users@grizzly.java.net

Re: Grizzly 1.9.18 upgrade troubles

From: Oleksiy Stashok <Oleksiy.Stashok_at_Sun.COM>
Date: Wed, 09 Sep 2009 18:53:58 +0200

Hi Alexei,

just wanted to answer the Copyable related part of your questions.
Basically, I'd say cloning wasn't good solution at all, and we are
going to make refactoring and remove cloning, copying etc.
I'll start to work on that tomorrow and nearest time hope to have that
completed (till the end of the week and beginning of next).

I'll let you know, once it will be completed.

Thanks.

WBR,
Alexey.

On Sep 8, 2009, at 20:28 , Alexei Dets wrote:

> Hi!
>
> In the company I work for we have some applications using grizzly
> framework
> and recently I've tried to update them from grizzly-1.9.5 to 1.9.18-
> M2 and
> unfortunately (and to say truth, unexpectedly) faced various issues.
> The
> main cause of the problem is the incompatible and AFAIK undocumented
> behavior change of the framework (Controller's readThreadCount) but I
> wonder that may be we are also doing something in completely
> wrong/suboptimal way. Unfortunately documentation is somewhat
> lacking :-(
>
> So, the issues:
>
> 1. After upgrade code perfectly compiles and runs without any
> warnings.
> Unfortunately, application just stops working. It listens to ports and
> that's all, it does nothing else. The cause of the problem is
> Controller.readThreadCount that is now by default > 0 and this opens
> a HUGE
> can of worms as I'll try to explain below. Setting readThreadCount
> to 0
> makes application work again.
>
> 2. This new default behavior with multiple SelectorHandler's and
> ReadControllers heavily relies on implementation of Copyable.copyTo
> method
> in SelectorHandler & SelectionKeyHandler. If application provides
> its own
> handlers then in general (if they have instance variables) these
> handlers
> MUST provide their own copyTo implementations. AFAIK this is not
> documented
> anywhere. Apart from this, copyTo (actually Cloner.clone) imposes
> additional restriction on implementations.
>
> 3. Classes implementing Copyable interface can't have final instance
> variables (copyTo will not be able to copy them).
>
> 4. Classes implementing Copyable interface can't have default
> constructor
> with the visibility different from public (this includes the case if
> the
> whole class has, say, default visibility). Cloner.clone will fail with
> IllegalAccessException because in case of default constructor it
> only checks
> for InstantiationException.
>
> 5. Even after fixes to constructors, final variables and copyTo
> implementations application will still not work because application's
> SelectorHandler is just silently ignored. The main reason (if not
> the only)
> to provide a custom SelectorHandler is to override onAcceptInterest
> method
> to do processing of incoming connections. This is what is showed in
> all
> code samples/tutorials on Grizzly I'm aware of and this is exactly
> that
> doesn't work any more because now accept is done by (hardcoded in
> Controller.start()) RoundRobinSelectorHandler and so custom
> onAcceptInterest method in provided SelectorHandler is never ever
> called.
>
> I'd like to here some suggestions about how this all is supposed to
> work,
> what we should change in our apps and any comments & suggestions on
> this in
> general. BTW, it is also not clear why Controller needs _copies_ of
> SelectorHandler/SelectionKeyHandler and can't use _one_ instance
> shared
> between threads.
>
> Alexei
>
>
> ---------------------------------------------------------------------
> To unsubscribe, e-mail: users-unsubscribe_at_grizzly.dev.java.net
> For additional commands, e-mail: users-help_at_grizzly.dev.java.net
>