Hi John,

Looking a little deeper, I see a remaining problem. I've got a test with three invocations:

  1. field and parameter violations

  2. no violations

  3. field and parameter violations

On the third invocation, the parameter violation is found by Resteasy, and then the EJB subsystem finds it again and throws a ConstaintViolationException. I unwrap the exception, collect the two exceptions (as I mentioned below - it turns out that org.hibernate.validator.internal.engine.ConstraintViolationImpl.hashCode() refers to the class in which the violation occurs, and I've got a plain class and a proxy class, so two essentially identical violations appear to not be equal), and send them back to the client. The problem is that, by throwing an exception, EJB is interrupting the normal flow and preventing org.jboss.resteasy.cdi.JaxrsInjectionTarget from finding field/property violations. That is, it's worse than just reporting duplicate violations - it's losing violations.

I can probably do some contorted thing to ignore the fact that EJB is throwing an exception in order to continue with the normal flow, but I'm thinking it would be much better to be able to turn off the validation in EJB. You were going to look into the log level of EJB exceptions. Have you seen anything about turning off validation? Or, if not, do you know someone who might know?

-Ron

Ron,

I have checked the fix in the https://github.com/ronsigal/Resteasy/tree/3.0_1749 branch and it does now pass the unit test and we see the correct response in the benchmark I am running. 

I still see the Exception being logged in the ejb3 subsystem as an ERROR, but I will investigate whether this should be logged as an error there, or whether a different log level would be more appropriate.

Thanks

John

On Sat, Nov 4, 2017 at 10:31 PM, Ron Sigal <rsigal@redhat.com> wrote:

Hi John,

Well, I thought I would have wrapped this up by now. I started by experimenting with @ApplicationScoped resources, but I just wasn't seeing what I expected. Then I started playing with @Stateless bean resources, and I still didn't see what I expected. That was all on the master branch. Today, I played with @Stateless resources on branch 3.0, and I finally reproduced what you've been seeing. It's vaguely familiar - EJB is doing its own validation in an interceptor. Anyway, I've updated GeneralValidatorImpl.checkForConstraintViolations(), and I've gotten rid of the 500 status by unwinding thrown Exceptions until I get a ConstraintViolationException. One problem I still have is that I'm getting two copies of the violations when I combine a couple of sources, even though there is an equals() method that I think should prevent that.

Anyway, I'm telling you now about a not quite complete solution because I'm going to take off next week. The solution is on branch 3.0_1749 in my personal github repository: https://github.com/ronsigal/Resteasy/tree/3.0_1749.

I'll clean things up when I get back. If you get a chance, let me know if the updated version fixes your problem.

-Ron

On 11/02/2017 03:21 AM, John O'Hara wrote:

Hi,

Does the test need to be in its own class, or can I add it to an existing class?  Should the test be in the integration testsuite, or is there a better module to put it in? org.jboss.resteasy.test.cdi.validation.ValidationWithCDITest seems to me to be a logical place to put it.

Also, when I run the integration testsuite, it is run against an existing WF runtime, afaics the tests use the version of resteasy jaxrs that is bundled with the runtime, are there any tests that run with the build artefacts?

Thanks

John

On Tue, Oct 31, 2017 at 1:18 AM, Ron Sigal <rsigal@redhat.com> wrote:

Hi John,

Ok, I'll start playing with RESTEASY-1749. Thanks.

Yes, master branch.

-Ron

On 10/30/2017 07:30 AM, John O'Hara wrote:

Ron,

I agree with your thinking. I think that we do need to check the parameters when the endpoint is called, and throw an appropriate exception.

I have opened a jira : https://issues.jboss.org/projects/RESTEASY/issues/RESTEASY-1749

I can create a test in the testsuite, would this need to be in the current master branch?

John

On Fri, Oct 27, 2017 at 5:21 PM, Ron Sigal <rsigal@redhat.com> wrote:

On 10/27/2017 08:33 AM, John O'Hara wrote:

After some more digging, I have found that this issue occurs if the the rest endpoint is also defined as an ejb managed bean. e.g.

@Stateless
@Path("/policyholder")
@Produces(MediaType.APPLICATION_JSON)
@Consumes(MediaType.APPLICATION_JSON)
public class PolicyHolderEndpoint {

In this case the JaxrsInjectionTarget.inject() is only called on the first endpoint invocation, and not on subsequent calls.

Hmmm. @Stateless, so it doesn't have @Request scope; i.e., it's created just once. I conjecture you wouldn't see this problem for a @Stateful session bean.

In the application the pusdo PolicyHolderEndpoint class is also a managed bean that is injected into other services as a single entry point for managing policyHolders.

Is it reasonable to expect a rest endpoint can also be defined as an ejb managed bean?

According to the JAX-RS spec:

10.2.4 Enterprise Java Beans (EJBs)

In a product that supports EJBs, an implementation MUST support the use of stateless and singleton ses- sion beans as root resource classes, providers and Application subclasses. JAX-RS annotations can be applied to methods in an EJB’s local interface or directly to methods in a no-interface EJB. Resource class annotations (like @Path ) MUST be applied to an EJB’s class directly following the annotation inheritance rules defined in Section 3.6.

I need to look into why weld isn't injecting the JaxrsInjectionTarget, but I think that my original question still stands in this use case.  If we are running in a CDI context, we already ascertain in MethodInjectorImpl.invoke() whether there are validation exceptions by calling [1]. Can we fail at this point, rather than delegating the validation to the CDI manager and waiting for a callback?

The CDI spec says we have to accumulate as many violations as possible before throwing an exception,so, in general, we still have to proceed to check field / property violations. BUT ... I think the scope is the source of the problem. If an EJB session bean is created just once, then field / parameter injections should be checked only once. On subsequent calls, it does, as you say, make sense to stop and throw an exception after checking parameters. SO ... if my reasoning makes sense, this is something that should be handled in Resteasy. If you agree, do you want to create a JIRA and I'll work on it? What do you think? -Ron

Thanks

[1] - https://github.com/resteasy/Resteasy/blob/master/resteasy-jaxrs/src/main/java/org/jboss/resteasy/core/MethodInjectorImpl.java#L119

On Fri, Oct 27, 2017 at 8:21 AM, John O'Hara <johara@redhat.com> wrote:

Ron,

I re-read your email and double checked the call to GeneralValidatorImpl.checkViolationsfromCDI().

I only see this being called once, on the first request.

I will dig some more to see why it is not being called on all requests

Thanks

John

On Fri, Oct 27, 2017 at 8:03 AM, John O'Hara <johara@redhat.com> wrote:

Thank you for your responses. If I fill you in on a bit of background about what I have experienced, then it might help clarify the issue I see.  

Background

I am running a JEE benchmark on EAP7.1 (using resteasy 3.0.24.Final). One of rest endpoints has a @Valid annotation on a method parameter for a complex object.  

@POST

@Path("/foo")

@Consumes(MediaType.APPLICATION_JSON)

public void addFoo(@Valid Foo bar) throws Exception {

The Foo class contains different validation annotations, such as @NotNull, @Size, @Pattern and a custom validation annotation.

One test determines how WF/EAP handles invalid requests. When I invoke the endpoint above I see a non-deterministic response from the app server.  If I invoke the endpoint with an invalid json object I receive a HTTP 400 exception as expected, but if I invoke the endpoint with a valid object, and *then* invoke the endpoint with an invalid object I receive a HTTP 500 error.

This exception only occurs on some property validation annotations and not all (e.g. @NotNull works as expected)

The test is being run in the context of CDI, and the HTTP 500 error is coming from the ejb3 subsystem. The ejb3 subsystem is wrapping the ConstraintValidationException raised by hibernate-validator.

Observations

In the call stack I capture [1], and WF thread dump [2] I can see;

1. The test is run in the context of CDI. The HTTP 500 response code message is caused by the ejb3 subsystem wrapping a ConstraintValidationException raised by hibernate-validator

2. The validator is not called in ResourceMethodInvoker.invokeOnTarget() [3] as isValidatable is false [4], a ResteasyViolationException is not thrown

3. The request is validated again here [5], the validation exceptions are not acted upon in this method call and the reflected method (which happens to be the Foo class constructor) is called here [6]

Questions

1. Why is ResourceMethodInvoker.isValidatable false in WF/EAP?

2. If isValidatable is false, and validation occurs again at MethodInjectorImpl.invoke() can we not check for validation errors and raise the correct exception there as well?

I have a standalone test that I will clean up and push out. There are some aspects of the test that can not be published publicly, so I need to modify them first.

Thanks,

John

[1] - see failing_4.stack  

[2] - see WF_stackTrace.out

[3] - https://github.com/resteasy/Resteasy/blob/master/resteasy-jaxrs/src/main/java/org/jboss/resteasy/core/ResourceMethodInvoker.java#L305

[4] - https://github.com/resteasy/Resteasy/blob/master/providers/resteasy-validator-provider-11/src/main/java/org/jboss/resteasy/plugins/validation/GeneralValidatorImpl.java#L223

[5] - https://github.com/resteasy/Resteasy/blob/master/resteasy-jaxrs/src/main/java/org/jboss/resteasy/core/MethodInjectorImpl.java#L119

[6] - https://github.com/resteasy/Resteasy/blob/master/resteasy-jaxrs/src/main/java/org/jboss/resteasy/core/MethodInjectorImpl.java#L140

--

JOHN O'HARA

PRINCIPAL SOFTWARE ENGINEER

Red Hat

TRIED. TESTED. TRUSTED.

@redhatway   @redhatinc   @redhatsnaps

--

JOHN O'HARA

PRINCIPAL SOFTWARE ENGINEER

Red Hat

TRIED. TESTED. TRUSTED.

@redhatway   @redhatinc   @redhatsnaps

_______________________________________________
resteasy-dev mailing list
resteasy-dev@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/resteasy-dev

-- 
My company's smarter than your company (unless you work for Red Hat)

_______________________________________________ resteasy-dev mailing list resteasy-dev@lists.jboss.org https://lists.jboss.org/mailman/listinfo/resteasy-dev

--

JOHN O'HARA

PRINCIPAL SOFTWARE ENGINEER

Red Hat

TRIED. TESTED. TRUSTED.

@redhatway   @redhatinc   @redhatsnaps

_______________________________________________
resteasy-dev mailing list
resteasy-dev@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/resteasy-dev

-- 
My company's smarter than your company (unless you work for Red Hat)

_______________________________________________ resteasy-dev mailing list resteasy-dev@lists.jboss.org https://lists.jboss.org/mailman/listinfo/resteasy-dev

--

JOHN O'HARA

PRINCIPAL SOFTWARE ENGINEER

Red Hat

TRIED. TESTED. TRUSTED.

@redhatway   @redhatinc   @redhatsnaps

_______________________________________________
resteasy-dev mailing list
resteasy-dev@lists.jboss.org
https://lists.jboss.org/mailman/listinfo/resteasy-dev

-- 
My company's smarter than your company (unless you work for Red Hat)

_______________________________________________ resteasy-dev mailing list resteasy-dev@lists.jboss.org https://lists.jboss.org/mailman/listinfo/resteasy-dev

--

JOHN O'HARA

PRINCIPAL SOFTWARE ENGINEER

Red Hat

TRIED. TESTED. TRUSTED.

@redhatway   @redhatinc   @redhatsnaps

-- 
My company's smarter than your company (unless you work for Red Hat)