On 08/21/2013 08:35 AM, Bartosz Baranowski wrote:
----- Original Message -----
> From: "Stefano Maestri" <smaestri(a)redhat.com>
> To: wildfly-dev(a)lists.jboss.org
> Sent: Monday, August 19, 2013 3:09:41 PM
> Subject: Re: [wildfly-dev] JCA service/model names a.k.a WFLY-1656
> I've started to review this patch and at a first look can't work as is.
> It needs at a first look at least changes to resource-adapters xsd,
> because it bases service names on "id" attribute, but actually this
> attribute is not mandatory in xsd. In case you define a resource-adapter
> configuration in standalone.xml for a rar having an internal config with
> ironjacamar.xml It drives to runtime error like this:
> 14:47:27,561 ERROR [org.jboss.msc.service.fail] (MSC service thread 1-7)
> MSC000001: Failed to start service jboss.ra.deployment."ra16outij.rar":
> org.jboss.msc.service.StartException in service
> org.jboss.msc.service.DuplicateServiceException: Service
> jboss.ra.ra16outij is already registered
Thats expected error, since such configuration/setup leads to same resource being
spawned at runtime.
And it's exactly reason because it would be better to make
IMHO, since it's a perfectly valid configuration.
> Or even defining 2 resource-adapter in standaole xml on the same rar
> archive, w/o defining id as xsd permit:
> 14:49:59,574 ERROR [org.jboss.as.controller.management-operation]
> (ServerService Thread Pool -- 11) JBAS014613: Operation ("add") failed -
> address: ([
> ("subsystem" => "resource-adapters"),
> ("resource-adapter" => "ra16outij.rar")
> ]) - failure description: "JBAS014803: Duplicate resource [
> (\"subsystem\" => \"resource-adapters\"),
> (\"resource-adapter\" => \"ra16outij.rar\")
> Both this use case are normal and very frequent because resource-adapter
> defined in standalone.xml should be considered multiple configuration
> (activation in jca slang) on the same deployed rar.
Actually I wasnt able to grasp this to full extent - if this kind of anonymous resource
definition is a requirement from specs or something forged in IJ or AS7/8 itself.
Spec require multiple activation, ID is a concept added in AS. I have to
double check, but should be fine to make it mandatory.
I still have to
> cover all use cases, but those 2 at least needs to be fixed. Probably
> the easiest solution is to make "id" mandatory, but I have to double
> check it is sufficient for programmatic activations (explicit call of
> service like HornetQ is doing) and ironjacamar.xml ones.
If its not covered in testsuite or it did not make server explode, I might missed it.
I think it is covered by test suite, but only adding it by DMR command
(so no anonymous ID since resource name become ID). So in fact 2
anonymous RA isn't covered by test suite.
Anyway, again, making ID mandatory should fix it.
> If you agree I'll take the issue and I'll merge your patch on my branch
> and rework it a bit to cover all those use cases and possibly to solve
> also WFLY-1776 (the problem here could be that xsd changes can't be
> backported and it is a EAP case)
What I was led to believe is that this step is not required. It is perfectly
fine if user has "anonymous" activation, without id, as long as there is only
In case there is need for more than one activation, users need to provide proper
configuration, that is
Well, we should at least give a more expressive error. A service start
error could be confusing (in case of IJ.xml above for example), and
maybe easier to have ID mandatory.
WRT hornetq, if Im not wrong, each activation from hornetq subsystem
has unique name, since
each config entry has it unique( I may be off here though ).
Also, Im not sure about MDB annotation, this part of integration looked a bit fuzzy.
I'll take care to check them. I've already reassigned issue to me.
Existing tests pass, but Im not quite sure how this should look like for "IDed"
The point is that current tests is doing everything using DMR commands,
so they are always using ID (see above). Probably the most consistent
behaviour is to make ID mandatory. I'm going in this direction, even if
I need to double check something.
If you want to redo something feel free to mold it.
Oki, I'll ping you for a review. Thanks for interest and the patch.
> On 08/06/2013 03:29 PM, Bartosz Baranowski wrote:
>> Finally I was able ( I think ) to nail the problem and forge fix. There are
>> still some intermittent CI failures, but other than that CI looks clear.
>> While I polish last parts it would be good to get feedback on changes,
>> since impact on JCA integration is quite big.
>> Fix removes all static/leaky code from ConnectorServices ( add relevant
>> code to MSC service ), move away from static hashmap checks in favor of
>> MSC service present/deployment/dependency and remove ability to create
>> more than one activation for distinct model address.
>> Relevant issues:
>> wildfly-dev mailing list
> wildfly-dev mailing list