I have a COM object, which I connect to, and I should recieve an event, which would confirm that connection is established. I write code and test it in F# interactive, and for some reason it wouldn't catch COM events when I use Async.RunSynchronously.
/// This class wraps COM event into F# Async-compatible event
type EikonWatcher(eikon : EikonDesktopDataAPI) =
let changed = new Event<_>()
do eikon.add_OnStatusChanged (fun e -> changed.Trigger true)
member self.StatusChanged = changed.Publish
/// My method
let ``will that connection work?`` () =
let eikon = EikonDesktopDataAPIClass() :> EikonDesktopDataAPI // create COM object
let a = async {
let watcher = EikonWatcher eikon // wrap it
eikon.Initialize() |> ignore // send connection request
let! result = Async.AwaitEvent watcher.StatusChanged // waiting event
printfn "%A" result // printing result
return result
}
// I use either first or second line of code, not both of them
Async.Start (Async.Ignore a) // does not hang, result prints
Async.RunSynchronously (Async.Ignore) a // hangs!!!
/// Running
``will that connection work?`` ()
At the same time, code works perfectly well with RunSynchronously when I insert it into console app.
What should I do so that to prevent that nasty behavior?
The code we write under within a single Thread (as in STA) feels like it is made of independant pieces each having their own life, but this is actually a fallacy : everything is mediated under a common event loop, which "linearizes" the various calls.
So everything we do, unless explicitely spoecified otherwise, is essentially single threaded and you can not wait for yourself without creating a deadlock.
When you specify Async.Start it does start a new, independant computation which runs on its own, a "thread".
Whereas When you call runsynchronously, it awaits on the same 'thread'.
Now if the event you are waiting, which feels like an independant thing, is actually 'linearized' by the same event loop, you are actually waiting for yourself, hence the deadlock.
Something useful if you want to wait "asynchronously", (aka wait for an event, but not actually block and leave the opportunity for any other task to perform work) you can use the following code within your async block :
async {
....
let! token = myAsyncTask |> Async.StartChild
let! result = token
....
}
Related
I've spent hours combing through documentation and tutorials, but can't figure out how to use ReactiveX to poll an external resource, or anything for that matter, every at an interval. Below is some code I wrote to get information from a REST API at an interval.
open System
open System.Reactive.Linq
module MyObservable =
let getResources =
async {
use client = new HttpClient()
let! response = client.GetStringAsync("http://localhost:8080/items") |> Async.AwaitTask
return response
} |> Async.StartAsTask
let getObservable (interval: TimeSpan) =
let f () = getResources.Result
Observable.Interval(interval)
|> Observable.map(fun _ -> f ())
To test this out, I tried subscribing to the Observable and waiting five seconds. It does receive something every second for five seconds, but the getResources is only called the first time and then the result is just used at each interval. How can I modify this to make the REST call at each interval instead of just the result of the first call being used over and over again?
let mutable res = Seq.empty
getObservable (new TimeSpan(0,0,1))
|> Observable.subscribe(fun (x: seq<string>) -> res <- res |> Seq.append x;)
|> ignore
Threading.Thread.Sleep(5000)
Don't use a Task. Tasks are what we call "hot", meaning that if you have a value of type Task in your hand, it means that the task is already running, and there is nothing you can do about it. In particular, this means you cannot restart it, or start a second instance of it. Once a Task is created, it's too late.
In your particular case it means that getResources is not "a way to start a task", but just "a task". Already started, already running.
If you want to start a new task every time, you have two alternatives:
First (the worse alternative), you could make getResources a function rather than a value, which you can do by giving it a parameter:
let getResources () =
async { ...
And then call it with that parameter:
let f () = getResources().Result
This will run the getResources function afresh every time you call f(), which will create a new Task every time and start it.
Second (a better option), don't use a Task at all. You're creating a perfectly good async computation and then turning it into a Task only to block on getting its result. Why? You can block on an async's result just as well!
let getResources = async { ... }
let getObservable interval =
let f () = getResources |> Async.RunSynchronously
...
This works, even though getResources is not a function, because asyncs, unlike Tasks, are what we call "cold". This means that, if you have an async in your hand, it doesn't mean that it's already running. async, unlike Task, represents not an "already running" computation, but rather "a way to start a computation". A corollary is that you can start it multiple times from the same async value.
One way to start it is via Async.RunSynchronously as I'm doing in my example above. This is not the best way, because it blocks the current thread until the computation is done, but it's equivalent to what you were doing with accessing the Task.Result property, which also blocks until the Task is done.
I'm struggling to figure out how to write a Task.ContinueWith within a Async computation expression.
Ultimately, I want to process a cancel without relying on the Cancel Exception. As a result, I thought I could use Task.ContinueWith.
However, I'm in-over-my-head on trying to write this.
Any suggestions?
let rec receiveXmlMessage connection (cancellation:CancellationToken) queue =
async {
use receiveCommand = new SqlCommand(receiveQuery, connection, CommandTimeout = 0)
let result = receiveCommand.ExecuteNonQueryAsync(cancellation)
result.ContinueWith(fun (someResult:Task<int>) -> CancellableResult.Cancelled // IDK... ) |> AsyncResult.fromAsync
}
Error Type mismatch. Expecting a
Task> -> 'b but given a
Async<'c> -> Async> The type 'Task>' does not match the type 'Async<'b>'
Sorry, this isn't necessarily an answer but wouldn't fit well as a comment.
Do you need to pass a cancellation token to your workflow? I think it might be better to use Async.CancellationToken to propagate the workflow's existing token:
async {
let! ct = Async.CancellationToken // gets the async workflow's current cancellation token
use receiveCommand = new SqlCommand(receiveQuery, connection, CommandTimeout = 0)
let! result = receiveCommand.ExecuteNonQueryAsync(ct) |> Async.AwaitTask
// do something with the result
}
By threading the async workflow's cancellation token through to the computation you care about, you enable cancelling the workflow to do the right thing - it can cancel the top-level workflow when transitioning from one step to another, or it can cancel the query execution if that's what's executing.
Then when you schedule your async via a method like Async.RunSynchronously or Async.Start you can pass in an existing cancellation token if you have one.
However, it's not completely clear to me if this is what you're actually trying to do, since you're using let rec but you haven't indicated any need for recursive logic.
I am trying to get the below code to work in a F# async workflow, but I am getting the error "Unexpected symbol '}' in expression". I am fairly new to both F# and async in general. What am I missing here.
let someFunction (req : HttpRequestMesssage) a b =
// code
async{
let! readToProvider =
req.Content.ReadAsMultipartAsync(provider)
|> Async.AwaitIAsyncResult
} |> Async.RunSynchronously
req.CreateResponse(HttpStatusCode.OK)
I worry that my previous answer wasn't quite what you want. What I supplied just got you through the compile error. But one thing about it, is that it does not run asynchronously. Task.Wait and Async.RunSynchronously will both block the running thread until the operation is complete.
If you want to actually be async, i.e. not blocking, you have to put the entire method, or at least the last part of it, into the async block, such that you're actually returning an async op to the caller. So the answer would be
let someFunction (req : HttpRequestMesssage) a b =
async {
let! readToProvider = (req.Content.ReadAsMultipartAsync provider) |> Async.AwaitIAsyncResult
return req.CreateResponse HttpStatusCode.OK
}
This option returns not the response, but an Async<Response>. So now the caller can decide how to run it, either blocking or truly asynchronously.
This way if you're using a web server that handles asynchronous requests, then you can simply connect this function to an endpoint (probably converting the Async to a Task at the point of connection, since most .net async web servers are written from C# perspective) and it'll run asynchronously without blocking a thread. Or if you're calling it from another async op you can do do! someFunction ... and it'll run asynchronously. But if the caller doesn't care and just wants to run synchronously, it can do someFunction ... |> Async.RunSynchronously. So you get more flexibility there. And you can always define let someFunctionSync ... = someFunction ... |> Async.RunSynchronously if that's the more common use case.
I'd recommend going this way unless you really want to enforce blocking.
You're doing it right. You're only getting the error because you're ending your async block with a let! expression. Change it to return!, or do! ... |> Async.Ignore and you'll be good.
Blocks in F# (neither workflows nor regular code blocks) should not end with let.
Of course if all you're really doing in the workflow is that one call, you don't need the workflow block at all (you never really should need to write a block for a single call). Just do
req.Content.ReadAsMultipartAsync provider
|> Async.AwaitIAsyncResult
|> Async.Ignore
|> Async.RunSynchronously
req.CreateResponse HttpStatusCode.OK
Or for that matter, just use the built in Tasks Wait, which does the same thing as Async.RunSynchronously:
(req.Content.ReadAsMultipartAsync provider).Wait()
I was pretty comfortable with how async cancellations where done in C# with the TPL, but I am a little bit confused in F#. Apparently by calling Async.CancelDefaultToken() is enough to cancel outgoing Async<'T> operations. But they are not cancelled as I expected, they just... vanishes... I cannot detect properly the cancellation and tear down the stack properly.
For example, I have this code that depends on a C# library that uses TPL:
type WebSocketListener with
member x.AsyncAcceptWebSocket = async {
let! client = Async.AwaitTask <| x.AcceptWebSocketAsync Async.DefaultCancellationToken
if(not(isNull client)) then
return Some client
else
return None
}
let rec AsyncAcceptClients(listener : WebSocketListener) =
async {
let! result = listener.AsyncAcceptWebSocket
match result with
| None -> printf "Stop accepting clients.\n"
| Some client ->
Async.Start <| AsyncAcceptMessages client
do! AsyncAcceptClients listener
}
When the CancellationToken passed to x.AcceptWebSocketAsync is cancelled, returns null, and then AsyncAcceptWebSocket method returns None. I can verify this with a breakpoint.
But, AsyncAcceptClients (the caller), never gets that None value, the method just ends, and "Stop accepting clients.\n" is never displayed on the console. If I wrap everything in a try\finally :
let rec AsyncAcceptClients(listener : WebSocketListener) =
async {
try
let! result = listener.AsyncAcceptWebSocket
match result with
| None -> printf "Stop accepting clients.\n"
| Some client ->
Async.Start <| AsyncAcceptMessages client
do! AsyncAcceptClients listener
finally
printf "This message is actually printed"
}
Then what I put in the finally gets executed when listener.AsyncAcceptWebSocket returns None, but the code I have in the match still doesn't. (Actually, it prints the message on the finally block once for each connected client, so maybe I should move to an iterative approach?)
However, if I use a custom CancellationToken rather than Async.DefaultCancellationToken, everything works as expected, and the "Stop accepting clients.\n" message is print on screen.
What is going on here?
There are two things about the question:
First, when a cancellation happens in F#, the AwaitTask does not return null, but instead, the task throws OperationCanceledException exception. So, you do not get back None value, but instead, you get an exception (and then F# also runs your finally block).
The confusing thing is that cancellation is a special kind of exception that cannot be handled in user code inside the async block - once your computation is cancelled, it cannot be un-cancelled and it will always stop (you can do cleanup in finally). You can workaround this (see this SO answer) but it might cause unexpected things.
Second, I would not use default cancellation token - that's shared by all async workflows and so it might do unexpected things. You can instead use Async.CancellationToken which gives you access to a current cancellation token (which F# automatically propagates for you - so you do not have to pass it around by hand as you do in C#).
EDIT: Clarified how F# async handles cancellation exceptions.
Knowing an RPC call to a server method that returns unit is a message passing call, I want to force the call to be asynchronous and be able to fire the next server call only after the first one has gone to the server.
Server code:
[<Rpc>]
let FirstCall value =
printfn "%s" value
async.Zero()
[<Rpc>]
let SecondCall() =
"test"
Client code:
|>! OnClick (fun _ _ -> async {
do! Server.FirstCall "test"
do Server.SecondCall() |> ignore
} |> Async.Start)
This seems to crash on the client since returning unit, replacing the server and client code to:
[<Rpc>]
let FirstCall value =
printfn "%s" value
async { return () }
let! _ = Server.FirstCall "test"
Didn't fix the problem, while the following did:
[<Rpc>]
let FirstCall value =
printfn "%s" value
async { return "" }
let! _ = Server.FirstCall "test"
Is there another way to force a message passing call to be asynchronous instead?
This is most definitely a bug. I added it here:
https://bugs.intellifactory.com/websharper/show_bug.cgi?id=468
Your approach is completely legit. Your workaround is also probably the best for now, e.g. instead of returning Async<unit> return Async<int> with a zero and ignore it.
We are busy with preparing the 2.4 release due next week and the fix will make it there. Thanks!
Also, in 2.4 we'll be dropping synchronous calls, so you will have to use Async throughout for RPC, as discussed in https://bugs.intellifactory.com/websharper/show_bug.cgi?id=467 -- primarily motivated by new targets (Android and WP7) that do not support sync AJAX.