().unwrap_or_else(|error| error.into_inner());
|
||||
if !allowed.contains(&state) {
|
||||
return Err(super::invalid(
|
||||
"任务状态已变化或正在应用安装文件,当前操作不可用",
|
||||
));
|
||||
}
|
||||
*state = next;
|
||||
// Publish under the transition lock, so a late pause cannot overwrite finalization.
|
||||
crate::event::emit::set_hosted_loading_state(&self.instance_id, next);
|
||||
self.changes.send_replace(next);
|
||||
Ok(())
|
||||
}
|
||||
|
||||
fn pause(&self) -> crate::Result<()> {
|
||||
self.transition(&[HostedTaskState::Running], HostedTaskState::Paused)
|
||||
}
|
||||
|
||||
fn resume(&self) -> crate::Result<()> {
|
||||
self.transition(&[HostedTaskState::Paused], HostedTaskState::Running)
|
||||
}
|
||||
|
||||
fn cancel(&self) -> crate::Result<()> {
|
||||
self.transition(
|
||||
&[HostedTaskState::Running, HostedTaskState::Paused],
|
||||
HostedTaskState::Canceling,
|
||||
)
|
||||
}
|
||||
|
||||
pub(super) fn begin_finalizing(&self) -> crate::Result<()> {
|
||||
self.transition(
|
||||
&[HostedTaskState::Running],
|
||||
HostedTaskState::Finalizing,
|
||||
)
|
||||
}
|
||||
|
||||
/// The operation may be dropped only before begin_finalizing. Restarting
|
||||
/// re-plans from verified cache objects and the downloader's partial files.
|
||||
pub(super) async fn run<T, F, Fut>(
|
||||
&self,
|
||||
mut operation: F,
|
||||
) -> crate::Result<T>
|
||||
where
|
||||
F: FnMut() -> Fut,
|
||||
Fut: Future<Output = crate::Result<T>>,
|
||||
{
|
||||
let mut changes = self.changes.subscribe();
|
||||
loop {
|
||||
let state = *changes.borrow_and_update();
|
||||
match state {
|
||||
HostedTaskState::Paused => {
|
||||
changes.changed().await.map_err(|_| stopped())?;
|
||||
continue;
|
||||
}
|
||||
HostedTaskState::Canceling => return Err(stopped()),
|
||||
_ => {}
|
||||
}
|
||||
let work = operation();
|
||||
tokio::pin!(work);
|
||||
loop {
|
||||
tokio::select! {
|
||||
biased;
|
||||
changed = changes.changed() => {
|
||||
changed.map_err(|_| stopped())?;
|
||||
match self.state() {
|
||||
HostedTaskState::Paused => break,
|
||||
HostedTaskState::Canceling => return Err(stopped()),
|
||||
_ => {},
|
||||
}
|
||||
}
|
||||
result = &mut work => {
|
||||
match self.state() {
|
||||
HostedTaskState::Paused => break,
|
||||
HostedTaskState::Canceling => return Err(stopped()),
|
||||
_ => return result,
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
pub(super) struct Registration {
|
||||
instance_id: String,
|
||||
pub(super) control: Arc<Control>,
|
||||
}
|
||||
|
||||
impl Registration {
|
||||
pub(super) fn new(instance_id: &str) -> crate::Result<Self> {
|
||||
use dashmap::mapref::entry::Entry;
|
||||
let control = Arc::new(Control::new(instance_id));
|
||||
match CONTROLS.entry(instance_id.to_owned()) {
|
||||
Entry::Occupied(_) => {
|
||||
return Err(super::invalid("此实例已有整合包下载任务"));
|
||||
}
|
||||
Entry::Vacant(entry) => {
|
||||
entry.insert(control.clone());
|
||||
}
|
||||
}
|
||||
Ok(Self {
|
||||
instance_id: instance_id.to_owned(),
|
||||
control,
|
||||
})
|
||||
}
|
||||
}
|
||||
|
||||
impl Drop for Registration {
|
||||
fn drop(&mut self) {
|
||||
CONTROLS.remove_if(&self.instance_id, |_, value| {
|
||||
Arc::ptr_eq(value, &self.control)
|
||||
});
|
||||
}
|
||||
}
|
||||
|
||||
fn lookup(instance_id: &str) -> crate::Result<Arc<Control>> {
|
||||
CONTROLS
|
||||
.get(instance_id)
|
||||
.map(|entry| entry.value().clone())
|
||||
.ok_or_else(|| super::invalid("整合包下载任务已结束,请重试"))
|
||||
}
|
||||
|
||||
pub async fn pause(instance_id: &str) -> crate::Result<()> {
|
||||
lookup(instance_id)?.pause()
|
||||
}
|
||||
pub async fn resume(instance_id: &str) -> crate::Result<()> {
|
||||
lookup(instance_id)?.resume()
|
||||
}
|
||||
pub async fn cancel(instance_id: &str) -> crate::Result<()> {
|
||||
lookup(instance_id)?.cancel()
|
||||
}
|
||||
|
||||
/// Stop the batch on its first error so stalled peers cannot hide failure.
|
||||
pub(super) async fn collect_downloads<S>(downloads: S) -> crate::Result<u64>
|
||||
where
|
||||
S: Stream<Item = crate::Result<u64>>,
|
||||
{
|
||||
tokio::pin!(downloads);
|
||||
let mut total = 0;
|
||||
while let Some(result) = downloads.next().await {
|
||||
total += result?;
|
||||
}
|
||||
Ok(total)
|
||||
}
|
||||
|
||||
#[cfg(test)]
|
||||
mod tests {
|
||||
use super::*;
|
||||
use futures::FutureExt;
|
||||
use std::sync::{
|
||||
Arc,
|
||||
atomic::{AtomicUsize, Ordering},
|
||||
};
|
||||
use tokio::sync::{Notify, Semaphore};
|
||||
|
||||
struct TransferGuard(Arc<AtomicUsize>, Arc<Notify>);
|
||||
|
||||
impl Drop for TransferGuard {
|
||||
fn drop(&mut self) {
|
||||
self.0.fetch_add(1, Ordering::SeqCst);
|
||||
self.1.notify_one();
|
||||
}
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn pause_drops_live_transfers_and_resume_restarts_before_stop() {
|
||||
let control = Arc::new(Control::new("pause-resume-test"));
|
||||
let starts = Arc::new(Semaphore::new(0));
|
||||
let drops = Arc::new(AtomicUsize::new(0));
|
||||
let dropped = Arc::new(Notify::new());
|
||||
let task = {
|
||||
let control = control.clone();
|
||||
let starts = starts.clone();
|
||||
let drops = drops.clone();
|
||||
let dropped = dropped.clone();
|
||||
tokio::spawn(async move {
|
||||
control
|
||||
.run(|| {
|
||||
let guard =
|
||||
TransferGuard(drops.clone(), dropped.clone());
|
||||
starts.add_permits(1);
|
||||
async move {
|
||||
let _guard = guard;
|
||||
std::future::pending::<crate::Result<()>>().await
|
||||
}
|
||||
})
|
||||
.await
|
||||
})
|
||||
};
|
||||
starts.acquire().await.unwrap().forget();
|
||||
control.pause().unwrap();
|
||||
dropped.notified().await;
|
||||
assert_eq!(drops.load(Ordering::SeqCst), 1);
|
||||
assert_eq!(control.state(), HostedTaskState::Paused);
|
||||
control.resume().unwrap();
|
||||
starts.acquire().await.unwrap().forget();
|
||||
control.cancel().unwrap();
|
||||
let result = task.await.unwrap();
|
||||
assert!(result.unwrap_err().to_string().contains("停止"));
|
||||
assert_eq!(drops.load(Ordering::SeqCst), 2);
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn stopping_an_already_paused_task_never_starts_a_transfer() {
|
||||
let control = Arc::new(Control::new("paused-stop-test"));
|
||||
control.pause().unwrap();
|
||||
let task = {
|
||||
let control = control.clone();
|
||||
tokio::spawn(async move {
|
||||
control
|
||||
.run(|| async {
|
||||
panic!("paused work was polled");
|
||||
Ok::<(), crate::Error>(())
|
||||
})
|
||||
.await
|
||||
})
|
||||
};
|
||||
control.cancel().unwrap();
|
||||
assert!(task.await.unwrap().is_err());
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn finalization_rejects_pause_and_stop() {
|
||||
let control = Control::new("finalization-test");
|
||||
control.begin_finalizing().unwrap();
|
||||
assert_eq!(control.state(), HostedTaskState::Finalizing);
|
||||
assert!(control.pause().is_err());
|
||||
assert!(control.cancel().is_err());
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn pause_and_finalization_have_only_one_winner() {
|
||||
for _ in 0..32 {
|
||||
let control = Arc::new(Control::new("finalization-race-test"));
|
||||
let barrier = Arc::new(std::sync::Barrier::new(2));
|
||||
let task = {
|
||||
let control = control.clone();
|
||||
let barrier = barrier.clone();
|
||||
std::thread::spawn(move || {
|
||||
barrier.wait();
|
||||
control.pause().is_ok()
|
||||
})
|
||||
};
|
||||
barrier.wait();
|
||||
let finalized = control.begin_finalizing().is_ok();
|
||||
assert_ne!(finalized, task.join().unwrap());
|
||||
}
|
||||
}
|
||||
|
||||
#[tokio::test]
|
||||
async fn first_failure_drops_stalled_peer_downloads() {
|
||||
let drops = Arc::new(AtomicUsize::new(0));
|
||||
let guard = TransferGuard(drops.clone(), Arc::new(Notify::new()));
|
||||
let downloads = futures::stream::iter(vec![
|
||||
async move {
|
||||
let _guard = guard;
|
||||
std::future::pending::<crate::Result<u64>>().await
|
||||
}
|
||||
.boxed(),
|
||||
async { Err(super::super::invalid("failed file")) }.boxed(),
|
||||
])
|
||||
.buffer_unordered(2);
|
||||
let result = collect_downloads(downloads).await;
|
||||
assert!(result.unwrap_err().to_string().contains("failed file"));
|
||||
assert_eq!(drops.load(Ordering::SeqCst), 1);
|
||||
}
|
||||
|
||||
#[test]
|
||||
fn registration_is_unique_and_released_on_drop() {
|
||||
let id = uuid::Uuid::new_v4().to_string();
|
||||
let registration = Registration::new(&id).unwrap();
|
||||
assert!(Registration::new(&id).is_err());
|
||||
drop(registration);
|
||||
assert!(Registration::new(&id).is_ok());
|
||||
}
|
||||
}
|
||||
Reference in new issue
Block a user