Thank you Matej

You are correct to bring up threading, this is done on a single thread - for performance reasons WebSphere keeps both ends of an RMI on a single thread for performance reasons.

But I confess this feels a little unintuitive to me. The fact we do everything on one thread feels like an implementation detail. What would happen if there was a new thread for processing an RMI? Would weld try to identify the correct pre-existing request context to associate with? Or require us to do so? When drafting this reply to you I noticed this in the spec that I missed the first time:

The context associated with a built-in normal scope propagates across local, synchronous Java method calls, including invocation of EJB local business methods. The context does not propagate across remote method invocations or to asynchronous processes such as JMS message listeners or EJB timer service timeouts.

Please could you help me understand how that "does not propagate" fits in what you've said before. Are you saying it really does come down to the fact we're handling RMIs on the same thread as the service() method that invokes them?

Regards
Benjamin



From:        Matej Novotny <manovotn@redhat.com>
To:        Benjamin Confino <BENJAMIC@uk.ibm.com>
Cc:        weld-dev@lists.jboss.org
Date:        01/06/2021 14:27
Subject:        [EXTERNAL] Re: [weld-dev] Request Scope Lifecycle and Remote EJB invocations



Hello, from your description and the diagram I gather that the whole thing happens inside a single service() invocation. The thing with CDI contexts is that they are thread-bound. Your context either is or isn't active on a thread. žžžž
Hello,

from your description and the diagram I gather that the whole thing happens inside a single service() invocation.

The thing with CDI contexts is that they are thread-bound. Your context either is or isn't active on a thread.
CDI defines several cases for which it has to be active - such as the service() invocation or the RMI - but it doesn't say anything about the context being a new one; just active, which holds true in your scenario.

Therefore, what you are seeing is IMO expected behavior.
While executing the sequence of actions, you are still on one thread and the original req. context (that is activated by an actual HTTP request) is supposed to live until the service() method ends.
By ending it mid-way, you'd violate its lifecycle. Not to mention that it could create some weird situations if you attempted to pause/stop/restart the same context several times in one request.

Also, the same logic is applied to whenever CDI spec requires some context to be active for certain action (async events, RMI, ...) - we basically check if it already is active and only in case it isn't we activate it.

That's my understanding of it, hope it helps :-)

Regards
Matej

On Tue, Jun 1, 2021 at 11:22 AM Benjamin Confino <BENJAMIC@uk.ibm.com> wrote:
Hello

I have a customer who sent me a pair of applications with the following behaviour (see also the attached flow diagram):
1.        A request is sent to a REST resource in Application A
2.        The endpoint makes a RMI call to ServiceB in Application B
3.        ServiceB makes an RMI to ServiceA in Application A
4.        ServiceA makes an RMI to ServiceB in Application B
5.        ServiceB calls back to the rest resource.
A @RequestScoped bean is injected into the REST Resource and Service A.

The customer has noticed the same instance of that bean is injected into both injection points. They believe the two should be separate request scoped contexts.

The CDI spec says the request context is alive : But it is not clear what happens if a remote method invocation is nested inside a service() method. Is that a separate request context or the same as the parent?


If they are seperate contexts, I notice that WeldInitialListener.requestInitialized requires a ServletRequestEvent, which would be tricky to acquire as part of processing an EJB remote method invocation. Would I need to find a way to access a ServletRequestEvent when creating the request context for a RMI?


Unless stated otherwise above:
IBM United Kingdom Limited - Registered in England and Wales with number 741598.
Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU
_______________________________________________
weld-dev mailing list -- weld-dev@lists.jboss.org
To unsubscribe send an email to weld-dev-leave@lists.jboss.org
%(web_page_url)slistinfo%(cgiext)s/%(_internal_name)s


Unless stated otherwise above:
IBM United Kingdom Limited - Registered in England and Wales with number 741598.
Registered office: PO Box 41, North Harbour, Portsmouth, Hampshire PO6 3AU