Application crashes on azure app service with signalr alpha

[yosigolan]

setup:
server: asp.net core app version 2.0
signalr 1.00 alpha
client: Angular 4.3

symptom: when deploying the server to azure app service, server crashes with 2 connections (the entire app crashes and restarts).
i couldn't receive any exception or error from azure app service, just: application stopped unexpectedly.

when using the same setup only with the old signalR (i know its not certified), the app does not crash.
i think its related to the number of connections, because i see it happens when i am connecting the second client and playing a bit.

sorry i dont have too much details, i tried getting some logs from azure but haven't succeeded after few hours...

[davidfowl]

When you get some logs, let us know. Turn up the application logs to trace. This isn't very actionable feedback at the moment.

[yosigolan]

Do you mean the application logs inside the asp.net core app? they are currently on Information for Microsoft. the last logs i saw from the signalr was connection closed and then the application crashes and restarts (on azure). if there was any warning on error i guess i would see it (since its above information).
i will try to reproduce it tomorrow locally, maybe than i will receive something.

Yosi

[davidfowl]

Any logs you can get, crank up the default logging level to "Trace" and look at either app insights (what I recommend) or use the stdout log

[yosigolan]

Hey @davidfowl , good news i think i got it.
i reproduced it locally. it happens only on iis (on kestrel works fine). when connecting with 2 clients an exception is thrown inside signalr (i see it stops in visual studio):

{System.NullReferenceException: Object reference not set to an instance of an object.
   at Microsoft.AspNetCore.SignalR.DefaultHubLifetimeManager`1.<WriteAsync>d__11.MoveNext()
--- End of stack trace from previous location where exception was thrown ---
   at System.Runtime.ExceptionServices.ExceptionDispatchInfo.Throw()
   at System.Runtime.CompilerServices.TaskAwaiter.HandleNonSuccessAndDebuggerNotification(Task task)
   at Travideo.AppsWebService.Hubs.TripPlanSystemViewConnector.<SetTripGeneralData>d__17.MoveNext() in C:\GitRepos\Backend\AppsWebService\AppsWebService\Hubs\TripPlanSystemViewConnector.cs:line 122}

once this exception is thrown the application is dead (this what caused it to crash on azure).
in the logs i get nothing... even when loglevel for Microsoft and system is debug.
let me know if you need anything else.
Yosi

[davidfowl]

@yosigolan yes that exception looks pretty deadly 😄 . Can you put your application on a github repository somewhere so we can take a look.

[yosigolan]

sorry, its a production product, you can view it here: www.seevoov.com
i cant share the code.
but if you want i can show to you on a private session and even we can debug it.

Yosi

[davidfowl]

If you can provide any details on how the error can be reproduced, it would be great. Can you reproduce it on demand now only when running behind IIS locally?

[yosigolan]

yes, i can reproduce it on demand on my machine locally using iis express. if you want we can make a conference call and i will let you debug it (you need to connect with the source).
the details i can provide are:

• signalR core with default options
• javascript client (i am using angular 4)
• start 2 connections
• send data from the server to the clients
• after few seconds the exception is thrown

Yosi

[davidfowl]

I'd like to figure it out without a conference call (chat in slack might be easier to start with http://tattoocoder.com/asp-net-core-slack-community/) Based on the stack trace there's a null ref in at Microsoft.AspNetCore.SignalR.DefaultHubLifetimeManager1.d__11.MoveNext()which looks like it might map to theWriteAsync` call.

Some things that might help:

at Travideo.AppsWebService.Hubs.TripPlanSystemViewConnector.d__17.MoveNext() in C:\GitRepos\Backend\AppsWebService\AppsWebService\Hubs\TripPlanSystemViewConnector.cs:line 122}

What does this method do (at line 122)?

1. What transports are being used?
2. Did you turn logging on trace at the application level? Can you provide those logs?

[yosigolan]

in line 122 there is this code:
await m_Client.InvokeAsync("setTripGeneralData",generalData);
thats all the function do. m_Client is IClientProxy which was created as follows: m_Client = m_TripPlanHubContext.Clients.Client(m_ConnectionId);
it was created in the constructor using the client connection id.
i checked, the connection id is valid (few calls before the crash works fine and arrives to the client)

i am using the defaults so eventually it uses websockes.
i turned logging to trace (i am suing serilog so its debug) but i dont see any log from signalr.

[davidfowl]

it was created in the constructor using the client connection id.

Do not do that. Just get it in the method every time. The constructor is too early to store anything to do with the hub.

i turned logging to trace (i am suing serilog so its debug) but i dont see any log from signalr.

Then something is wrong. This is what your Program.cs should look like:

WebHost.CreateDefaultBuilder(args)
        .UseStartup<Startup>()
        .ConfigureLogging(logging =>
        {
             logging.SetMinimumLevel(LogLevel.Trace);
        }
        .Build();

That will set the logging to trace. There's an output window in visual studio that will show everything when running on IIS Express:

[yosigolan]

Thanks for the answer.
| Do not do that. Just get it in the method every time. The constructor is too early to store anything to do with the hub. -
are you saying that this is the cause? why does the constructor is too early? the connection was made earlier in the application lifetime. in the constructor i am injecting the context: IHubContext tripPlanHubContext
and then just to make the methods nicer i am creating the client member from the context.
you saying that this is forbidden? i should do this : var client = m_TripPlanHubContext.Clients.Client(m_ConnectionId); in every method that wants to send data to the client?

regarding the logs, when using serilog the configuration looks different.
anyway, i do see Microsoft logging (in debug) but not signalr. from signalr i am seeing 1 line (repeatedly) that look like this: Using HubProtocol '"json"'. from Microsoft.AspNetCore.SignalR.HubEndPoint. in info

Yosi

[davidfowl]

| Do not do that. Just get it in the method every time. The constructor is too early to store anything to do with the hub. -
are you saying that this is the cause? why does the constructor is too early? the connection was made earlier in the application lifetime. in the constructor i am injecting the context: IHubContext tripPlanHubContext

The constructor is too early because none of the hub properties have been set yet. You're injecting the IHubContext into the Hub itself? If yes why?

regarding the logs, when using serilog the configuration looks different.
anyway, i do see Microsoft logging (in debug) but not signalr. from signalr i am seeing 1 line (repeatedly) that look like this: Using HubProtocol '"json"'. from Microsoft.AspNetCore.SignalR.HubEndPoint. in info

Just make sure you add the ConfigureLogging code that sets the level to Trace.

The problem is I can't decipher what your application is doing. I need to see the source code or you need to debug it more so I can get more details. It would help if you wrote the most basic idomatic hub code and see if that reproduces the issue:

• Remove the IHubContext from the Hub constructor, it's redundant and not required. There's a property on the Hub that gives you access to the context.
• Use Clients.Client(...).InvokeAsync() in each method instead of using a field.

After making those changes, see if the problem can be reproduced.

[yosigolan]

Let me explain a bit, i think the naming of my classes confused you:

• i am not injecting the hub context to the hub.
• the flow of my app is to send data to the client upon a call from the client to some method in the rest controller.
• since i dident find any way to inject the hub into a controller, my hub code is empty. instead i wrote a different class with the methods that sends that data to the client (lets call it sender). i am injecting the hub context into the controller class and the controller creates the sender with the hub context and the connection id and then initiates a send method on it.
• the exception occurred when calling a method on the sender that initiated the action await m_Client.InvokeAsync("setTripGeneralData",generalData);
• is there a better way of doing the above?
• please note that using kestrel everything works perfectly. with iis it works at start and then fails after few minutes.
• also, with the old signal r i am doing the same and it works great.

Yosi

[davidfowl]

@yosigolan I see. Just do you know, I'm not trying to get you to change your code to fix things, I'm trying to narrow down the problem. There is a bug here, I'm just trying to figure out how I can reproduce the problem so I can fix it. I don't want to force you to debug the SignalR source code yourself so the goal is to narrow the things down so we can figure out what the issue is.

the exception occurred when calling a method on the sender that initiated the action await m_Client.InvokeAsync("setTripGeneralData",generalData);

We need to figure out what's causing the null ref. If you never store m_Client and always resolve the client before calling the method, does it work?

also, with the old signal r i am doing the same and it works great.

Yes, this is an alpha release and there are LOTS of bugs and changes that still need to be made. If you're running it in production that's on you 😄, there is no support for this until it ships. We appreciate the feedback regardless.