akkadotnet/akka.net

Port of Akka actors for .NET

rsinohara

@rsinohara

I just needed to read the code after a few days off it :)

Aaron Stannard

@Aaronontheweb

and the ignore messages is synchronous

so it was just coincidence that it worked at all - whereas when you ran the full suite you eventually hit a case where you lost the race

rsinohara

@rsinohara

That's why this test is the only one that had this issue, as others I have ignores use TestActors

Maciek Misztal

@mmisztal1980

@Aaronontheweb After a 3rd Become, I'm trying to do a Self.Tell(new Msg()) but that Msg is never received, even though the Receive<Msg>(...) is set in that Become(...) handler

rsinohara

@rsinohara

btw congrats Aaron, besides the nice work on akka.net, you're an amazing speaker

rsinohara

@rsinohara

Hmm, I cannot call methods on TestActor.UnderlyingActor... What is the recommended way to make a test actor become something, without having to go through all the state changes?

(Or is the only way to create a message handler just for this?)

Bartosz Sypytkowski

@Horusiath

@rsinohara think about this a little differently. Become essentially builds a state machine. C# await also builds a state machine:

DoSomething();
await DoSomethingAsync(); // in this continuation the behavior switches
DoSomethingElse();

How do you want to check if DoSomethingElse() has been called without calling first two methods before?

rsinohara

@rsinohara

@Horusiath I'm not sure what you mean. Suppose my actor goes S1->S2->S3. I want to test S3 behavior, without having to interact with the actor to make it follow nominal path from S1 to S3.

I just want to set that state, then test its behavior, not test the state change flow.

Bartosz Sypytkowski

@Horusiath

what I'm trying to say, is that's impossible unless you already have some start point, which allows you to omit are intermediate steps

just like you cannot execute only the last line of the method without going through every statement before

rsinohara

@rsinohara

Well I can create a message handler that will directly call the specific Become... but I'll be creating a handler just to accommodate tests - which I'd avoid unless there is no way

Bartosz Sypytkowski

@Horusiath

when writing tests for akka, I usually write the whole initialization code (if it's repeatable, make a separate method from it)

rsinohara

@rsinohara

If Underlying actor exposed methods, that would be as simples as calling a method that Becomes what I need for the test in question.

A hack would be to do it through a property setter/getter (ewww)

Bartosz Sypytkowski

@Horusiath

if I make a shortcut to test only the specific behavior without steps that I need to make on the way, then I'm not truly testing that behavior. Simply because the tested scenario is artificial and doesn't occur in reality

heh, testing only the behavior is something that's going to happen, when akka-typed goes out :P

rsinohara

@rsinohara

I get what you mean now

Maybe I should rethink my testing startegy

Is there any material on actor (unit) testing?

I'd love a book about that.

Bartosz Sypytkowski

@Horusiath

the biggest reference is akka repository - around 4000 examples ;P

rsinohara

@rsinohara

Aren't they too low level?

Bartosz Sypytkowski

@Horusiath

it depends, most of them are actually integration tests - using the testkit itself

rsinohara

@rsinohara

Interesting then. I'll take a look. Thanks :)

Jeff

@jpierson

Is there a way to create a ConsistentHashingGroup router where the hash id determined by the sender as opposed to something in each message?

On second thought, perhaps it would be better to have some other actor responsible for providing another actor for direct communication (routee) based on a consistent hash of the sender without using a router. Is this an existing pattern already that has a name? I'm running into this where I have interfaces between external services where I wanted ordered retry logic but don't want to force everything to be processed in serial, just in proper order based on the sender to keep the sender/receiver order guarantees.

Aaron Stannard

@Aaronontheweb

I think I'll be able to add the ability to log anything on the Scheduler that wasn't executed at the time it was shutdown

as part of some of these changes

Michael

@michaelsg

anyone have familiarity with the F# api? I have one nagging question about the implementation of pipe operator |!>.

Bartosz Sypytkowski

@Horusiath

@michaelsg what's up?

Michael

@michaelsg

Had a question about the implementation of the pipeTo (|!>) operator - how it's supposed to work. I gathered that it was supposed to be non-blocking but it uses Async.StartWithContinuations which starts the async on the current thread. I think it only work right for the things like in the examples of its use - that are already-running .net things wrapped in asyncs.

if I create an f# async - not just something wrapping a task or async pattern, it blocks.

Bartosz Sypytkowski

@Horusiath

what are you creating F# async for?

Michael

@michaelsg

to do background work :-) sorry. not trying to be cheeky. I don't know what you mean.

Bartosz Sypytkowski

@Horusiath

I'm just asking for the use case ;) If you're working with something that already starts an async, it should be good to go.

Michael

@michaelsg

so that's the thing. Async (the kind you make with async {}) aren't running when they are created. and starting one does not return an async - it returns unit.

Bartosz Sypytkowski

@Horusiath

yep, it may be a problem ;) Probably something to be fixed - can you set a github issue on that?

Michael

@michaelsg

async { do! Async.Sleep 50000.; return AMessageToSend("hello") } |!> mbx.Self

you agree that shouldn't block?

Corneliu

@corneliutusnea

guys, any guidance on deploying Akka cluster on Azure?

Maciek Misztal

@mmisztal1980

depends on what you need?

Corneliu

@corneliutusnea

anything to make my life easy-ier :smile: e.g. how do deploy lighthouse-es, experiences, dealing with azure vm virtual ips

Maciek Misztal

@mmisztal1980

right, so far I've used lighthouse deployed on to static IaaS VMs to act as a "fixed cluster anchor" with static public IPs. when setting up the config, you need to take both the public and the private IPs into account :) As for nodes joining the cluster - those can be pretty much anything... old Azure Service Roles, WebJobs, Functions?, ServiceFabric processes, Containers - take your pick :)

Corneliu

@corneliutusnea

thanks, that's already good advice

I'm thinking of doing everything on VMs just to keep it simpler

Yin Zhang

@melcloud

Hi all. I would like to have a bounded stash to cache incoming messages when the database is not available. Since the BoundedStash is not implemented, can I use the ICircularBuffer from Helios to achieve this instead of writing my own circular buffer? Or there is any other ways to do it?

Jeff

@jpierson

@mmisztal1980 From what could find Azure WebJobs don't support any protocol other than HTTP specifically and only over the typical ports. If you understand this differently I'd be interested.

Where communities thrive

People

Repo info

Activity