#tokio-docs

1 messages · Page 1 of 1 (latest)

umbral thistle
#

Me and a friend use tokio already for a long time and know to "never block in an async context/runtime". So when we talk with newcomers to Rust and async/await, we tell them so too. But we don't just want to tell "don't do it", we'd like to say "look at this article, site. This shows what happens if you do".

We were unable to find an up-to-date and precise article that does exactly that, for easy sharing. I remember things like the current thread might get taken completely off the runtime and any pending tasks might get stuck with it, never to recover. And similar things like that, maybe from some older recordings of Jon Gjengset (pre tokio 1.0).

So would just love to know if there is a nice article out there that explores what would actually happen (or can happen) if you execute blocking calls on the tokio runtime. Best if it even even explores whether it might be okay to block for shorter times, and what happens a long blocking call completes (is the runtime able to recover and what not).

If anybody has a good article on it, please let me know.

errant star
umbral thistle
#

Thank you. I know and read that article. But I feel it's not fully covering what bad things might happen.

This can be a major issue because it means that other tasks on the same runtime will stop running until the thread is no longer being blocked.

Which means technically, all async tasks on the thread are halted until the blocking call finishes. But the following sample after that statement shows it in such a simple example, that it doesn't seem to break anything, just running things in serial instead of in parallel.

That is obviously not good, but not too bad either, is it? (as the execution of async tasks continues after the blocking call)

The default Tokio runtime spawns one thread per CPU core, and you will typically have around 8 CPU cores. This is enough that you can miss the issue when testing locally, but sufficiently few that you will very quickly run out of threads when running the code for real.

So I guess the worst case is, if you have all async threads blocking on a long running function call, it halts all async tasks. But again, wouldn't it recover after the blocking calls finish?

errant star
#

However, in general, if something is blocking but then stops blocking, then the runtime will recover.

#

You don't end up with problems forever unless something blocks forever.

umbral thistle
#

Thank you, that's a really interesting issue. Also, learned that the tokio runtime has a single "IO owner thread". Didn't know that.
Thank yo so much for your time. Very much appreciated 🙇

errant star
umbral thistle
#

So in the issue, all tasks halt, because we were unlucky that the blocking thread was the one with the IO driver. Hence, all other threads don't wake up until the blocking call finishes, so the thread with the IO driver can notify the other threads again?

errant star
umbral thistle
#

Oh, because only one thread is responsible for polling the IO driver, and that one is busy with a blocking call 💡

errant star
umbral thistle
#

Definitely really bad for a web server to block, as the whole request processing might halt.

In case of CLI apps, it seems to not be too bad if a few blocking calls happen here and there? I mean, it always depends, but the typical CLI app, where it runs a single command and shuts down, or even some dashboard CLI might just freeze for a moment (unless you have extremely long blocking calls of course)

errant star
umbral thistle
#

For file I/O bound tasks yes, but cli tools like link checkers or load testers, there's surely a benefit in using async.

Or, in our case:. A newcomer to Rust, that grabbed reqwest as it's in the top results for http clients, without understanding the whole async vs blocking in Rust yet. For his simple cli app a blocking http client would have been more than sufficient, but he didn't know.

He mixed that with blocking reads from stdin, so we told him to better not do that. As it turns out now, in his case it wouldn't really have mattered, but probably good to steer towards the mindset of never do blocking calls in async in general 😅

plucky bloom
#

async is also oftentimes a much better paradigm for CLIs that have to manage a bunch of child processes…

#

not necessarily because of performance, but because you might want things like cancellation/timeouts, processing output from child processes in the order they complete, etc

static nexus
alpine comet
#

accept?

static nexus
ancient summit
#

@errant star @static nexus I know you both have wonderful blogs that you link to in here pretty often to great effect, would it maybe make sense to mirror some of that content on tokio.rs?

#

Maybe not in blog form but translated with credit into a tutorial/guide page?

static nexus
#

Sure, I'd be happy to. I haven't actually written anything about Tokio directly, mostly about tonic and axum, not sure if that belongs on tokio.rs?

ancient summit
#

Yeah thats a good point...I wonder where we should put http stuff @desert scaffold any ideas? like I guess hyper.rs or tokio.rs its confusing a bit lol

errant star
#

I have pondered doing so before, but haven't taken the time to do it. I don't mind doing so.

errant star
#

Axum in particular is certainly a Tokio project — it's under the Tokio github org.

ancient summit
#

yeah

static nexus
#

Ok cool. I think I'm gonna do a second extensions post with more examples taken from discussions here. Maybe after I'll clean it up and open a PR to add on Tokio.

rugged drum
#

Hi, is the docs regarding tokio::time::Sleep not being Unpin outdated?

errant star
rugged drum
#

pin_project seems to generate Unpin impl

errant star
#

It looks bad in the documentation, but the where bound is false so it doesn't apply.

rugged drum
#

okay, it's just confusing on first glance.

reef eagle
#

Hi - I am a new Tokio user.
I am reading the tutorial, and I feel that I am more confused than I should be. It could just be that I am new to Tokio, but it feels like the tutorial is mixing up the introduction of new topics and the code that we need to write in the tutorial. It is not clear when or where code needs to be created.

Am I the only new user that feels like this, or is there room for improvement?

full depot
#

I agree, And actually I think don't need mini redis.

reef eagle
full depot
reef eagle
# full depot yes~

The more I read, the more confusing everything is. For example, on this page https://tokio.rs/tokio/tutorial/framing there is a heading called "The Buf trait", but there is nothing in the text that shows an implementation of the Buf trait or any other trait.

When adding the read_frame method on the Connection struct, the code is missing the impl Connection statement.

Rust is a really hard language to learn, and async/await is also tough to grasp. It is important that we keep the documentation clean.

stray merlin
#

Coincidentally I am reading that same page for the very first time. I don't find it particularly confusing. I think the doc is good enough as long as I can understand what a Buftrait is in principle and that I can reconstruct my own implementation by reading the doc. Seeing the code for Buf trait would then be a confirmation/verification of my understanding. My 2 cents.

stray merlin
#

On https://tokio.rs/tokio/tutorial/async, the futures crate is used under the Mini Tokio section before it was later introduced (ie added as a dependency futures = "0.3") under the Updating Mini Tokio section. Is this intentional?

delicate flame
#

hey guys, I have some question about the tokio::spawn, the docs say tokio::spawn(async{}) will create a task , and tokio::select! run on the same task, then how to explain this example?

# 
use tokio::sync::oneshot;

#[tokio::main]
async fn main() {
    let (tx1, rx1) = oneshot::channel();
    let (tx2, rx2) = oneshot::channel();

    tokio::spawn(async {
        let _ = tx1.send("one");
    });

    tokio::spawn(async {
        let _ = tx2.send("two");
    });

    tokio::select! {
        val = rx1 => {
            println!("rx1 completed first with {:?}", val);
        }
        val = rx2 => {
            println!("rx2 completed first with {:?}", val);
        }
    }
}
#

doesn't it run on two taks?

rose kite
plucky bloom
delicate flame
#

@plucky bloom so you mean in main task, you can spawn multi tasks, select! will collect all spawned tasks in the main task?

#

if so, I think this part needs a more precise description, not task and task , it's very confusing

humble vapor
#

select! isn't "collecting" anything, no

#

you can only run async code in Tokio if it's running as a "task": this is just the equivalent of a regular OS thread.

for the above code, the #[tokio::main] macro is doing something that you can conceptually think of as:

tokio::spawn(async {
    let (tx1, rx1) = oneshot::channel();
    let (tx2, rx2) = oneshot::channel();
    ...
});
#

so that's the first task, as Eliza mentioned: the "main" task. then there's the two tasks being spawned via tokio::spawn. all three tasks are separate.

#

select! is a macro for trying to await multiple futures and doing something with whichever future, out of all of the futures being select!-ed, completes first. since that code is in in main(), it runs on the "main" task.

wintry beacon
#

Good morning, all.. I have a quick question regarding a snippet included in the tutorial documentation. Is this the proper channel for that type of question?

errant star
wintry beacon
#

Thank you, @errant star ! I have resolved what I was planning to ask. This is tutorial is so well thought out, learning lots a long the way

subtle marten
#

white list?

alpine comet
#

?

lyric oasis
errant star
lyric oasis
#

@errant starHow about adding a search bar to tokio.rs?

errant star
lyric oasis
#

Many open source projects use the algoria search api for their static documentation sites, witch is free for open source projects

#

I'm not sure if tokio.rs needs a search feat, if you think it does but don't know what way I can try to investigate@errant star

errant star
#

If there's a good way of adding it, then I don't mind

gleaming tree
#

It used to have a search feature provided by a third party service

lament karma
#

Hi. Is there a new good book covering Axum in detail? Paid books are ok too

lethal berry
stoic ruin
clear sable
#

How does one reconnect a tokio-tungstenite websocket connection? Say it's already connected but I need it to be reconnected, gracefully or not?

humble pagoda
hushed root
#

I'm a little bit confused by https://docs.rs/tokio/latest/tokio/sync/mpsc/index.html - which says "The bounded variant has a limit [...] if this limit is reached, trying to send another message will wait until a message is received from the channel." - it sounds like Sender will block when the channel is full.

But then at the end of the next paragraph, it says "If the bounded channel is at capacity, the send is rejected and the task will be notified when additional capacity is available. In other words, the channel provides backpressure." - which sounds like it doesn't block.

Is it just that I'm not thinking in terms of Tokio "asynchronous tasks", and that what the second paragraph means with "the send is rejected" is that the "asynchronous task" is blocked and later notified to retry the send? Such that the calling code does indeed block, this "reject & later notify" being just the implementation detail of what's happening behind the scenes for the blocked send?

#

Hmm. I guess this became more a "noob" question than a "docs" question. 🙂

errant star
#

anyway, the send method will block until it is able to send the message

gleaming tree
#

I will need to figure out how to increase max permutations though

hushed root
gleaming tree
hushed root
night schooner
#

I have a question about the following (module doc of tokio::fs):

Tasks run by worker threads should not block, as this could delay servicing reactor events. Portable filesystem operations are blocking, however. This module offers adapters which use a blocking annotation to inform the runtime that a blocking operation is required. When necessary, this allows the runtime to convert the current thread from a worker to a backup thread, where blocking is acceptable.

#

Is there a way to tell/know which operations will block (e. g. spawn a backup thread) on which platforms?

errant star
night schooner
#

They all use blocking fs internally? 🤐 I had thought they would use epoll on Linux and Kqueue on Macos which are, IIRC, block-free

errant star
#

Epoll and kqueue don't support normal files.

#

They just claim that the file is always ready even if the operation would block.

night schooner
#

I see, thanks for the info

alpine comet
#

you can use kqueue with aio to do some fs operations on bsd, but that has other restrictions and difficulties

#

(mac is bsd)

night schooner
#

So on platforms without io_uring, blocking file IO is as good as you're gonna get?

errant star
#

Tokio doesn't use io_uring for tokio::fs, even if it is available.

night schooner
alpine comet
errant star
stoic ruin
#

So https://docs.rs/tokio-macros/latest/tokio_macros/attr.test.html the builder and runtime links are broken. I tried to fix them via changing to [`tokio::runtime::Runtime`] following the docs from tokio-util but that only seems to work if the item is a dependency and doesn't work for dev-dependencies. Also, doing a [target.'cfg(doc)'.dependencies] and adding tokio as a dependency for docs won't work because of circular dependency issues fingergun

errant star
stoic ruin
#

hmm seems a bit of a UX pain when I google for the tokio main macro docs and the macro crate docs page is the first result sunglass_cry

#

wonder if there's a solution that works for both ThinkingFast

supple acorn
half stump
#

hello，firends。can i use egui in tokio？

fierce timber
# half stump hello，firends。can i use egui in tokio？

If you mean using tokio within an egui application, yes it's possible. There's a few ways to go about it but one of them is to pass a Runtime to your application, spawn tasks for async work and communicate back to the main thread through channels.

half stump
minor hatch
#

Sorry I didn’t realize how long ago this was posted

fiery wagon
#

Hello!
I want to work on this issue #6263 ~
I wrote a quick document describing my thought process before I start implementing anything. I also prepared some questions in advance to better understand the scope of work
It would be great if someone could review, and perhaps answer those questions~

half stump
gentle leaf
#

I've been trying to understand codecs and how to use them. I started at https://docs.rs/tokio-util/latest/tokio_util/codec/index.html, which got me thinking that what I need is either a LinesCodec or LengthDelimitedCodec (I'm just playing around so doesn't really matter which one) and a FramedRead to use it with a stream. I followed through to https://docs.rs/tokio-util/latest/tokio_util/codec/length_delimited/index.html which shows me how to configure a decoder, and https://docs.rs/tokio-util/latest/tokio_util/codec/struct.FramedRead.html which tells me what FramedRead has.

But I still haven't really seen any concrete example that ties everything together and shows me exactly how to read a frame from a stream. I'm gathering that I actually need tokio_stream::StreamExt so I can do .next() on my FramedRead. This does not seem to be explained anywhere in the docs I've seen so far, and I only figured it out looking at https://users.rust-lang.org/t/understanding-framed-in-tokio/42449/2

Am I missing something obvious? This seems like such a fantastic feature to have, but nothing in the docs (at least whatever I've seen so far) seem to really help me to become successful.

errant star
craggy epoch
#

The front page of Bytes docs (https://docs.rs/bytes/latest/bytes/) have Buf and BufMut points to a 404. But local cargo doc generates the correct link

Perhaps re-triggering the docs ci job fix this?

errant star
thorny sierra
#

Hi - the following async move in the Tokio docs (https://tokio.rs/tokio/tutorial/spawning) seems redundant:

use tokio::net::TcpListener;

#[tokio::main]
async fn main() {
    let listener = TcpListener::bind("127.0.0.1:6379").await.unwrap();

    loop {
        let (socket, _) = listener.accept().await.unwrap();
        // A new task is spawned for each inbound socket. The socket is
        // moved to the new task and processed there.
        tokio::spawn(async move {  // <----------------------------- THIS ONE
            process(socket).await;
        });
    }
}

The process function takes a TcpStream, not a reference to one (and returns a future), so the following works just fine:

tokio::spawn(process(socket));

Am I missing something?

errant star
warm spindle
#

Is there any good doc explaining how Tokio channels are implemented?

thorny light
#

howdi2e, just following the tut. bit confused ny https://tokio.rs/tokio/tutorial/spawning under subheading "Tasks" it seems the example is incorrect. or i am misunderstanding it. The text talks about "joinhandle" but its not in the example

errant star
thorny light
#

Thanks @errant star

burnt horizon
#

see what Rust Rover is showing here, I think, the people maintaining tokio have to change the documentation comments

image.png
#

Changing that really had a huge performance boost

shell carbon
#

Has there been any thoughts about formalizing cancellation safety signalling? I know Alice did a lot of work to document all of Tokio's APIs and their respective cancellation safety posture. I was thinking it'd would be interesting to have this be machine checkable. Something like

#[cancel_safe]
async fn foo() -> Bar;

struct Foo {}

#[cancel_safe]
impl Future for Foo {}

with an accompanying linter(ideally just part of cargo clippy), alternatively some external tool

#

The derive macro wouldn't need to do anything, it just exists as a marker

errant star
#

Also, when select! isn't used in a loop, you often don't need cancel safety. E.g., if the server is shutting down, it doesn't matter that data on some connection is lost when you cancel it.

shell carbon
#

I was thinking that the cancel_safe macro doesn't have to do anything at runtime, it would only serve as a marker for a linter tool

#

Also, when select! isn't used in a loop, you often don't need cancel safety. E.g., if the server is shutting down, it doesn't matter that data on some connection is lost when you cancel it.
True true

errant star
#

That's an interesting idea.

shell carbon
# 
#[cancel_safe(required = false)]
tokio::select! {} 

# Check the source tree based on these annotations, assume any future that isn't explicitly marked is cancel safe
cargo ensure-cancel-safe --strategy=assume-cancel-safe
shell carbon
#

Feels to me like it would need to be a stateful multipass lint(one pass to collect all types marked with cancel_safe and then another to check select! and other constructs where cancel safety matters)

shell carbon
#

I asked in the clippy Zulip channel about feasibility

shell carbon
#

I thought some more, and it would be pretty tricky for any async fn futures were the callsite is far from the usage

shell carbon
#

Messed around some more with this, ended up with a type level solution

gleaming tree
#

@shell carbon very nice, this has the potential of being super helpful. I will carve out some time to look more into this

gleaming tree
gleaming tree
#

I would really like to have some sort of cancel safety related tooling / lints by default for Tokio apps. The biggest challenge I see is how to manage backwards compatibility with select! Emitting a warning could work...

errant star
gleaming tree
#

100% it cannot be outlawed

#

But, I think saying you should only pass cancel-safe futures to select is fair

#

"I don't care about the end state because it is isolated and dropped" is "cancel-safe"

errant star
#

ah, hm

#

that means that cancel-safe futures may be composed of stuff that is not cancel-safe, but I guess that's fine

gleaming tree
#

Right, you are just saying "this is cancel safe because I say so"

#

@errant star another thing that I realized, cancel safety and panic safety are very similar (same thing?)

#

if a future isn't cancel safe, it probably isn't panic safe either

#

it just is that cancellation happens a lot more as part of the regular execution of code

errant star
#

I guess both are cases where code abrubtle gets stopped somewhere.

#

Cancel safety just only gets stopped at awaits, and panic safety is when you only get stopped at things that can panic.

gleaming tree
#

right, panic safety would be stronger than cancellation safety

errant star
#

either way, some sort of macro could be cool

gleaming tree
#

I agree, it also is worth thinking if it would be possible to minimize having to annotate fns

#

it would probably be too much friction if users have to annotate most of their fns

errant star
#

well, they only need to annotate what goes into a select

gleaming tree
#

@errant star In theory, it would cascade out. A method is cancel safe only if the methods it calls are cancel safe

#

like, you want to be able to catch an issue with introduced non-cancel safe code in the call stack

errant star
# 
fn my_func(&mut self) {
    self.num_active_calls += 1;
    my_cancel_safe_fn().await;
    self.num_active_calls -= 1;
}
gleaming tree
#

@errant star sorry, I didn't mean it was a sufficient condition, just necessary

#

if all .await calls in a method are cancel safe, then the method might be cancel safe. If one .await call is not cancel safe, then the method is not cancel safe (forgetting the case where a user doesn't care if the method is technically cancel safe)

shell carbon
shell carbon
gleaming tree
#

@shell carbon yeah... but if it is a lint, it might be possible to just default to warnings

#

anyway, it is worth exploring more imo

shell carbon
#

I'm not sure about my initial idea of using the derive macro only as a lint tool. I think it would break down across crate boundaries etc. The nice thing about how I implemented it is that it becomes a type level concern

shell carbon
errant star
#

problem is, if you require require_cancel_safe to apply it, nobody will use it

gleaming tree
#

Also, if you just annotate the outer method, what will happen is someone will come by later and accidentally make the call stack not cancel safe

#

So, the ideal situation is, there is some recursive check through the call stack

errant star
shell carbon
#

That said it should be possible to do recursively too, more work at compile time though. Also might require a lot of annotations unless it can be determined automatically in some instances

gleaming tree
#

Yeah, in my mind, the difference for real world devs using tokio, is manually implementing Send, Sync, and AssertUnwindSafe basically never happens, and (at least Send, Sync) it is unsafe

#

I'm coming from the POV of teams of devs that are having this issue

#

where you can write safe async rust code that is not cancel safe

#

and it is a pretty common issue in practice (based on my observations)

shell carbon
#

I'm hoping, but haven't verified, that the wrapping future I used can always be optimised out by the compiler. If it can't that's another problem

shell carbon
gleaming tree
#

@shell carbon I think the "easy" work around would be to have it conditionally happen based on a cfg flag

#

and then there can be a cargo check-cancel-safe

#

or something like that

#

or, just have it done w/ debug assertions, etc...

shell carbon
gleaming tree
#

w/ debug assertions is probably the best route

#

I think the most obvious recursive implementation of a #[cancel_safe] macro would require all .await calls to be be calling methods annotated w/ #[cancel_safe] as well or wrap the .await w/ cancel_safe(foo()).await (or something)

#

but that would be pretty noisy...

#

the question is, can we decide if an async method is cancel safe via static analysis. If so, another option would be to use a clippy lint type thing I think?

#

Anyway, those are my thoughts. I would love to see the question of "cancel safety" lints explored more

shell carbon
#

I chatted a bit with the clippy folks on Zulip and it's possible to look for the mere presence of an attribute without needing to change the resulting types as I did. clippy::has_significant_drop is an example of this kind of intert attribute

#

Also, it's possible to check across crate boundaries

#

Since clippy runs after macro expansion something is needed to identify the callsites where futures needs to be checked for cancel safety e.g. select! after expansion

errant star
shell carbon
craggy epoch
shell carbon
shell carbon
craggy epoch
errant star
#

The main issue with the trait approach is implementing it for futures returned by async fn.

craggy epoch
#

Right. That requires changes from rustc.

#

Well, I think a clippy lint is beneficial for others as well.

#

Feel free to ping me in zulip/here if you need any help with the clippy part, if you decide to go that route @shell carbon

shell carbon
#

The marker trait would be really nice and my POC does support it, but it comes at the cost of having to add a wrapping future to all async fns which is a bit crap for compile times and potentially runtime

umbral spruce
errant star
umbral spruce
errant star
#

It's fine ... if you are careful.

#

at least in the case of dashmap

#

I don't know whether the other crates have lock guards that are Send

umbral spruce
#

We sure did try to make it work, and we could not. My theory (from practictioner perspective), is that some crates just are not a good match with Tokio, and are best to be avoided.

errant star
#

curious to hear your perspective

umbral spruce
#

I love this post, and I think we had once an email exchange where I asked if a subset of the content or link to the post could be added to Tokio docs, but did not get around to suggest those changes.

errant star
#

Ah, yes, now that you mention it, I remember that you emailed me about something along those lines.

umbral spruce
#

Given the sad experience with dashmap (and more importantly, given that others have bad experiences as well), I went for the easy solution, which is to just remove it from the official Tokio docs..

errant star
#

Just to confirm, the issue you ran into is that it's too easy to end up blocking the thread / deadlocking?

umbral spruce
#

Yeah, I did, I would rather not have anyone ever have to go to the deadlock party

#

The issue we had was with an actor based prototype, where we managed the objects in the memory with dashmap, and even with heavy weeding and best effort so to say, we could not have a solution that would not deadlock even under the lightest load.

#

Glad to have survived that, to tell the story, but yeah, I don't wish that to anyone. I also find it interesting, that even knowing about these issues, we just could not make it work.

errant star
#

Do you use spawn_local?

umbral spruce
#

I can't remember, we buried that prototype somewhere. Could take a look later though. 🙏

errant star
#

I just looked through the two other libraries, and it looks like only flurry has non-Send guards.

umbral spruce
#

In any case, it might be helpful to have lists of crates that are known to play well, and crates that could be risky, and the reasons why. Something digestible for the lazy or busy developer who might be building with tokio for the first time.

errant star
#

How about this: Since I have you now, I merge the other PR first, and then you rebase yours on top to remove dashmap and leapfrog from the list. That way, we merge both PRs, and the list ends up with a recommendation of flurry, which doesn't use deadlock-prone guards.

umbral spruce
errant star
#

ok, I merged it

umbral spruce
#

Will fix my PR by EOD

umbral spruce
errant star
#

I guess it's fine if you link to the draft for now.

umbral spruce
#

Looking at this page here https://tokio.rs/tokio/tutorial/shared-state, we might consider the following changes:

  • Moving "Tasks, threads and contention" after the part on restructuring code, so that restructuring code is right after the compiler error
  • Add a warning on Send MutexGuards in the compiler error section
  • Add a link to the draft post in that section
  • Mention Send MutexGuards again in the mutex crate section.
  • Add link to the draft post there as well
#

These could be quick fixes, that would help guide more developers on the right path.

#

I can't help asking though, if the mini redis example itself, should not be refactored to use the wrapped mutex pattern as well? That might help to drive the point home. After all, I'm sure quite a few folks just copy whatever code the tutorial presents.

#

Also I have some additional questions on this page...

#

A common error is to unconditionally use `tokio::sync::Mutex` from within async code.

  • Why is this an error? Is it because it is slower, and should be more or less the last option?
#
  • Is it so, that holding a Send MutexGuard across an await will deadlock only in the multithreaded runtime?
#

(And no, we did not use the wrapped mutex pattern on our ill-faithed actor prototype, that used dashmap. We probably should have.)

#

I'm also curious about when to go for wrapped mutex, and when might actor model be suitable. DB (unless in the process' memory) means IO, so should the DB be rather managed by an actor, than a wrapped mutex?

#

Ah, of course, an external database is not really shared mutable state from the application's perspective, in the same sense. So it would not count as an IO resource, typically, I guess.

spring crypt
# umbral spruce - Is it so, that holding a Send MutexGuard across an await will deadlock only in...

It can deadlock in both current and multi thread runtimes. It’s actually easier to get a std mutex to deadlock in a current thread runtime because the deadlock is caused by the task that holds the lock being scheduled on the same worker as the task which is trying to obtain the lock (for worst case in a way that it will never get stolen - which means either a current thread runtime or that the task is in the lifo slot or maybe the other workers are all idle).

spring crypt
umbral spruce
spring crypt
umbral spruce
umbral spruce
spring crypt
umbral spruce
#

In general, what do you think about sanitizing all the mutex examples, so that they would all have wrapped mutexes, instead of passing Arc<Mutex<_>> around? I believe this would make it less likely that tokio users end up with problematic code. (I'm volunteering of course, if the timeline is flexible.)

umbral spruce
#

@errant star @spring crypt Any comments/feedback?

spring crypt
umbral spruce
umbral spruce
errant star
spring crypt
ocean wing
#

Hello Guys
Hopping this is the right place to ask
I'm reading the tutorial (https://tokio.rs/tokio/tutorial/channels) and using the oneshot
example I got the following question:
based on the image's code
on the t2 spaws
tx2 is a cloned refence from tx but using oneshot channel kind there is no way to have multiple senders.
Am I missing something?
(Maybe It can be possible using Arc)

image.png
errant star
#

so you can have more than one

ocean wing
#

Yes, the screnshot is from the refactor section once is using oneshoot

#

that code is the example code

#

on the website

#

btw thanks for your response Alice

errant star
#

I don't really understand the question

ocean wing
#

Sorry
The situation is:
I was confused with the example reading again the article I got my error
Thanks for your help and sorry if my words were rude (My english is not perfect :S )

swift bronze
#

hi

#

I am dracula

#

I can't read the doc because its too shiny

image.png
#

we want dark theme please

errant star
swift bronze
vernal cape
brisk timber
errant star
brisk timber
#

Dark mode

errant star
#

Does anyone want to take the time to update the website to the new next.js? Otherwise I'm just going to disable dependabot again. I don't really think staying up to date brings any real value, so I don't care to spend the time myself.

brisk timber
#

yeah - I'll take a look. Do you have any constraints that aren't obvious on this? Or just make it simple to merge and pass CI.

errant star
#
  1. Pass CI.
  2. Don't change how the website looks.
brisk timber
#

👍

errant star
#

Thanks!

brisk timber
#

https://github.com/tokio-rs/website/pull/790 is all the npm packages except next.js and bulma as a preperation for the next.js upgrade. I'd suggest that it's worthwhile merging and then doing the other two on top of that. I didn't see any obvious breaks. in testing.

errant star
brisk timber
#

Thanks

brisk timber
#

Bulma is still a bit of a pain as 1.0 broke a bunch of CSS variables which are defined in the site's scss file, so I'm punting that one for now. I'd recommend this would need someone that enjoys spelunking through css / scss / etc. to upgrade it.

#

That said, bulma 1.0 has dark mode theme support built-in, so this negates the need for the dark mode PR.

errant star
#

You can just swap bulma@0.9.4/css/bulma.min.css with bulma@1.0.0/css/bulma.min.css and everything will work. Things will look slightly different, but they will still work.
https://bulma.io/documentation/start/migrating-to-v1/

This is crazy to me. We had a designer spend a lot of time on putting together the current look. Changing the design because of this kind of thing is bullshit.

brisk timber
#

If I was going to make some PRs to change the docs, is there a preference for larger PRs with multiple commits moving things around, or small single focus PRs. The latter can get annoying to work on especially when there's dependent changes between each commit, so I'd prefer the former in this case if that's ok with the powers that be.

alpine comet
#

"github, i am once again asking you to give us stacked diffs"

alpine comet
brisk timber
#

Even "github, please allow me to manually choose which commits are shown in the diff view" would suffice there.

brisk timber
frigid hollow
#

Hi guys, I'm a beginner learning tokio tutorial. I got an issue in the image example, which can be found here: https://tokio.rs/tokio/tutorial/async#a-few-loose-ends. I noticed that the main function returns before the delay future completes. I don't know if it's intensional because it's just a demonstration. And there is still race condition even though we wrap the waker in a mutex, as it might be updated after the timer thread returns. This could result in mis-scheduling an expired future. This example is really confusing, and I still don't know how to handle this scenario 😩

Screenshot_2024-12-23_at_20.48.13.png
errant star
frigid hollow
errant star
frigid hollow
ocean wing
#

Hello community
I'll be talking about multithreading and Async in Rust
It will be for a community meetup, just sharing what I know, just for fun
Do you know if there are diagrams/architectures related tokio?
(I'm looking for visual representations of the runtime/workers/global queue/etc)
Thanks for your help 🙂

void spindle
void spindle
# ocean wing Hello community I'll be talking about multithreading and Async in Rust It will ...

You didn't ask, but I highly recommend including concepts from alice's blog on blocking: https://ryhl.io/blog/async-what-is-blocking/

That is IME one of the harder concepts for people to grasp when reasoning about how/when to use tokio. Essentially, that a thread != a task, that threads pre-empt each other, whereas tasks execute sequentially on a given worker thread and therefore can have very unpleasant effects if one task takes a long time.

I tend to introduce this by discussing the difference between concurrency and parallelism.

ocean wing
#

Thanks for your help 🙂

#

Really appreciate it

#

The meetup will be in Spanish
I'll share the link to the streaming just in case anyone want to attend 🙂

#

The schedule is:
January 17, 2025
7:00 pm UTC - 6 (Mexico central time)

void spindle
#

Nice, probably toss it in #tokio-users as well for visibility

swift bronze
#

error: failed to select a version for the requirement tokio-macros = "~2.5.0"
candidate versions found which didn't match: 2.4.0, 2.3.0, 2.2.0, ...
location searched: crates.io index
required by package tokio v1.43.0
... which satisfies dependency tokio = "^1.43.0" (locked to 1.43.0) of package Omama v0.1.0 (/run/media/allawiii/projects/zed/Rust/Omama)
if you are looking for the prerelease package it needs to be specified explicitly
tokio-macros = { version = "0.2.0-alpha.6" }

#

[dependencies]

tokio = { version = "1.43.0", features = ["full"] }

errant star
swift bronze
#

cargo clean?

errant star
#

rm Cargo.lock

#

I don't think cargo clean deletes it

swift bronze
#

yes cargo clean doesnot delete it

#

thanks every thing works properly

cinder schooner
#

hi guys. What are the good reading to understand how tokio works internally? like does it creates threads/processes, when, how? is there something like event-loop?

gleaming tree
#

The source code probably 🙂 there are some blog posts that cover it though, if you look a few years back.

acoustic elk
cinder schooner
acoustic elk
# cinder schooner what is pollster and mio?

Two crates that are good to familiarise yourself with. Pollster is the smallest functional async runtime, that is reasonable to read through. Mio is an async-io abstraction layer that tokio uses

brisk timber
placid cove
#

is it true that u cannot change 1- the number of threads in rust tokio during the runtime; 2-tokio involves non premptive threads

alpine comet
#

this is for work on the docs of tokio

swift bronze
#

so, tokio::io::copy ,
when reqwesting large file from web ,
it doesnt download the whole file in the memory first!!?
but instead it stream each arrived chunk directly to the harddisk ?

errant star
#

Yes

swift bronze
#

thank you

# 
async fn write_to_disk(mut tool: Response, full_path: PathBuf) -> OResult<PathBuf> {
    let mut file = File::create(&full_path).await?;
    let mut writer = BufWriter::new(file);
    copy(&mut tool.bytes(), &mut writer);

    Ok(full_path)
}
#

tool.chunk() returns bytes::Bytes object

errant star
#

Calling tool.chunk() reads the entire thing into memory, so don't do that.

#

or ... no

#

anyway

#

hold on

#

your screenshot shows a function called tool.bytes(), not tool.chunk()

#

which is it?

swift bronze
#

rustc: {async fn body of reqwest::Response::bytes()} cannot be unpinned
within impl futures_util::Future<Output = std::result::Result<bytes::bytes::Bytes, reqwest::Error>>, the trait std::marker::Unpin is not implemented for {async fn body of reqwest::Response::bytes()}
consider using the pin! macro
consider using Box::pin if you need to access the pinned value outside of the current scope

#

returns future of Result

errant star
#
  1. If you have a Bytes object, then you don't use tokio::io::copy. You just call writer.write_all(&my_bytes_obj).
  2. If you want to stream the data, then you don't use .bytes().
#

What you want is chunk, not bytes.

#

taking its example

# 
let mut res = reqwest::get("https://hyper.rs").await?;

while let Some(chunk) = res.chunk().await? {
    println!("Chunk: {chunk:?}");
}
#

and updating it:

swift bronze
#

I want , every arrived peice of data to be directly stored on disk , no need to hold them in the memory

errant star
# 
let mut res = reqwest::get("https://hyper.rs").await?;

while let Some(chunk) = res.chunk().await? {
    writer.write_all(&chunk).await?;
}
#

this will do what you want

swift bronze
# 
async fn write_to_disk(mut tool: Response, full_path: PathBuf) -> OResult<PathBuf> {
    let mut file = File::create(&full_path).await?;

    while let Some(chunk) = tool.chunk().await? {
        file.write_all(&chunk).await?;
    }

    Ok(full_path)
}

is this the correct way?

errant star
#

Call file.flush().await? after the loop. Otherwise it looks good

swift bronze
#

why I need to flush it ?

#

what flush means here ?

errant star
#

It means to wait for writing to finish.

swift bronze
#

ooh

#

thats great point

#

flush functionality also exists in stdout()

#

now I understood what it means , thanks

#

so when the last chunk reaches its destination the file buffer sends something like acknowledgment to let the program know that all the data were written successfully alright ?
or something like that .

errant star
#

Tokio's file performs the writing on a background thread. Calling file.write just starts the write without waiting for it.

swift bronze
#

so no need to call flush function?

errant star
#

yes need to call flush function

swift bronze
#

understood

#

thank you

swift bronze
#

I am facing a nightmare here :

image.png
#

rustc: type annotations needed
cannot satisfy `_: std::future::Future

errant star
frosty cave
#

<@&1207366540299735151> Got 1365779859837816892 with a Steam scam

image.png
#

Not the right channel, just the one I had clicked on, sorry

lunar canopy
#

I finished The Rust Programming Language ("the book") last weekend, this weekend I'm starting the Tokio tutorial here: https://tokio.rs/tokio/tutorial

I'm a mid-career software engineer (C, Obj-C, Swift), a Rust beginner, and a Tokio beginner. Are #1324762623006347264 and #tokio-users the best place for beginner questions if they arise?

compact wagon
brisk timber
spiral spade
#

because Semaphore::acquire* are async fns, I realized they can't reserve your slot in the fairness queue until the result is polled at least once. is this something tokio has (intentionally?) committed to (vs leaving open the possibility of replacing the async fns with -> impl Future fns that reserve the queue before constructing the future) and should the answer to that question be explained in the docs?

acoustic elk
rough spindle
#

Could someone please clarify if it is true that Tokio can handle tasks preemptively, and if so, how this wouldn’t potentially cause deadlocks in some cases?

errant star
rough spindle
# errant star Tasks are suspended only at `.await` points.

Sorry for the inconvenience, but do you happen to have a link to somewhere I can read in detail about how Tokio handles tasks preemptively? The .await thing makes a lot of sense to me, but I understand that it’s the task that yields cooperatively

errant star
rough spindle
#

Gracias

rough spindle
#

Hey, dumb question here! I've been experimenting a bit and it seems like the main task (the one from block_on or #[tokio::main]) isn't affected by work stealing and doesn’t migrate between threads. Is that right?

acoustic elk
#

But importantly, the block_on thread (if using a multi_threaded runtime) only executes that one future - it does not participate in the runtime at all

rough spindle
acoustic elk
#

If you use a current_thread runtime, then block_on does also run other tasks

rough spindle
mint otter
#

Ou

rough spindle
#

Hi, I know this place is for talking about Tokio, but maybe someone could give me a quick answer. Could someone tell me if NLL depends on variance to coerce lifetimes?

acoustic elk
rough spindle
# acoustic elk The borrow checker is necessarily aware of variance. However, lifetimes are tech...

If I didn't explain it well, it's actually the exact opposite—because Rust cannot determine its exact lifetime at a point, it extends it to its true duration. It happened to me when taking a &mut inside a type that made it invariant, and I can't reference it again even though I didn't use the type that borrowed it anymore. Not sure if I explained myself clearly or just spoke a tongue twister

azure jolt
#

Alguém tem o link da Tokio

tacit loom
#

Po

tacit loom
flint ledge
#

Rola

rough spindle
#

Hi, could you help me with this question:
It’s about what happens when the blocking thread pool is completely in use. Let’s assume we are on Linux. For example, let’s take Notify without going into too much detail and see if I’m correct: when calling notified().await, Tokio likely spawns one of its blocking pool threads and puts it to sleep using a futex. Then, once someone wakes it up, it reschedules the task that triggered it.
If this is the case, then a non-blocking worker would actually block if it needs to use a thread from the blocking pool and that pool is completely occupied.
Maybe Notify can be handled without spawning any threads, but it’s just to illustrate the idea.

errant star
#

There are no primitives in Tokio that work like the one you mentioned. Tokio only uses spawn_blocking in cases where the operation is something that will exit on its own eventually without outside input.

#

Such as reading a file.

#

Notify works by storing a Waker for the task that is waiting. You can think of a Waker as a closure that enqueues the async task to the queue of tasks that need to be polled again.

#

This way, Notify can function without taking up a blocking thread.

rough spindle
dull sail
#

Oh, that's how you alert end-users what is required to enable a particular module, etc?

gleaming tree
#

yep

dull sail
#

That is so brilliant.

#

Thank you! I don't know why I couldn't seem to find a straight answer to it. I appreciate it. I'm taking inspiration from tokio with regard to properly documenting my library. Now it makes sense why you have all the ...cfg... derive macros or what not for feature gates

#

Errrr, what I mean is now it all makes sense. I know that those macros are not the same thing. Just saying, I understand how everything is related now. Thanks again!

errant star
#

@dull sail Note that this also requires some stuff in your Cargo.toml and lib.rs if you want to use it.

dull sail
#

Right. I felt like I was just missing a single piece to a puzzle. I'll make sure that I take a closer look and really understand everything.

#

You are the best 🙂

steel wedge
#

yes

errant star
#

they are the same

bright otter
#

it waits a certain amount of time

#

it’s not very complicated

#

iirc, and i probably don’t, tokio just has a thread pool of native threads

gleaming tree
#

Sleep is “green thread” it is very low overhead

errant star
steel wedge
#

I guess it's just missing tokio::spawn ferrisThink

lethal berry
#

yeah I suppose, but like its just an example. Its not meant to be real

lethal berry
steel wedge
lethal berry
minor hatch
#

does tokio use linux's epoll somewhere?

errant star
minor hatch
#

task::spawn?

errant star
minor hatch
#

what advantage one gets by using tokio::sync::Mutex over std::sync::Mutex?

gleaming tree
minor hatch
#

how expensive is tokio::spawn? is it okay to use 1000 tasks at once?

sinful condor
#

tokio::spawn is like, one heap allocation and a couple atomic ops

errant star
minor hatch
errant star
minor hatch
#

understandable thx

solar river
#

I've written a draft of a small note warning about a footgun in fs, thought it might be best to get some opinions here first before I send the PR:

//! **Warning**: These adapters may create a large number of temporary tasks,
//! especially when reading large files. When performing a lot of operations
//! in one batch, it may be significantly faster to use [`spawn_blocking`]
//! directly:
//!
//! `\``
//! # #[tokio::main]
//! # async fn main() {
//! use std::io::{BufReader, BufRead};
//! let line_count = tokio::task::spawn_blocking(move || -> Result<usize, std::io::Error> {
//!     let file = std::fs::File::open("/usr/share/dict/words")?;
//!     let line_count = BufReader::new(file).lines().count();
//!     Ok(line_count)
//! }).await?;
//! # }
#

gah discord's text formatting

#

I'm not hugely pleased about hardcoding a filepath that might not be there on all platforms, but I'm not quite sure what else to put there that would work.

errant star
errant star
solar river
solar river
errant star
#

Meh. It's fine. If you want, you can add a hidden main function 🤷‍♀️

solar river
#

mh, not sure if it's worth making them work on all platforms, if compile test is fine I'll just do that.

errant star
#

Fine with me.

cosmic orbit
#

tx.send("sending from first handle").await

_ = tx2.send("sending from second handle").await;

#

would fix the compiler's warnning

#

I apologize for the presentation..