inheritance of cdi scopes

classic Classic list List threaded Threaded
44 messages Options
123
Reply | Threaded
Open this post in threaded view
|

Re: inheritance of cdi scopes

Reza Rahman
Even in the most conservative reading of this, the spec is clearly not disallowing it.

> On Mar 7, 2016, at 10:05 AM, Mark Struberg <[hidden email]> wrote:
>
> The question is whether the spec does allow us to do it. And if other containers consequently do it as well.
>
> If it does then I will implement it in TomEE.
>
> LieGrue,
> strub
>
>
>> Am 07.03.2016 um 14:06 schrieb Reza Rahman <[hidden email]>:
>>
>> What this is saying is that it is not recommended to use them because of the possible life-cycle mismatch. If they are not supposed to work at all, the specification would have simply stated it won't work.
>>
>> Anyway personally I have no reason to further discuss this. I'm going to try to find a way to get this done for developers sooner rather than later. If TomEE does not want to do it, too bad for developers.
>>
>>> On Mar 7, 2016, at 3:49 AM, Romain Manni-Bucau <[hidden email]> wrote:
>>>
>>> "
>>> Tasks that are submitted to a managed instance of ExecutorService may still be running after the lifecycle of the submitting component. Therefore, CDI beans with a scope of @RequestScoped, @SessionScoped, or @ConversationScoped are not recommended to use as tasks as it cannot be guaranteed that the tasks will complete before the CDI context is destroyed.
>>> "
>>>
>>> States that the context is not inherited, is that what you mean?
>>>
>>>
>>>
>>> Romain Manni-Bucau
>>> @rmannibucau |  Blog | Github | LinkedIn | Tomitriber
>>>
>>> 2016-03-07 5:57 GMT+01:00 Reza Rahman <[hidden email]>:
>>> The specification currently references pretty much all the major CDI scopes specifically with the issue of propagation and lifecycle in mind. Please see section 2.3.
>>>
>>>> On Mar 6, 2016, at 11:53 PM, Mark Struberg <[hidden email]> wrote:
>>>> Specifically
>>>> The containers mimic ejb for propagation for a good reason!
>>>> No session e.g. , new TX, etc
>>>>
>>>> Sadly the concurrency utilis only mention @ApplicationScoped, so the Request Context not only doesn't get propagated (which is good), but also doesn't get set up (which is crap).
>>>>
>>>> LieGrue,
>>>> Strub
>>>>
>>>>> Am 06.03.2016 um 23:03 schrieb John D. Ament <[hidden email]>:
>>>>>
>>>>> I agree, in a sense, with what you're saying.  There's nothing in this spec that says it wouldn't be propagated.  However, there's nothing in this spec that states clearly that CDI contexts are propagated.
>>>>>
>>>>> If you look at the RI, the RI only seems to propagate transaction state.  Considering the age of the spec, I'm not surprised to see that.  The worst part is that right now, outside of the ASF, all other EE7 impls seem to be using the RI for concurrency.
>>>>>
>>>>> I'm fairly certain that from this spec's standpoint, the only thing that's actually propagated is the transaction.
>>>>>
>>>>> John
>>>>>
>>>>> On Sun, Mar 6, 2016 at 4:50 PM Reza Rahman <[hidden email]> wrote:
>>>>> I am re-reading the spec end to end again right now. So far it seems I have remembered everything correctly.
>>>>>
>>>>> You should read over section 2.3. What it is saying is that a container implementing the Java EE concurrency utilities should ensure whatever contextual information is needed for managed components to work correctly should be propagated automatically. For the correct implementation of CDI scopes, this should also mean any currently active scopes. The section you are referring to is basically implying that thinking that it is possible to use these scoped beans in tasks (albeit not reliably since beans could go out of scope before the thread finishes - for example if the request ends).
>>>>>
>>>>> This does not have anything to do with the context service per se. The context service is an SPI of sorts to allow end user developers to do for their own applications what the container does behind the scenes for managed component context propagation.
>>>>>
>>>>> I'll read over the entire spec to see if there is anything to contradict this. If that's not the case what Romain is describing is most likely an implementation specific bug that did not take into account CDI scope propagation.
>>>>>
>>>>>> On Mar 6, 2016, at 4:23 PM, John D. Ament <[hidden email]> wrote:
>>>>>>
>>>>>> Reza,
>>>>>>
>>>>>> I read through the concurrency utils spec.  Was there a specific section you had in mind?  The only references to CDI were near the beginning warning users to not use Request/Session scoped beans as tasks since the outer most context may be destroyed before the work is done.
>>>>>>
>>>>>> I have a feeling what you're referring to is the context service: http://docs.oracle.com/javaee/7/api/javax/enterprise/concurrent/ContextService.html
>>>>>>
>>>>>> If that's the case, then basically this should work OOTB right?
>>>>>>
>>>>>> Task task = new MyTask();
>>>>>> task = contextService.createContextualProxy(task, Task.class);
>>>>>> executor.submit(task);
>>>>>>
>>>>>> // now magically the context should be prop'd?
>>>>>>
>>>>>> Is that about right?
>>>>>>
>>>>>> John
>>>>>>
>>>>>> On Sun, Mar 6, 2016 at 3:30 PM Reza Rahman <[hidden email]> wrote:
>>>>>> Have you actually looked at the EE concurrency spec text in detail? What does it say about managed component context propagation?
>>>>>>
>>>>>> Without actually doing that further discussing this is just taking shots in the dark. As an implementer it should not surprise you that this might simply be a bug because the person implementing the concurrency utilities for the EE runtime was not told about what to copy over into the new thread for CDI to work correctly.
>>>>>>
>>>>>>> On Mar 6, 2016, at 3:06 PM, Romain Manni-Bucau <[hidden email]> wrote:
>>>>>>>
>>>>>>>
>>>>>>> 2016-03-06 20:59 GMT+01:00 Reza Rahman <[hidden email]>:
>>>>>>> As far as I know this is precisely the sort of thing that the EE concurrency spec is intended for. It is supposed to copy over everything from the underlying thread local context into the new thread for all EE managed components to function. Since CDI beans are also EE container managed, it also applies to CDI beans as well. The EE vendor is supposed to make sure this works properly.
>>>>>>>
>>>>>>> I don't think the concurrency utilities specifically lists APIs for which thread context propagation should work. If this doesn't work in a specific implementation it's most likely because they didn't take CDI into account in their own EE concurrency implementation.
>>>>>>>
>>>>>>>
>>>>>>> That's what I wanted/would like. CDI TCK breaks it quite easily and @RequestScoped which is *used* today is sadly  a @ThreadLocalScoped badly named. So to solve it we would need another scope as I mentionned several times on this list 100% matching servlet instances lifecycles (on a pure CDI side we have the same issue for sessions which are recycled during a request, the session scope is corrupted *by spec* in term of user behavior).
>>>>>>>
>>>>>>>> On Mar 6, 2016, at 2:45 PM, John D. Ament <[hidden email]> wrote:
>>>>>>>>
>>>>>>>> The section of the spec you link to makes no references to threads.  6.3 makes some notes about normal scopes and threads, and specifically says that a context is bound to one or more threads.
>>>>>>>>
>>>>>>>> I think what's happened is that over the years, people have simply bound HTTP Request == single thread, but when async processing was introduced no one thought to clarify that the spawning of a child thread from the original HTTP request retains the parent's context.
>>>>>>>>
>>>>>>>> This is another requested feature, but looks more like a bug or gap in the spec.
>>>>>>>>
>>>>>>>> John
>>>>>>>>
>>>>>>>> On Sun, Mar 6, 2016 at 2:37 PM Romain Manni-Bucau <[hidden email]> wrote:
>>>>>>>> 2016-03-06 20:25 GMT+01:00 Reza Rahman <[hidden email]>:
>>>>>>>> Let's see. I suspect the specification text for EE concurrency is generic enough for implementations to also be able to cover CDI scopes or any other Java EE API context propagation needs. This means the issue needs to be solved at the individual implementation level. Changing anything in the spec is probably just unnecessary ceremony in this case.
>>>>>>>>
>>>>>>>>
>>>>>>>> Then 1. concurrency- utility can't be reliable for "EE" users, 2. CDI still prevent it to work since it would violate the spec to propagate it while request scope is bound to another thread (http://docs.jboss.org/cdi/spec/1.1/cdi-spec.html#request_context handles async listener but not the main AsyncContext part).
>>>>>>>>
>>>>>>>>> On Mar 6, 2016, at 2:15 PM, Romain Manni-Bucau <[hidden email]> wrote:
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> 2016-03-06 19:42 GMT+01:00 Reza Rahman <[hidden email]>:
>>>>>>>>> This frankly surprises me. I'll check the specification text. This might indeed just be an implementation bug. The EE concurrency utilities are supposed to be copying all relevant context. If this is an issue than it has to be that it is not copying enough of the HTTP request context for CDI to work.
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> The issue is not technical since I got it working but needed to reverse. From my understanding ee concurrency utilities was done in a time CDI was not there so it just ignored it somehow and it hasnt been updated when integrated to the spec. Now with the wording of the CDI - and TCK - it is impossible to make it working since request scope is bound the thre request thread - and not the request. Side note: same applies to session scope and conversation.
>>>>>>>>>
>>>>>>>>> Surely the Red Hat folks can quickly shed some light here since they implement essentially this whole stack?
>>>>>>>>>
>>>>>>>>>> On Mar 6, 2016, at 1:30 PM, Romain Manni-Bucau <[hidden email]> wrote:
>>>>>>>>>>
>>>>>>>>>> 2016-03-06 19:20 GMT+01:00 Reza Rahman <[hidden email]>:
>>>>>>>>>> Can you kindly try to make the example a bit simpler? It's important to make the case for how likely this is supposed to occur in most business applications.
>>>>>>>>>>
>>>>>>>>>> Also, other than making sure that the executor service is propagating thread local request contexts correctly what other solution are you proposing? Did you check the specification? How sure are you that this isn't simply an implementation bug?
>>>>>>>>>>
>>>>>>>>>> As far as I know the executor service is supposed to be preserving all relevant parts of the EE context?
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> Not in concurrency-utilities for EE at least. That was the first impl I did then Mark pointed out it was violating CDI spec and request scope definition. There is a kind of contracdiction there cause concurrency-utilities doesn't integrate with CDI at all but we can also see it the opposite way: CDI doesn't provide any way to propagate a context in another thread. Both point of view are valid so we need to see where we tackle it.
>>>>>>>>>>
>>>>>>>>>>> On Mar 6, 2016, at 12:35 PM, Romain Manni-Bucau <[hidden email]> wrote:
>>>>>>>>>>>
>>>>>>>>>>> does https://gist.github.com/rmannibucau/d55fce47b001185dca3e help?
>>>>>>>>>>>
>>>>>>>>>>> Idea is to give an API to make:
>>>>>>>>>>>
>>>>>>>>>>>                public void complete() {
>>>>>>>>>>>                    try {
>>>>>>>>>>>                        asyncContext.complete();
>>>>>>>>>>>                    } finally {
>>>>>>>>>>>                        auditContext.end();
>>>>>>>>>>>                    }
>>>>>>>>>>>                }
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> working without hacky and almost impossible context pushing (cause of injections nature you are not supposed to know what to push in the context when going async).
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> Romain Manni-Bucau
>>>>>>>>>>> @rmannibucau |  Blog | Github | LinkedIn | Tomitriber
>>>>>>>>>>>
>>>>>>>>>>> 2016-03-06 16:40 GMT+01:00 Reza Rahman <[hidden email]>:
>>>>>>>>>>> Can you kindly share an annotated code example of the proposed solution so we can all follow and discuss this?
>>>>>>>>>>>
>>>>>>>>>>> On Mar 6, 2016, at 9:31 AM, Romain Manni-Bucau <[hidden email]> wroteshar:
>>>>>>>>>>>
>>>>>>>>>>>> Hi guys,
>>>>>>>>>>>>
>>>>>>>>>>>> spoke on concurrency utilities about the ability to inherit a cdi scope. Idea is to follow request scope more than cdi spec allows. First thought it was a concurrency utilities thing but Reza mentionned can be a CDI one so here it is.
>>>>>>>>>>>>
>>>>>>>>>>>> Sample:
>>>>>>>>>>>> In a servlet i get MyBean which is @RequestScoped, I do some set on it. The i go async (AsyncContext) and trigger a task in another thread. It would be neat - and mandatory in some case by the loose coupling nature of CDI - to get the *same* MyBean *instance* in this thread. With a direct dependency you can easily use message passing pattern - but you loose the loose coupling cause you need to know until which level you unwrap, think t principal case which has 2-3 proxies!. However in practice you have a lot of undirect dependencies, in particular with enterprise concerns (auditing, security...) so you can't really do it easily/naturally.
>>>>>>>>>>>>
>>>>>>>>>>>> Bonus:
>>>>>>>>>>>> One very verbose way is to be able to kind of push/pop an existing context in a thread - wrappers doing it on a Runnable/Consumer/Function/... would be neat.
>>>>>>>>>>>>
>>>>>>>>>>>> Question:
>>>>>>>>>>>> Would CDI handle it in 2.0?
>>>>>>>>>>>>
>>>>>>>>>>>> Side note: this is really about the fact to reuse a "context context" (its current instances map) in another thread the more transparently possible and match the user vision more than a technical question for now.
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> Romain Manni-Bucau
>>>>>>>>>>>> @rmannibucau |  Blog | Github | LinkedIn | Tomitriber
>>>>>>>>>>>> _______________________________________________
>>>>>>>>>>>> cdi-dev mailing list
>>>>>>>>>>>> [hidden email]
>>>>>>>>>>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>>>>>>>>>>
>>>>>>>>>>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>>>>>>>>>>>
>>>>>>>>>>> _______________________________________________
>>>>>>>>>>> cdi-dev mailing list
>>>>>>>>>>> [hidden email]
>>>>>>>>>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>>>>>>>>>
>>>>>>>>>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>>>>>>>>>>
>>>>>>>>>> _______________________________________________
>>>>>>>>>> cdi-dev mailing list
>>>>>>>>>> [hidden email]
>>>>>>>>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>>>>>>>>
>>>>>>>>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>>>>>>>>>
>>>>>>>>> _______________________________________________
>>>>>>>>> cdi-dev mailing list
>>>>>>>>> [hidden email]
>>>>>>>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>>>>>>>
>>>>>>>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>>>>>>>>
>>>>>>>> _______________________________________________
>>>>>>>> cdi-dev mailing list
>>>>>>>> [hidden email]
>>>>>>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>>>>>>
>>>>>>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>>>>>>>> _______________________________________________
>>>>>>>> cdi-dev mailing list
>>>>>>>> [hidden email]
>>>>>>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>>>>>>
>>>>>>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>>>>>>>
>>>>>>> _______________________________________________
>>>>>>> cdi-dev mailing list
>>>>>>> [hidden email]
>>>>>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>>>>>
>>>>>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>>>>>> _______________________________________________
>>>>>> cdi-dev mailing list
>>>>>> [hidden email]
>>>>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>>>>
>>>>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>>>>>> _______________________________________________
>>>>>> cdi-dev mailing list
>>>>>> [hidden email]
>>>>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>>>>
>>>>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>>>>> _______________________________________________
>>>>> cdi-dev mailing list
>>>>> [hidden email]
>>>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>>>
>>>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>>>>> _______________________________________________
>>>>> cdi-dev mailing list
>>>>> [hidden email]
>>>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>>>
>>>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>>>
>>> _______________________________________________
>>> cdi-dev mailing list
>>> [hidden email]
>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>
>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>> _______________________________________________
>> cdi-dev mailing list
>> [hidden email]
>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>
>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>

_______________________________________________
cdi-dev mailing list
[hidden email]
https://lists.jboss.org/mailman/listinfo/cdi-dev

Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
Reply | Threaded
Open this post in threaded view
|

Re: inheritance of cdi scopes

Reza Rahman
Reading over the CDI spec definition for request scoped beans, I am a tad confused. When are request scoped beans being destroyed right now? Are they just bound to the Servlet request thread and destroyed as soon as the service method returns? In case of asynchronous Servlets, are they kept around until the real HTTP request actually completes the same way the underlying HTTP connection is kept around? Or is that too difficult because it would require integration at a very low level with the Servlet implementation?

There's some language around asynchronous completion right now but it's not very clear what actually happens. Does the onComplete, etc asynchronous callback basically create new request scoped instances?

> On Mar 7, 2016, at 10:28 AM, Reza Rahman <[hidden email]> wrote:
>
> Even in the most conservative reading of this, the spec is clearly not disallowing it.
>
>> On Mar 7, 2016, at 10:05 AM, Mark Struberg <[hidden email]> wrote:
>>
>> The question is whether the spec does allow us to do it. And if other containers consequently do it as well.
>>
>> If it does then I will implement it in TomEE.
>>
>> LieGrue,
>> strub
>>
>>
>>> Am 07.03.2016 um 14:06 schrieb Reza Rahman <[hidden email]>:
>>>
>>> What this is saying is that it is not recommended to use them because of the possible life-cycle mismatch. If they are not supposed to work at all, the specification would have simply stated it won't work.
>>>
>>> Anyway personally I have no reason to further discuss this. I'm going to try to find a way to get this done for developers sooner rather than later. If TomEE does not want to do it, too bad for developers.
>>>
>>>> On Mar 7, 2016, at 3:49 AM, Romain Manni-Bucau <[hidden email]> wrote:
>>>>
>>>> "
>>>> Tasks that are submitted to a managed instance of ExecutorService may still be running after the lifecycle of the submitting component. Therefore, CDI beans with a scope of @RequestScoped, @SessionScoped, or @ConversationScoped are not recommended to use as tasks as it cannot be guaranteed that the tasks will complete before the CDI context is destroyed.
>>>> "
>>>>
>>>> States that the context is not inherited, is that what you mean?
>>>>
>>>>
>>>>
>>>> Romain Manni-Bucau
>>>> @rmannibucau |  Blog | Github | LinkedIn | Tomitriber
>>>>
>>>> 2016-03-07 5:57 GMT+01:00 Reza Rahman <[hidden email]>:
>>>> The specification currently references pretty much all the major CDI scopes specifically with the issue of propagation and lifecycle in mind. Please see section 2.3.
>>>>
>>>>> On Mar 6, 2016, at 11:53 PM, Mark Struberg <[hidden email]> wrote:
>>>>> Specifically
>>>>> The containers mimic ejb for propagation for a good reason!
>>>>> No session e.g. , new TX, etc
>>>>>
>>>>> Sadly the concurrency utilis only mention @ApplicationScoped, so the Request Context not only doesn't get propagated (which is good), but also doesn't get set up (which is crap).
>>>>>
>>>>> LieGrue,
>>>>> Strub
>>>>>
>>>>>> Am 06.03.2016 um 23:03 schrieb John D. Ament <[hidden email]>:
>>>>>>
>>>>>> I agree, in a sense, with what you're saying.  There's nothing in this spec that says it wouldn't be propagated.  However, there's nothing in this spec that states clearly that CDI contexts are propagated.
>>>>>>
>>>>>> If you look at the RI, the RI only seems to propagate transaction state.  Considering the age of the spec, I'm not surprised to see that.  The worst part is that right now, outside of the ASF, all other EE7 impls seem to be using the RI for concurrency.
>>>>>>
>>>>>> I'm fairly certain that from this spec's standpoint, the only thing that's actually propagated is the transaction.
>>>>>>
>>>>>> John
>>>>>>
>>>>>> On Sun, Mar 6, 2016 at 4:50 PM Reza Rahman <[hidden email]> wrote:
>>>>>> I am re-reading the spec end to end again right now. So far it seems I have remembered everything correctly.
>>>>>>
>>>>>> You should read over section 2.3. What it is saying is that a container implementing the Java EE concurrency utilities should ensure whatever contextual information is needed for managed components to work correctly should be propagated automatically. For the correct implementation of CDI scopes, this should also mean any currently active scopes. The section you are referring to is basically implying that thinking that it is possible to use these scoped beans in tasks (albeit not reliably since beans could go out of scope before the thread finishes - for example if the request ends).
>>>>>>
>>>>>> This does not have anything to do with the context service per se. The context service is an SPI of sorts to allow end user developers to do for their own applications what the container does behind the scenes for managed component context propagation.
>>>>>>
>>>>>> I'll read over the entire spec to see if there is anything to contradict this. If that's not the case what Romain is describing is most likely an implementation specific bug that did not take into account CDI scope propagation.
>>>>>>
>>>>>>> On Mar 6, 2016, at 4:23 PM, John D. Ament <[hidden email]> wrote:
>>>>>>>
>>>>>>> Reza,
>>>>>>>
>>>>>>> I read through the concurrency utils spec.  Was there a specific section you had in mind?  The only references to CDI were near the beginning warning users to not use Request/Session scoped beans as tasks since the outer most context may be destroyed before the work is done.
>>>>>>>
>>>>>>> I have a feeling what you're referring to is the context service: http://docs.oracle.com/javaee/7/api/javax/enterprise/concurrent/ContextService.html
>>>>>>>
>>>>>>> If that's the case, then basically this should work OOTB right?
>>>>>>>
>>>>>>> Task task = new MyTask();
>>>>>>> task = contextService.createContextualProxy(task, Task.class);
>>>>>>> executor.submit(task);
>>>>>>>
>>>>>>> // now magically the context should be prop'd?
>>>>>>>
>>>>>>> Is that about right?
>>>>>>>
>>>>>>> John
>>>>>>>
>>>>>>> On Sun, Mar 6, 2016 at 3:30 PM Reza Rahman <[hidden email]> wrote:
>>>>>>> Have you actually looked at the EE concurrency spec text in detail? What does it say about managed component context propagation?
>>>>>>>
>>>>>>> Without actually doing that further discussing this is just taking shots in the dark. As an implementer it should not surprise you that this might simply be a bug because the person implementing the concurrency utilities for the EE runtime was not told about what to copy over into the new thread for CDI to work correctly.
>>>>>>>
>>>>>>>> On Mar 6, 2016, at 3:06 PM, Romain Manni-Bucau <[hidden email]> wrote:
>>>>>>>>
>>>>>>>>
>>>>>>>> 2016-03-06 20:59 GMT+01:00 Reza Rahman <[hidden email]>:
>>>>>>>> As far as I know this is precisely the sort of thing that the EE concurrency spec is intended for. It is supposed to copy over everything from the underlying thread local context into the new thread for all EE managed components to function. Since CDI beans are also EE container managed, it also applies to CDI beans as well. The EE vendor is supposed to make sure this works properly.
>>>>>>>>
>>>>>>>> I don't think the concurrency utilities specifically lists APIs for which thread context propagation should work. If this doesn't work in a specific implementation it's most likely because they didn't take CDI into account in their own EE concurrency implementation.
>>>>>>>>
>>>>>>>>
>>>>>>>> That's what I wanted/would like. CDI TCK breaks it quite easily and @RequestScoped which is *used* today is sadly  a @ThreadLocalScoped badly named. So to solve it we would need another scope as I mentionned several times on this list 100% matching servlet instances lifecycles (on a pure CDI side we have the same issue for sessions which are recycled during a request, the session scope is corrupted *by spec* in term of user behavior).
>>>>>>>>
>>>>>>>>> On Mar 6, 2016, at 2:45 PM, John D. Ament <[hidden email]> wrote:
>>>>>>>>>
>>>>>>>>> The section of the spec you link to makes no references to threads.  6.3 makes some notes about normal scopes and threads, and specifically says that a context is bound to one or more threads.
>>>>>>>>>
>>>>>>>>> I think what's happened is that over the years, people have simply bound HTTP Request == single thread, but when async processing was introduced no one thought to clarify that the spawning of a child thread from the original HTTP request retains the parent's context.
>>>>>>>>>
>>>>>>>>> This is another requested feature, but looks more like a bug or gap in the spec.
>>>>>>>>>
>>>>>>>>> John
>>>>>>>>>
>>>>>>>>> On Sun, Mar 6, 2016 at 2:37 PM Romain Manni-Bucau <[hidden email]> wrote:
>>>>>>>>> 2016-03-06 20:25 GMT+01:00 Reza Rahman <[hidden email]>:
>>>>>>>>> Let's see. I suspect the specification text for EE concurrency is generic enough for implementations to also be able to cover CDI scopes or any other Java EE API context propagation needs. This means the issue needs to be solved at the individual implementation level. Changing anything in the spec is probably just unnecessary ceremony in this case.
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> Then 1. concurrency- utility can't be reliable for "EE" users, 2. CDI still prevent it to work since it would violate the spec to propagate it while request scope is bound to another thread (http://docs.jboss.org/cdi/spec/1.1/cdi-spec.html#request_context handles async listener but not the main AsyncContext part).
>>>>>>>>>
>>>>>>>>>> On Mar 6, 2016, at 2:15 PM, Romain Manni-Bucau <[hidden email]> wrote:
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> 2016-03-06 19:42 GMT+01:00 Reza Rahman <[hidden email]>:
>>>>>>>>>> This frankly surprises me. I'll check the specification text. This might indeed just be an implementation bug. The EE concurrency utilities are supposed to be copying all relevant context. If this is an issue than it has to be that it is not copying enough of the HTTP request context for CDI to work.
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> The issue is not technical since I got it working but needed to reverse. From my understanding ee concurrency utilities was done in a time CDI was not there so it just ignored it somehow and it hasnt been updated when integrated to the spec. Now with the wording of the CDI - and TCK - it is impossible to make it working since request scope is bound the thre request thread - and not the request. Side note: same applies to session scope and conversation.
>>>>>>>>>>
>>>>>>>>>> Surely the Red Hat folks can quickly shed some light here since they implement essentially this whole stack?
>>>>>>>>>>
>>>>>>>>>>> On Mar 6, 2016, at 1:30 PM, Romain Manni-Bucau <[hidden email]> wrote:
>>>>>>>>>>>
>>>>>>>>>>> 2016-03-06 19:20 GMT+01:00 Reza Rahman <[hidden email]>:
>>>>>>>>>>> Can you kindly try to make the example a bit simpler? It's important to make the case for how likely this is supposed to occur in most business applications.
>>>>>>>>>>>
>>>>>>>>>>> Also, other than making sure that the executor service is propagating thread local request contexts correctly what other solution are you proposing? Did you check the specification? How sure are you that this isn't simply an implementation bug?
>>>>>>>>>>>
>>>>>>>>>>> As far as I know the executor service is supposed to be preserving all relevant parts of the EE context?
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> Not in concurrency-utilities for EE at least. That was the first impl I did then Mark pointed out it was violating CDI spec and request scope definition. There is a kind of contracdiction there cause concurrency-utilities doesn't integrate with CDI at all but we can also see it the opposite way: CDI doesn't provide any way to propagate a context in another thread. Both point of view are valid so we need to see where we tackle it.
>>>>>>>>>>>
>>>>>>>>>>>> On Mar 6, 2016, at 12:35 PM, Romain Manni-Bucau <[hidden email]> wrote:
>>>>>>>>>>>>
>>>>>>>>>>>> does https://gist.github.com/rmannibucau/d55fce47b001185dca3e help?
>>>>>>>>>>>>
>>>>>>>>>>>> Idea is to give an API to make:
>>>>>>>>>>>>
>>>>>>>>>>>>               public void complete() {
>>>>>>>>>>>>                   try {
>>>>>>>>>>>>                       asyncContext.complete();
>>>>>>>>>>>>                   } finally {
>>>>>>>>>>>>                       auditContext.end();
>>>>>>>>>>>>                   }
>>>>>>>>>>>>               }
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> working without hacky and almost impossible context pushing (cause of injections nature you are not supposed to know what to push in the context when going async).
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> Romain Manni-Bucau
>>>>>>>>>>>> @rmannibucau |  Blog | Github | LinkedIn | Tomitriber
>>>>>>>>>>>>
>>>>>>>>>>>> 2016-03-06 16:40 GMT+01:00 Reza Rahman <[hidden email]>:
>>>>>>>>>>>> Can you kindly share an annotated code example of the proposed solution so we can all follow and discuss this?
>>>>>>>>>>>>
>>>>>>>>>>>> On Mar 6, 2016, at 9:31 AM, Romain Manni-Bucau <[hidden email]> wroteshar:
>>>>>>>>>>>>
>>>>>>>>>>>>> Hi guys,
>>>>>>>>>>>>>
>>>>>>>>>>>>> spoke on concurrency utilities about the ability to inherit a cdi scope. Idea is to follow request scope more than cdi spec allows. First thought it was a concurrency utilities thing but Reza mentionned can be a CDI one so here it is.
>>>>>>>>>>>>>
>>>>>>>>>>>>> Sample:
>>>>>>>>>>>>> In a servlet i get MyBean which is @RequestScoped, I do some set on it. The i go async (AsyncContext) and trigger a task in another thread. It would be neat - and mandatory in some case by the loose coupling nature of CDI - to get the *same* MyBean *instance* in this thread. With a direct dependency you can easily use message passing pattern - but you loose the loose coupling cause you need to know until which level you unwrap, think t principal case which has 2-3 proxies!. However in practice you have a lot of undirect dependencies, in particular with enterprise concerns (auditing, security...) so you can't really do it easily/naturally.
>>>>>>>>>>>>>
>>>>>>>>>>>>> Bonus:
>>>>>>>>>>>>> One very verbose way is to be able to kind of push/pop an existing context in a thread - wrappers doing it on a Runnable/Consumer/Function/... would be neat.
>>>>>>>>>>>>>
>>>>>>>>>>>>> Question:
>>>>>>>>>>>>> Would CDI handle it in 2.0?
>>>>>>>>>>>>>
>>>>>>>>>>>>> Side note: this is really about the fact to reuse a "context context" (its current instances map) in another thread the more transparently possible and match the user vision more than a technical question for now.
>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>> Romain Manni-Bucau
>>>>>>>>>>>>> @rmannibucau |  Blog | Github | LinkedIn | Tomitriber
>>>>>>>>>>>>> _______________________________________________
>>>>>>>>>>>>> cdi-dev mailing list
>>>>>>>>>>>>> [hidden email]
>>>>>>>>>>>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>>>>>>>>>>>
>>>>>>>>>>>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>>>>>>>>>>>>
>>>>>>>>>>>> _______________________________________________
>>>>>>>>>>>> cdi-dev mailing list
>>>>>>>>>>>> [hidden email]
>>>>>>>>>>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>>>>>>>>>>
>>>>>>>>>>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>>>>>>>>>>>
>>>>>>>>>>> _______________________________________________
>>>>>>>>>>> cdi-dev mailing list
>>>>>>>>>>> [hidden email]
>>>>>>>>>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>>>>>>>>>
>>>>>>>>>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>>>>>>>>>>
>>>>>>>>>> _______________________________________________
>>>>>>>>>> cdi-dev mailing list
>>>>>>>>>> [hidden email]
>>>>>>>>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>>>>>>>>
>>>>>>>>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>>>>>>>>>
>>>>>>>>> _______________________________________________
>>>>>>>>> cdi-dev mailing list
>>>>>>>>> [hidden email]
>>>>>>>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>>>>>>>
>>>>>>>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>>>>>>>>> _______________________________________________
>>>>>>>>> cdi-dev mailing list
>>>>>>>>> [hidden email]
>>>>>>>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>>>>>>>
>>>>>>>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>>>>>>>>
>>>>>>>> _______________________________________________
>>>>>>>> cdi-dev mailing list
>>>>>>>> [hidden email]
>>>>>>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>>>>>>
>>>>>>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>>>>>>> _______________________________________________
>>>>>>> cdi-dev mailing list
>>>>>>> [hidden email]
>>>>>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>>>>>
>>>>>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>>>>>>> _______________________________________________
>>>>>>> cdi-dev mailing list
>>>>>>> [hidden email]
>>>>>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>>>>>
>>>>>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>>>>>> _______________________________________________
>>>>>> cdi-dev mailing list
>>>>>> [hidden email]
>>>>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>>>>
>>>>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>>>>>> _______________________________________________
>>>>>> cdi-dev mailing list
>>>>>> [hidden email]
>>>>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>>>>
>>>>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>>>>
>>>> _______________________________________________
>>>> cdi-dev mailing list
>>>> [hidden email]
>>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>>
>>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>>> _______________________________________________
>>> cdi-dev mailing list
>>> [hidden email]
>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>
>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>>

_______________________________________________
cdi-dev mailing list
[hidden email]
https://lists.jboss.org/mailman/listinfo/cdi-dev

Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
Reply | Threaded
Open this post in threaded view
|

Re: inheritance of cdi scopes

arjan tijms
Hi,

On Mon, Mar 7, 2016 at 8:08 PM, Reza Rahman <[hidden email]> wrote:
Reading over the CDI spec definition for request scoped beans, I am a tad confused. When are request scoped beans being destroyed right now? Are they just bound to the Servlet request thread and destroyed as soon as the service method returns?

In case of a Servlet request (request scoped beans are also tied to other kinds of "requests"), it's indeed not clear. In practice it looks like the moment between the first ServletRequestListener#requestInitialized and ServletRequestListener#requestDestroyed.

The exact scope is troublesome for security too, since in most severs the request scope (and session scope and application scope) is active when a SAM is called (the SAM gets an HttpServletRequest after all), but this is not the case in all servers. E.g. in Liberty the RequestScope starts AFTER a SAM is called.

Kind regards,
Arjan Tijms


 
In case of asynchronous Servlets, are they kept around until the real HTTP request actually completes the same way the underlying HTTP connection is kept around? Or is that too difficult because it would require integration at a very low level with the Servlet implementation?

There's some language around asynchronous completion right now but it's not very clear what actually happens. Does the onComplete, etc asynchronous callback basically create new request scoped instances?

> On Mar 7, 2016, at 10:28 AM, Reza Rahman <[hidden email]> wrote:
>
> Even in the most conservative reading of this, the spec is clearly not disallowing it.
>
>> On Mar 7, 2016, at 10:05 AM, Mark Struberg <[hidden email]> wrote:
>>
>> The question is whether the spec does allow us to do it. And if other containers consequently do it as well.
>>
>> If it does then I will implement it in TomEE.
>>
>> LieGrue,
>> strub
>>
>>
>>> Am 07.03.2016 um 14:06 schrieb Reza Rahman <[hidden email]>:
>>>
>>> What this is saying is that it is not recommended to use them because of the possible life-cycle mismatch. If they are not supposed to work at all, the specification would have simply stated it won't work.
>>>
>>> Anyway personally I have no reason to further discuss this. I'm going to try to find a way to get this done for developers sooner rather than later. If TomEE does not want to do it, too bad for developers.
>>>
>>>> On Mar 7, 2016, at 3:49 AM, Romain Manni-Bucau <[hidden email]> wrote:
>>>>
>>>> "
>>>> Tasks that are submitted to a managed instance of ExecutorService may still be running after the lifecycle of the submitting component. Therefore, CDI beans with a scope of @RequestScoped, @SessionScoped, or @ConversationScoped are not recommended to use as tasks as it cannot be guaranteed that the tasks will complete before the CDI context is destroyed.
>>>> "
>>>>
>>>> States that the context is not inherited, is that what you mean?
>>>>
>>>>
>>>>
>>>> Romain Manni-Bucau
>>>> @rmannibucau |  Blog | Github | LinkedIn | Tomitriber
>>>>
>>>> 2016-03-07 5:57 GMT+01:00 Reza Rahman <[hidden email]>:
>>>> The specification currently references pretty much all the major CDI scopes specifically with the issue of propagation and lifecycle in mind. Please see section 2.3.
>>>>
>>>>> On Mar 6, 2016, at 11:53 PM, Mark Struberg <[hidden email]> wrote:
>>>>> Specifically
>>>>> The containers mimic ejb for propagation for a good reason!
>>>>> No session e.g. , new TX, etc
>>>>>
>>>>> Sadly the concurrency utilis only mention @ApplicationScoped, so the Request Context not only doesn't get propagated (which is good), but also doesn't get set up (which is crap).
>>>>>
>>>>> LieGrue,
>>>>> Strub
>>>>>
>>>>>> Am 06.03.2016 um 23:03 schrieb John D. Ament <[hidden email]>:
>>>>>>
>>>>>> I agree, in a sense, with what you're saying.  There's nothing in this spec that says it wouldn't be propagated.  However, there's nothing in this spec that states clearly that CDI contexts are propagated.
>>>>>>
>>>>>> If you look at the RI, the RI only seems to propagate transaction state.  Considering the age of the spec, I'm not surprised to see that.  The worst part is that right now, outside of the ASF, all other EE7 impls seem to be using the RI for concurrency.
>>>>>>
>>>>>> I'm fairly certain that from this spec's standpoint, the only thing that's actually propagated is the transaction.
>>>>>>
>>>>>> John
>>>>>>
>>>>>> On Sun, Mar 6, 2016 at 4:50 PM Reza Rahman <[hidden email]> wrote:
>>>>>> I am re-reading the spec end to end again right now. So far it seems I have remembered everything correctly.
>>>>>>
>>>>>> You should read over section 2.3. What it is saying is that a container implementing the Java EE concurrency utilities should ensure whatever contextual information is needed for managed components to work correctly should be propagated automatically. For the correct implementation of CDI scopes, this should also mean any currently active scopes. The section you are referring to is basically implying that thinking that it is possible to use these scoped beans in tasks (albeit not reliably since beans could go out of scope before the thread finishes - for example if the request ends).
>>>>>>
>>>>>> This does not have anything to do with the context service per se. The context service is an SPI of sorts to allow end user developers to do for their own applications what the container does behind the scenes for managed component context propagation.
>>>>>>
>>>>>> I'll read over the entire spec to see if there is anything to contradict this. If that's not the case what Romain is describing is most likely an implementation specific bug that did not take into account CDI scope propagation.
>>>>>>
>>>>>>> On Mar 6, 2016, at 4:23 PM, John D. Ament <[hidden email]> wrote:
>>>>>>>
>>>>>>> Reza,
>>>>>>>
>>>>>>> I read through the concurrency utils spec.  Was there a specific section you had in mind?  The only references to CDI were near the beginning warning users to not use Request/Session scoped beans as tasks since the outer most context may be destroyed before the work is done.
>>>>>>>
>>>>>>> I have a feeling what you're referring to is the context service: http://docs.oracle.com/javaee/7/api/javax/enterprise/concurrent/ContextService.html
>>>>>>>
>>>>>>> If that's the case, then basically this should work OOTB right?
>>>>>>>
>>>>>>> Task task = new MyTask();
>>>>>>> task = contextService.createContextualProxy(task, Task.class);
>>>>>>> executor.submit(task);
>>>>>>>
>>>>>>> // now magically the context should be prop'd?
>>>>>>>
>>>>>>> Is that about right?
>>>>>>>
>>>>>>> John
>>>>>>>
>>>>>>> On Sun, Mar 6, 2016 at 3:30 PM Reza Rahman <[hidden email]> wrote:
>>>>>>> Have you actually looked at the EE concurrency spec text in detail? What does it say about managed component context propagation?
>>>>>>>
>>>>>>> Without actually doing that further discussing this is just taking shots in the dark. As an implementer it should not surprise you that this might simply be a bug because the person implementing the concurrency utilities for the EE runtime was not told about what to copy over into the new thread for CDI to work correctly.
>>>>>>>
>>>>>>>> On Mar 6, 2016, at 3:06 PM, Romain Manni-Bucau <[hidden email]> wrote:
>>>>>>>>
>>>>>>>>
>>>>>>>> 2016-03-06 20:59 GMT+01:00 Reza Rahman <[hidden email]>:
>>>>>>>> As far as I know this is precisely the sort of thing that the EE concurrency spec is intended for. It is supposed to copy over everything from the underlying thread local context into the new thread for all EE managed components to function. Since CDI beans are also EE container managed, it also applies to CDI beans as well. The EE vendor is supposed to make sure this works properly.
>>>>>>>>
>>>>>>>> I don't think the concurrency utilities specifically lists APIs for which thread context propagation should work. If this doesn't work in a specific implementation it's most likely because they didn't take CDI into account in their own EE concurrency implementation.
>>>>>>>>
>>>>>>>>
>>>>>>>> That's what I wanted/would like. CDI TCK breaks it quite easily and @RequestScoped which is *used* today is sadly  a @ThreadLocalScoped badly named. So to solve it we would need another scope as I mentionned several times on this list 100% matching servlet instances lifecycles (on a pure CDI side we have the same issue for sessions which are recycled during a request, the session scope is corrupted *by spec* in term of user behavior).
>>>>>>>>
>>>>>>>>> On Mar 6, 2016, at 2:45 PM, John D. Ament <[hidden email]> wrote:
>>>>>>>>>
>>>>>>>>> The section of the spec you link to makes no references to threads.  6.3 makes some notes about normal scopes and threads, and specifically says that a context is bound to one or more threads.
>>>>>>>>>
>>>>>>>>> I think what's happened is that over the years, people have simply bound HTTP Request == single thread, but when async processing was introduced no one thought to clarify that the spawning of a child thread from the original HTTP request retains the parent's context.
>>>>>>>>>
>>>>>>>>> This is another requested feature, but looks more like a bug or gap in the spec.
>>>>>>>>>
>>>>>>>>> John
>>>>>>>>>
>>>>>>>>> On Sun, Mar 6, 2016 at 2:37 PM Romain Manni-Bucau <[hidden email]> wrote:
>>>>>>>>> 2016-03-06 20:25 GMT+01:00 Reza Rahman <[hidden email]>:
>>>>>>>>> Let's see. I suspect the specification text for EE concurrency is generic enough for implementations to also be able to cover CDI scopes or any other Java EE API context propagation needs. This means the issue needs to be solved at the individual implementation level. Changing anything in the spec is probably just unnecessary ceremony in this case.
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> Then 1. concurrency- utility can't be reliable for "EE" users, 2. CDI still prevent it to work since it would violate the spec to propagate it while request scope is bound to another thread (http://docs.jboss.org/cdi/spec/1.1/cdi-spec.html#request_context handles async listener but not the main AsyncContext part).
>>>>>>>>>
>>>>>>>>>> On Mar 6, 2016, at 2:15 PM, Romain Manni-Bucau <[hidden email]> wrote:
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> 2016-03-06 19:42 GMT+01:00 Reza Rahman <[hidden email]>:
>>>>>>>>>> This frankly surprises me. I'll check the specification text. This might indeed just be an implementation bug. The EE concurrency utilities are supposed to be copying all relevant context. If this is an issue than it has to be that it is not copying enough of the HTTP request context for CDI to work.
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> The issue is not technical since I got it working but needed to reverse. From my understanding ee concurrency utilities was done in a time CDI was not there so it just ignored it somehow and it hasnt been updated when integrated to the spec. Now with the wording of the CDI - and TCK - it is impossible to make it working since request scope is bound the thre request thread - and not the request. Side note: same applies to session scope and conversation.
>>>>>>>>>>
>>>>>>>>>> Surely the Red Hat folks can quickly shed some light here since they implement essentially this whole stack?
>>>>>>>>>>
>>>>>>>>>>> On Mar 6, 2016, at 1:30 PM, Romain Manni-Bucau <[hidden email]> wrote:
>>>>>>>>>>>
>>>>>>>>>>> 2016-03-06 19:20 GMT+01:00 Reza Rahman <[hidden email]>:
>>>>>>>>>>> Can you kindly try to make the example a bit simpler? It's important to make the case for how likely this is supposed to occur in most business applications.
>>>>>>>>>>>
>>>>>>>>>>> Also, other than making sure that the executor service is propagating thread local request contexts correctly what other solution are you proposing? Did you check the specification? How sure are you that this isn't simply an implementation bug?
>>>>>>>>>>>
>>>>>>>>>>> As far as I know the executor service is supposed to be preserving all relevant parts of the EE context?
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> Not in concurrency-utilities for EE at least. That was the first impl I did then Mark pointed out it was violating CDI spec and request scope definition. There is a kind of contracdiction there cause concurrency-utilities doesn't integrate with CDI at all but we can also see it the opposite way: CDI doesn't provide any way to propagate a context in another thread. Both point of view are valid so we need to see where we tackle it.
>>>>>>>>>>>
>>>>>>>>>>>> On Mar 6, 2016, at 12:35 PM, Romain Manni-Bucau <[hidden email]> wrote:
>>>>>>>>>>>>
>>>>>>>>>>>> does https://gist.github.com/rmannibucau/d55fce47b001185dca3e help?
>>>>>>>>>>>>
>>>>>>>>>>>> Idea is to give an API to make:
>>>>>>>>>>>>
>>>>>>>>>>>>               public void complete() {
>>>>>>>>>>>>                   try {
>>>>>>>>>>>>                       asyncContext.complete();
>>>>>>>>>>>>                   } finally {
>>>>>>>>>>>>                       auditContext.end();
>>>>>>>>>>>>                   }
>>>>>>>>>>>>               }
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> working without hacky and almost impossible context pushing (cause of injections nature you are not supposed to know what to push in the context when going async).
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> Romain Manni-Bucau
>>>>>>>>>>>> @rmannibucau |  Blog | Github | LinkedIn | Tomitriber
>>>>>>>>>>>>
>>>>>>>>>>>> 2016-03-06 16:40 GMT+01:00 Reza Rahman <[hidden email]>:
>>>>>>>>>>>> Can you kindly share an annotated code example of the proposed solution so we can all follow and discuss this?
>>>>>>>>>>>>
>>>>>>>>>>>> On Mar 6, 2016, at 9:31 AM, Romain Manni-Bucau <[hidden email]> wroteshar:
>>>>>>>>>>>>
>>>>>>>>>>>>> Hi guys,
>>>>>>>>>>>>>
>>>>>>>>>>>>> spoke on concurrency utilities about the ability to inherit a cdi scope. Idea is to follow request scope more than cdi spec allows. First thought it was a concurrency utilities thing but Reza mentionned can be a CDI one so here it is.
>>>>>>>>>>>>>
>>>>>>>>>>>>> Sample:
>>>>>>>>>>>>> In a servlet i get MyBean which is @RequestScoped, I do some set on it. The i go async (AsyncContext) and trigger a task in another thread. It would be neat - and mandatory in some case by the loose coupling nature of CDI - to get the *same* MyBean *instance* in this thread. With a direct dependency you can easily use message passing pattern - but you loose the loose coupling cause you need to know until which level you unwrap, think t principal case which has 2-3 proxies!. However in practice you have a lot of undirect dependencies, in particular with enterprise concerns (auditing, security...) so you can't really do it easily/naturally.
>>>>>>>>>>>>>
>>>>>>>>>>>>> Bonus:
>>>>>>>>>>>>> One very verbose way is to be able to kind of push/pop an existing context in a thread - wrappers doing it on a Runnable/Consumer/Function/... would be neat.
>>>>>>>>>>>>>
>>>>>>>>>>>>> Question:
>>>>>>>>>>>>> Would CDI handle it in 2.0?
>>>>>>>>>>>>>
>>>>>>>>>>>>> Side note: this is really about the fact to reuse a "context context" (its current instances map) in another thread the more transparently possible and match the user vision more than a technical question for now.
>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>> Romain Manni-Bucau
>>>>>>>>>>>>> @rmannibucau |  Blog | Github | LinkedIn | Tomitriber
>>>>>>>>>>>>> _______________________________________________
>>>>>>>>>>>>> cdi-dev mailing list
>>>>>>>>>>>>> [hidden email]
>>>>>>>>>>>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>>>>>>>>>>>
>>>>>>>>>>>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>>>>>>>>>>>>
>>>>>>>>>>>> _______________________________________________
>>>>>>>>>>>> cdi-dev mailing list
>>>>>>>>>>>> [hidden email]
>>>>>>>>>>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>>>>>>>>>>
>>>>>>>>>>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>>>>>>>>>>>
>>>>>>>>>>> _______________________________________________
>>>>>>>>>>> cdi-dev mailing list
>>>>>>>>>>> [hidden email]
>>>>>>>>>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>>>>>>>>>
>>>>>>>>>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>>>>>>>>>>
>>>>>>>>>> _______________________________________________
>>>>>>>>>> cdi-dev mailing list
>>>>>>>>>> [hidden email]
>>>>>>>>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>>>>>>>>
>>>>>>>>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>>>>>>>>>
>>>>>>>>> _______________________________________________
>>>>>>>>> cdi-dev mailing list
>>>>>>>>> [hidden email]
>>>>>>>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>>>>>>>
>>>>>>>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>>>>>>>>> _______________________________________________
>>>>>>>>> cdi-dev mailing list
>>>>>>>>> [hidden email]
>>>>>>>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>>>>>>>
>>>>>>>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>>>>>>>>
>>>>>>>> _______________________________________________
>>>>>>>> cdi-dev mailing list
>>>>>>>> [hidden email]
>>>>>>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>>>>>>
>>>>>>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>>>>>>> _______________________________________________
>>>>>>> cdi-dev mailing list
>>>>>>> [hidden email]
>>>>>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>>>>>
>>>>>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>>>>>>> _______________________________________________
>>>>>>> cdi-dev mailing list
>>>>>>> [hidden email]
>>>>>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>>>>>
>>>>>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>>>>>> _______________________________________________
>>>>>> cdi-dev mailing list
>>>>>> [hidden email]
>>>>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>>>>
>>>>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>>>>>> _______________________________________________
>>>>>> cdi-dev mailing list
>>>>>> [hidden email]
>>>>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>>>>
>>>>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>>>>
>>>> _______________________________________________
>>>> cdi-dev mailing list
>>>> [hidden email]
>>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>>
>>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>>> _______________________________________________
>>> cdi-dev mailing list
>>> [hidden email]
>>> https://lists.jboss.org/mailman/listinfo/cdi-dev
>>>
>>> Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
>>

_______________________________________________
cdi-dev mailing list
[hidden email]
https://lists.jboss.org/mailman/listinfo/cdi-dev

Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.


_______________________________________________
cdi-dev mailing list
[hidden email]
https://lists.jboss.org/mailman/listinfo/cdi-dev

Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
Reply | Threaded
Open this post in threaded view
|

Re: inheritance of cdi scopes

Tomas Remes
In reply to this post by Romain Manni-Bucau

Hi,

If you feel that CDI TCK breaks something then raise an issue please.

Tom

----- Original Message -----
From: "Romain Manni-Bucau" <[hidden email]>
To: "Reza Rahman" <[hidden email]>
Cc: [hidden email]
Sent: Sunday, March 6, 2016 9:06:59 PM
Subject: Re: [cdi-dev] inheritance of cdi scopes


2016-03-06 20:59 GMT+01:00 Reza Rahman < [hidden email] > :



As far as I know this is precisely the sort of thing that the EE concurrency spec is intended for. It is supposed to copy over everything from the underlying thread local context into the new thread for all EE managed components to function. Since CDI beans are also EE container managed, it also applies to CDI beans as well. The EE vendor is supposed to make sure this works properly.

I don't think the concurrency utilities specifically lists APIs for which thread context propagation should work. If this doesn't work in a specific implementation it's most likely because they didn't take CDI into account in their own EE concurrency implementation.


That's what I wanted/would like. CDI TCK breaks it quite easily and @RequestScoped which is *used* today is sadly a @ThreadLocalScoped badly named. So to solve it we would need another scope as I mentionned several times on this list 100% matching servlet instances lifecycles (on a pure CDI side we have the same issue for sessions which are recycled during a request, the session scope is corrupted *by spec* in term of user behavior).



On Mar 6, 2016, at 2:45 PM, John D. Ament < [hidden email] > wrote:




The section of the spec you link to makes no references to threads. 6.3 makes some notes about normal scopes and threads, and specifically says that a context is bound to one or more threads.

I think what's happened is that over the years, people have simply bound HTTP Request == single thread, but when async processing was introduced no one thought to clarify that the spawning of a child thread from the original HTTP request retains the parent's context.

This is another requested feature, but looks more like a bug or gap in the spec.

John

On Sun, Mar 6, 2016 at 2:37 PM Romain Manni-Bucau < [hidden email] > wrote:



2016-03-06 20:25 GMT+01:00 Reza Rahman < [hidden email] > :



Let's see. I suspect the specification text for EE concurrency is generic enough for implementations to also be able to cover CDI scopes or any other Java EE API context propagation needs. This means the issue needs to be solved at the individual implementation level. Changing anything in the spec is probably just unnecessary ceremony in this case.


Then 1. concurrency- utility can't be reliable for "EE" users, 2. CDI still prevent it to work since it would violate the spec to propagate it while request scope is bound to another thread ( http://docs.jboss.org/cdi/spec/1.1/cdi-spec.html#request_context handles async listener but not the main AsyncContext part).



On Mar 6, 2016, at 2:15 PM, Romain Manni-Bucau < [hidden email] > wrote:





2016-03-06 19:42 GMT+01:00 Reza Rahman < [hidden email] > :



This frankly surprises me. I'll check the specification text. This might indeed just be an implementation bug. The EE concurrency utilities are supposed to be copying all relevant context. If this is an issue than it has to be that it is not copying enough of the HTTP request context for CDI to work.


The issue is not technical since I got it working but needed to reverse. From my understanding ee concurrency utilities was done in a time CDI was not there so it just ignored it somehow and it hasnt been updated when integrated to the spec. Now with the wording of the CDI - and TCK - it is impossible to make it working since request scope is bound the thre request thread - and not the request. Side note: same applies to session scope and conversation.



Surely the Red Hat folks can quickly shed some light here since they implement essentially this whole stack?

On Mar 6, 2016, at 1:30 PM, Romain Manni-Bucau < [hidden email] > wrote:




2016-03-06 19:20 GMT+01:00 Reza Rahman < [hidden email] > :



Can you kindly try to make the example a bit simpler? It's important to make the case for how likely this is supposed to occur in most business applications.

Also, other than making sure that the executor service is propagating thread local request contexts correctly what other solution are you proposing? Did you check the specification? How sure are you that this isn't simply an implementation bug?

As far as I know the executor service is supposed to be preserving all relevant parts of the EE context?


Not in concurrency-utilities for EE at least. That was the first impl I did then Mark pointed out it was violating CDI spec and request scope definition. There is a kind of contracdiction there cause concurrency-utilities doesn't integrate with CDI at all but we can also see it the opposite way: CDI doesn't provide any way to propagate a context in another thread. Both point of view are valid so we need to see where we tackle it.



On Mar 6, 2016, at 12:35 PM, Romain Manni-Bucau < [hidden email] > wrote:




does https://gist.github.com/rmannibucau/d55fce47b001185dca3e help?

Idea is to give an API to make:

public void complete () {
        try {
        asyncContext . complete();
        } finally {
        auditContext . end();
        }
        }

working without hacky and almost impossible context pushing (cause of injections nature you are not supposed to know what to push in the context when going async).



Romain Manni-Bucau
@rmannibucau | Blog | Github | LinkedIn | Tomitriber

2016-03-06 16:40 GMT+01:00 Reza Rahman < [hidden email] > :



Can you kindly share an annotated code example of the proposed solution so we can all follow and discuss this?

On Mar 6, 2016, at 9:31 AM, Romain Manni-Bucau < [hidden email] > wroteshar:




Hi guys,

spoke on concurrency utilities about the ability to inherit a cdi scope. Idea is to follow request scope more than cdi spec allows. First thought it was a concurrency utilities thing but Reza mentionned can be a CDI one so here it is.

Sample:
In a servlet i get MyBean which is @RequestScoped, I do some set on it. The i go async (AsyncContext) and trigger a task in another thread. It would be neat - and mandatory in some case by the loose coupling nature of CDI - to get the *same* MyBean *instance* in this thread. With a direct dependency you can easily use message passing pattern - but you loose the loose coupling cause you need to know until which level you unwrap, think t principal case which has 2-3 proxies!. However in practice you have a lot of undirect dependencies, in particular with enterprise concerns (auditing, security...) so you can't really do it easily/naturally.

Bonus:
One very verbose way is to be able to kind of push/pop an existing context in a thread - wrappers doing it on a Runnable/Consumer/Function/... would be neat.

Question:
Would CDI handle it in 2.0?

Side note: this is really about the fact to reuse a "context context" (its current instances map) in another thread the more transparently possible and match the user vision more than a technical question for now.


Romain Manni-Bucau
@rmannibucau | Blog | Github | LinkedIn | Tomitriber



_______________________________________________
cdi-dev mailing list
[hidden email]
https://lists.jboss.org/mailman/listinfo/cdi-dev 

Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 ( http://www.apache.org/licenses/LICENSE-2.0.html ). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.

_______________________________________________
cdi-dev mailing list
[hidden email]
https://lists.jboss.org/mailman/listinfo/cdi-dev 

Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 ( http://www.apache.org/licenses/LICENSE-2.0.html ). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.


_______________________________________________
cdi-dev mailing list
[hidden email]
https://lists.jboss.org/mailman/listinfo/cdi-dev 

Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 ( http://www.apache.org/licenses/LICENSE-2.0.html ). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.


_______________________________________________
cdi-dev mailing list
[hidden email]
https://lists.jboss.org/mailman/listinfo/cdi-dev

Note that for all code provided on this list, the provider licenses the code under the Apache License, Version 2 (http://www.apache.org/licenses/LICENSE-2.0.html). For all other ideas provided on this list, the provider waives all patent and other intellectual property rights inherent in such information.
123