Revision control

Copy as Markdown

Other Tools

#![warn(rust_2018_idioms)]
use std::rc::Rc;
use std::sync::Arc;
use tokio_util::task;
/// Simple test of running a !Send future via spawn_pinned
#[tokio::test]
async fn can_spawn_not_send_future() {
let pool = task::LocalPoolHandle::new(1);
let output = pool
.spawn_pinned(|| {
// Rc is !Send + !Sync
let local_data = Rc::new("test");
// This future holds an Rc, so it is !Send
async move { local_data.to_string() }
})
.await
.unwrap();
assert_eq!(output, "test");
}
/// Dropping the join handle still lets the task execute
#[test]
fn can_drop_future_and_still_get_output() {
let pool = task::LocalPoolHandle::new(1);
let (sender, receiver) = std::sync::mpsc::channel();
let _ = pool.spawn_pinned(move || {
// Rc is !Send + !Sync
let local_data = Rc::new("test");
// This future holds an Rc, so it is !Send
async move {
let _ = sender.send(local_data.to_string());
}
});
assert_eq!(receiver.recv(), Ok("test".to_string()));
}
#[test]
#[should_panic(expected = "assertion failed: pool_size > 0")]
fn cannot_create_zero_sized_pool() {
let _pool = task::LocalPoolHandle::new(0);
}
/// We should be able to spawn multiple futures onto the pool at the same time.
#[tokio::test]
async fn can_spawn_multiple_futures() {
let pool = task::LocalPoolHandle::new(2);
let join_handle1 = pool.spawn_pinned(|| {
let local_data = Rc::new("test1");
async move { local_data.to_string() }
});
let join_handle2 = pool.spawn_pinned(|| {
let local_data = Rc::new("test2");
async move { local_data.to_string() }
});
assert_eq!(join_handle1.await.unwrap(), "test1");
assert_eq!(join_handle2.await.unwrap(), "test2");
}
/// A panic in the spawned task causes the join handle to return an error.
/// But, you can continue to spawn tasks.
#[tokio::test]
async fn task_panic_propagates() {
let pool = task::LocalPoolHandle::new(1);
let join_handle = pool.spawn_pinned(|| async {
panic!("Test panic");
});
let result = join_handle.await;
assert!(result.is_err());
let error = result.unwrap_err();
assert!(error.is_panic());
let panic_str: &str = *error.into_panic().downcast().unwrap();
assert_eq!(panic_str, "Test panic");
// Trying again with a "safe" task still works
let join_handle = pool.spawn_pinned(|| async { "test" });
let result = join_handle.await;
assert!(result.is_ok());
assert_eq!(result.unwrap(), "test");
}
/// A panic during task creation causes the join handle to return an error.
/// But, you can continue to spawn tasks.
#[tokio::test]
async fn callback_panic_does_not_kill_worker() {
let pool = task::LocalPoolHandle::new(1);
let join_handle = pool.spawn_pinned(|| {
panic!("Test panic");
#[allow(unreachable_code)]
async {}
});
let result = join_handle.await;
assert!(result.is_err());
let error = result.unwrap_err();
assert!(error.is_panic());
let panic_str: &str = *error.into_panic().downcast().unwrap();
assert_eq!(panic_str, "Test panic");
// Trying again with a "safe" callback works
let join_handle = pool.spawn_pinned(|| async { "test" });
let result = join_handle.await;
assert!(result.is_ok());
assert_eq!(result.unwrap(), "test");
}
/// Canceling the task via the returned join handle cancels the spawned task
/// (which has a different, internal join handle).
#[tokio::test]
async fn task_cancellation_propagates() {
let pool = task::LocalPoolHandle::new(1);
let notify_dropped = Arc::new(());
let weak_notify_dropped = Arc::downgrade(&notify_dropped);
let (start_sender, start_receiver) = tokio::sync::oneshot::channel();
let (drop_sender, drop_receiver) = tokio::sync::oneshot::channel::<()>();
let join_handle = pool.spawn_pinned(|| async move {
let _drop_sender = drop_sender;
// Move the Arc into the task
let _notify_dropped = notify_dropped;
let _ = start_sender.send(());
// Keep the task running until it gets aborted
futures::future::pending::<()>().await;
});
// Wait for the task to start
let _ = start_receiver.await;
join_handle.abort();
// Wait for the inner task to abort, dropping the sender.
// The top level join handle aborts quicker than the inner task (the abort
// needs to propagate and get processed on the worker thread), so we can't
// just await the top level join handle.
let _ = drop_receiver.await;
// Check that the Arc has been dropped. This verifies that the inner task
// was canceled as well.
assert!(weak_notify_dropped.upgrade().is_none());
}
/// Tasks should be given to the least burdened worker. When spawning two tasks
/// on a pool with two empty workers the tasks should be spawned on separate
/// workers.
#[tokio::test]
async fn tasks_are_balanced() {
let pool = task::LocalPoolHandle::new(2);
// Spawn a task so one thread has a task count of 1
let (start_sender1, start_receiver1) = tokio::sync::oneshot::channel();
let (end_sender1, end_receiver1) = tokio::sync::oneshot::channel();
let join_handle1 = pool.spawn_pinned(|| async move {
let _ = start_sender1.send(());
let _ = end_receiver1.await;
std::thread::current().id()
});
// Wait for the first task to start up
let _ = start_receiver1.await;
// This task should be spawned on the other thread
let (start_sender2, start_receiver2) = tokio::sync::oneshot::channel();
let join_handle2 = pool.spawn_pinned(|| async move {
let _ = start_sender2.send(());
std::thread::current().id()
});
// Wait for the second task to start up
let _ = start_receiver2.await;
// Allow the first task to end
let _ = end_sender1.send(());
let thread_id1 = join_handle1.await.unwrap();
let thread_id2 = join_handle2.await.unwrap();
// Since the first task was active when the second task spawned, they should
// be on separate workers/threads.
assert_ne!(thread_id1, thread_id2);
}