*I want to move ILoadNotificationPlugin from being this “catch all” thing that controls the entire assembly to something that can be used to fire up code; effectively “OnLoad” and “OnUnload”. To dynamically control loading, we’ll probably reflect on the individual plugins looking for attributes or perhaps some sort of static load function.* I meant to do something like that probably using spring to inject method names to be invoked. Also event listening (wich is I believe a generic need) could be managed this way and benefit from some abstraction. con.pop(1) is probably the most frequently written line by every plugin developer, probably some abstraction (an event started with his thread and the fs event passed as an argument?) could make code more elegant On Fri, Sep 11, 2009 at 00:19, Michael Giagnocavo <mgg@giagnocavo.net>wrote:
Well, we have absolutely no idea what the background thread is doing. It might be critical, and the fix is trivial: put a try/catch on it. This is the model all .NET applications have. Background threads doing bad things should always take down the process.
However, that’s a good point about Load() failing. The approach taken is more or less how FreeSWITCH handles things in general now. If a module has an error, the switch just logs and goes on. I’m not really in favour of this, and suggested at least a “required” attribute in the modules.conf that would prevent the switch from loading if the module fails.
The fix is probably to create an attribute you can apply to the plugin classes that indicate what kind of failure handling you want. For the assembly, we’d add an attribute with an enumeration like:
- Default (scan, call ILoadNotificationPlugin, log errors if they occur)
- NoLoad (don’t load the assembly)
- Critical (stop the switch if there’s an exception during loading)
That’d provide the control you want for loading. We could do something similar for App/Api plugins.
I want to move ILoadNotificationPlugin from being this “catch all” thing that controls the entire assembly to something that can be used to fire up code; effectively “OnLoad” and “OnUnload”. To dynamically control loading, we’ll probably reflect on the individual plugins looking for attributes or perhaps some sort of static load function.
How’s that sound?
*From:* freeswitch-users-bounces@lists.freeswitch.org [mailto: freeswitch-users-bounces@lists.freeswitch.org] *On Behalf Of *Josh Rivers *Sent:* Thursday, September 10, 2009 12:48 PM
*To:* freeswitch-users@lists.freeswitch.org *Subject:* Re: [Freeswitch-users] Subscribing to events in managed C# / .NET
I'm only concerned with the difference in treatment.
public class CrashFreeSWITCH : ILoadNotificationPlugin { public bool Load() { ThreadPool.QueueUserWorkItem((o) => { throw new NotImplementedException(); }); return true; } }
Crashes the entire switch, terminating all calls and disconnecting from the PSTN.
public class CrashFreeSWITCH : ILoadNotificationPlugin { public bool Load() { throw new NotImplementedException(); return true; } }
Logs a message to the console and doesn't load the module, while leaving the switch operating.
In my experience, exceptions in multi-threaded code: a) happen, b) are hard to diagnose. Is the best behavior for the environment to crash, providing no diagnostic information? That's hard in development, and even harder in production. I suppose 'terminate switch on fault' should be an option, to allow the operating system to restart the whole process on fault conditions, but if that is the intended result, shouldn't any fault do the same thing? What if the billing was happening in my second code block?
Normally, I'd trap the ThreadException and UnhandledExceptions in my application, so that my code could choose the correct actions to perform should the application get into an unknown state. This can include terminating the whole application, but also logging diagnostic information, trying to save uncommitted data, and sending notifications of the failure.
Is 'crash if it's a thread, but not if it's not' good because it's the way the module works now, or is it a better design for a reason I'm not understanding?
On Wed, Sep 9, 2009 at 11:09 PM, Michael Giagnocavo <mgg@giagnocavo.net> wrote:
Well, a segfault in voicemail would do the same thing.
Suppose your plugin runs a thread that does something important, like billing or so on. That thread fails – do you really want it to go on?
Anyways, the solution is simple enough, handle your exceptions J. Every plugin can decide what it wants to do here.
-Michael
*From:* freeswitch-users-bounces@lists.freeswitch.org [mailto: freeswitch-users-bounces@lists.freeswitch.org] *On Behalf Of *Josh Rivers *Sent:* Wednesday, September 09, 2009 10:41 PM
*To:* freeswitch-users@lists.freeswitch.org *Subject:* Re: [Freeswitch-users] Subscribing to events in managed C# / .NET
The question is whether the CLR should take down the whole phone server due to an unhandled exception...definitely the CLR should terminate...but shouldn't it just log the exception to the console, not crash the core?
On Wed, Sep 9, 2009 at 6:30 PM, Michael Giagnocavo <mgg@giagnocavo.net> wrote:
That’s by design. If a thread fails, and there’s no handler, then the application could be in a corrupted state, so the CLR takes down the process.
I think there is a .NET 1.0 compat switch you can enable in the config if you like exceptions to be silently ignored J.
-Michael
*From:* freeswitch-users-bounces@lists.freeswitch.org [mailto: freeswitch-users-bounces@lists.freeswitch.org] *On Behalf Of *Josh Rivers *Sent:* Wednesday, September 09, 2009 6:39 PM
*To:* freeswitch-users@lists.freeswitch.org *Subject:* Re: [Freeswitch-users] Subscribing to events in managed C# / .NET
I have a new thought on the crashes...I'm able to crash FreeSWITCH any time I like, just by having an exception in a thread.
public class CrashFreeSWITCH : ILoadNotificationPlugin
{
public bool Load()
{
ThreadPool.QueueUserWorkItem((o) => { throw new NotImplementedException(); });
return true;
}
}
Perhaps Application.ThreadException or AppDomain.UnhandledException need to be trapped?
On Wed, Sep 9, 2009 at 4:51 PM, Michael Giagnocavo <mgg@giagnocavo.net> wrote:
Looks like the event object goes straight to pinvokes, so a null result just crashes?
If it’s null, you should get a NullReferenceException. The C# compiler should callvirt the property getter and that’ll do a null check. If that isn’t happening, that’d be an interesting optimization somewhere along the line.
-Michael
*From:* freeswitch-users-bounces@lists.freeswitch.org [mailto: freeswitch-users-bounces@lists.freeswitch.org] *On Behalf Of *Josh Rivers *Sent:* Wednesday, September 09, 2009 3:01 PM
*To:* freeswitch-users@lists.freeswitch.org
*Subject:* Re: [Freeswitch-users] Subscribing to events in managed C# / .NET
A new discovery:
public bool Load()
{
ThreadPool.QueueUserWorkItem((o) =>
{
Log.WriteLine(LogLevel.Notice, "Thread Starting. ");
EventConsumer con = new EventConsumer("all", "");
while (true)
{
Event ev = con.pop(0);
if (ev == null) continue;
Log.WriteLine(LogLevel.Notice, "Event: " + ev.serialized_string);
}
});
return true;
}
Does not crash. (Adding the null check prevents crash.) The backgrounded loop runs fine. Looks like the event object goes straight to pinvokes, so a null result just crashes?
I like the idea of a 'startup-script' for mod_managed. It would also be excellent if there was an event or message informing the background code to terminate nicely when the module reloads.
--Josh
On Wed, Sep 9, 2009 at 12:57 PM, Jeff Lenk <jlenk@frontiernet.net> wrote:
I think the problem here is that the loader only keeps this method in scope until completion then it drops the remoted connection. Therefore you should not use threads in this method. Michael please correct me if I am wrong here.
As an example of the failure simply just put a Sleep(10000) call in the thread and you will see the failure.
As Michael said this method was only designed to allow the option to opt out of being loaded.
In order to support this perhaps a configuration flag simular to the lua "startup-script" should be added.
Here is the error I get with the loop I mentioned. -Josh [image: Capture.PNG]
On Tue, Sep 8, 2009 at 5:05 AM, Michael Giagnocavo <mgg@giagnocavo.net>wrote:
Hi,
Can you please elaborate on the crash you receive when you queue a thread during load?
Thanks,
Michael
-- View this message in context: http://n2.nabble.com/Subscribing-to-events-in-managed-C-NET-tp3573619p361319... Sent from the freeswitch-users mailing list archive at Nabble.com.
_______________________________________________ FreeSWITCH-users mailing list FreeSWITCH-users@lists.freeswitch.org http://lists.freeswitch.org/mailman/listinfo/freeswitch-users UNSUBSCRIBE:http://lists.freeswitch.org/mailman/options/freeswitch-users http://www.freeswitch.org
_______________________________________________ FreeSWITCH-users mailing list FreeSWITCH-users@lists.freeswitch.org http://lists.freeswitch.org/mailman/listinfo/freeswitch-users UNSUBSCRIBE:http://lists.freeswitch.org/mailman/options/freeswitch-users http://www.freeswitch.org
_______________________________________________ FreeSWITCH-users mailing list FreeSWITCH-users@lists.freeswitch.org http://lists.freeswitch.org/mailman/listinfo/freeswitch-users UNSUBSCRIBE:http://lists.freeswitch.org/mailman/options/freeswitch-users http://www.freeswitch.org
_______________________________________________ FreeSWITCH-users mailing list FreeSWITCH-users@lists.freeswitch.org http://lists.freeswitch.org/mailman/listinfo/freeswitch-users UNSUBSCRIBE:http://lists.freeswitch.org/mailman/options/freeswitch-users http://www.freeswitch.org
_______________________________________________ FreeSWITCH-users mailing list FreeSWITCH-users@lists.freeswitch.org http://lists.freeswitch.org/mailman/listinfo/freeswitch-users UNSUBSCRIBE:http://lists.freeswitch.org/mailman/options/freeswitch-users http://www.freeswitch.org