In my Spring Boot WebFlux application, I am using Reactor. I have set spring.sleuth.reactor.instrumentation-type=manual and am using #ContinueSpan on a service method. I see from the code that the span is created, started and ended appropriately in a reactive way.
Later, in the Flux I need to extract the trace information.
The code I am using injects Tracer and I use Tracer.currentSpan().context() to try to get to the trace context. I can successfully get the currentSpan() until I hit a call to an R2DBC repository method, after which the currentSpan is null. I can make the span "current" by annotating the repository method, but I do not want to do that unless it's necessary. I'd like to understand the underlying "problem".
I have also looked at CurrentTraceContext in the Reactor Subscription Context and see that it answers as Tracer does. Supporting both seems to be a threadlocal-ly supported trace context.
Oddly, if I look at TraceContext in the Subscriber context, the trace, parent, and span ids are there. It appears that WebFluxSleuthOperators.currentTraceContext(Context...) does this - so I have to believe that this is the appropriate vehicle for obtaining the trace context.
So, a few questions:
Is WebFluxSleuthOperators.currentTraceContext(Context...) [TraceContext in the Reactor Subscription context] the proper way to get the up-to-date trace context?
Looking at ReactorSleuthMethodInvocationProcessor, a reference to the current span and trace context are given to the SpanSubscriber which puts them in the Subscriber context. As mentioned earlier, subscribe() and next() are invoked within the context of that span. Why would a method such as a call to a r2dbc repository method effectively "erase" the tracer.currentSpan() but leave the trace context alone?
I'd love to understand this a bit more deeply and will look at the source more. But any insight right now is greatly appreciated. Thank you very much in advance.
To access the current span in a Reactor flow in Spring, use the Span or TraceContext found in the Subscriber Context.
Mono.deferContextual(contextView -> Mono.just(contextView.get(TraceContext.class)))
Or, better yet, use WebFluxSleuthOperators.currentTraceContext(Context.of(contextView)))
Accessing the span through the Tracer bean may not produce the current span - as the current thread may be different than the one that originated the span and brought it into scope. This was confirmed through debugging.
Related
I was wondering if there is some way to provide two methods to Micronaut which are guaranteed to run before and after a request was passed to the handler.
In my case, this would be to initialize some thread-local data.
I know this would also be possible to put in the handler itself but putting the same lines of code in every handler isn't really the greatest solution.
Use a filter - see the docs at https://docs.micronaut.io/latest/guide/#filters
Recently, I am studying the problem of task scheduling in Flink. My purpose is to schedule subtasks to a slot of the specified node according to my own needs by modifying some source codes of the scheduling part. Through remote debugging and checking the source code, I found the following method call stack, most of which I can't understand (the comments are a little less), especially in this method: org.apache.flink.runtime.jobmaster.slotpool.SchedulerImpl#allocateMultiTaskSlot. I guess the code that allocates slots to tasks is around here. Because it is too difficult to read the source code, I have to ask you for help. Of course, if there is a better way to achieve my needs, please specify one or two. Sincerely look forward to your reply! Thank you very much!!!
The method call stack is as follows(The version of Flink I use is 1.11.1):
org.apache.flink.runtime.jobmaster.JobMaster#startJobExecution
org.apache.flink.runtime.jobmaster.JobMaster#resetAndStartScheduler
org.apache.flink.runtime.jobmaster.JobMaster#startScheduling
org.apache.flink.runtime.scheduler.SchedulerBase#startScheduling
org.apache.flink.runtime.scheduler.DefaultScheduler#startSchedulingInternal
org.apache.flink.runtime.scheduler.strategy.EagerSchedulingStrategy#startScheduling
(This is like the method call chain of PipelinedRegionSchedulingStrategy class. In order to simply write it as the method call chain of EagerSchedulingStrategy class, it should have no effect)
org.apache.flink.runtime.scheduler.strategy.EagerSchedulingStrategy#allocateSlotsAndDeploy
org.apache.flink.runtime.scheduler.DefaultScheduler#allocateSlotsAndDeploy
org.apache.flink.runtime.scheduler.DefaultScheduler#allocateSlots
org.apache.flink.runtime.scheduler.DefaultExecutionSlotAllocator#allocateSlotsFor
org.apache.flink.runtime.executiongraph.SlotProviderStrategy.NormalSlotProviderStrategy#allocateSlot
org.apache.flink.runtime.jobmaster.slotpool.SchedulerImpl#allocateSlot
org.apache.flink.runtime.jobmaster.slotpool.SchedulerImpl#allocateSlotInternal
org.apache.flink.runtime.jobmaster.slotpool.SchedulerImpl#internalAllocateSlot
org.apache.flink.runtime.jobmaster.slotpool.SchedulerImpl#allocateSharedSlot
org.apache.flink.runtime.jobmaster.slotpool.SchedulerImpl#allocateMultiTaskSlot
(I feel that this is the key to allocate slot for subtask, that is, execution vertex, but there is no comment, and I don't understand the process idea, so I can't understand it.)
In an integration test, my advice of #AfterReturning is wrongly executed while in the test I mock it to throw TimeoutException, and the arg passed to the aspect is null.
My advice:
#AfterReturning("execution(* xxxxxx" +
"OrderKafkaProducerService.sendOrderPaidMessage(..)) && " +
"args(order)")
public void orderComplete(CheckoutOrder order) { // order is null when debugging
metricService.orderPaidKafkaSent();
log.trace("Counter inc: order paid kafka"); // this line of log is shown in console
metricService.orderCompleted();
log.trace("Order complete! {}", order.getId()); // this line is not, because NPE
}
And my test:
// mocking
doThrow(new ServiceException(new TimeoutException("A timeout occurred"), FAILED_PRODUCING_ORDER_MESSAGE))
.when(orderKafkaProducerService).sendOrderPaidMessage(any()); // this is where advice is executed, which is wrong
...
// when
(API call with RestAssured, launch a real HTTP call to endpoint; service is called during this process)
// then
verify(orderKafkaProducerService).sendOrderPaidMessage(any(CheckoutOrder.class)); // it should be called
verify(metricService, never()).orderCompleted(); // but we are throwing, not returning, we should not see this advice executed
This test is failing because of NPE(order is null).
In debugging, I find that when I was mocking, I already execute the advice, and at this point, any() has no value yet, is null, so NPE. But I don't think the advice should execute while mocking. How can I avoid that while testing?? This is absurd for me.
Currently Spring test support does not explicitly handle the situation that an injected mock or spy (which is a proxy subclass via Mockito) might actually be an AOP target later on (i.e. proxied and thus subclassed again via CGLIB).
There are several bug tickets related to this topic for Spring, Spring Boot and Mockito. Nobody has done anything about it yet. I do understand why the Mockito maintainers won't include Spring-specific stuff into their code base, but I do not understand why the Spring people do not improve their testing tools.
Actually when debugging your failing test and inspecting kafkaService, you will find out the following facts:
kafkaService.getClass() is com.example.template.services.KafkaService$MockitoMock$92961867$$EnhancerBySpringCGLIB$$8fc4fe95
kafkaService.getClass().getSuperclass() is com.example.template.services.KafkaService$MockitoMock$92961867
kafkaService.getClass().getSuperclass().getSuperclass() is class com.example.template.services.KafkaService
In other words:
kafkaService is a CGLIB Spring AOP proxy.
The AOP proxy wraps a Mockito spy (probably a ByteBuddy proxy).
The Mockito spy wraps the original object.
Besides, changing the wrapping order to make the Mockito spy the outermost object would not work because CGLIB deliberately makes its overriding methods final, i.e. you cannot extend and override them again. If Mockito was just as restrictive, the hierarchical wrapping would not work at all.
Anyway, what can you do?
Either you use a sophisticated approach like described in this tutorial
or you go for the cheap solution to explicitly unwrap an AOP proxy via AopTestUtils.getTargetObject(Object). You can call this method safely because if the passed candidate object is not a Spring proxy (internally easy to identify because it implements the Advised interface which also gives access to the target object), it just returns the passed object again.
In your case the latter solution would look like this:
#Test
void shouldCompleteHappyPath() {
// fetch spy bean by unwrapping the AOP proxy, if any
KafkaService kafkaServiceSpy = AopTestUtils.getTargetObject(kafkaService);
// given mocked
doNothing().when(kafkaServiceSpy).send(ArgumentMatchers.any());
// when (real request)
testClient.create().expectStatus().isOk();
// then
verify(kafkaServiceSpy).send(ArgumentMatchers.any());
verify(metricService).incKafka();
}
This has the effect that when(kafkaServiceSpy).send(ArgumentMatchers.any()) no longer triggers the aspect advice because kafkaServiceSpy is no longer an AOP proxy. The auto-wired bean kafkaService still is, though, thus AOP gets triggered as expected, but no longer unwantedly while recording the mock interaction.
Actually, for the verification you could even use kafkaService instead of the spy and only unwrap the spy when recording the interaction you want to verify later:
#Test
void shouldCompleteHappyPath() {
// given mocked
doNothing()
.when(
// fetch spy bean by unwrapping the AOP proxy, if any
AopTestUtils.<KafkaService>getTargetObject(kafkaService)
)
.send(ArgumentMatchers.any());
// when(real request)
testClient.create().expectStatus().isOk();
// then
verify(kafkaService).send(ArgumentMatchers.any());
verify(metricService).incKafka();
}
P.S.: Without your MCVE I would never have been able to debug this and find out what the heck was going on. This proves again that asking questions including an MCVE is the best thing you can do for yourself because it helps you get answers to questions which otherwise probably would remain unanswered.
Update: After I had mentioned this problem under the similar closed issue Spring Boot #6871, one of the maintainers has by himself created Spring Boot #22281 which specifically addresses your problem here. You might want to watch the new issue in order to find out if/when it can be fixed.
I don't understand the GTask functionality? why do I need this?
In my mind it is like callback.. you set a callback to a source in some context and this callback is then called when event is happening.
In general, i'm a bit confused about what is a Context and a Task in GLib and why do we need them.
In my understanding there is a main loop (only 1?) that can run several contexts (what is a context?) and each context is related to several sources which in their turn have callbacks that are like handlers.
So can someone please make some sense for me in it all.
I don't understand the GTask functionality? why do I need this? In my mind it is like callback.. you set a callback to a source in some context and this callback is then called when event is happening.
The main functionality GTask exposes is easily and safely running a task in a thread and returning the result back to the main thread.
In general, i'm a bit confused about what is a Context and a Task in GLib and why do we need them. In my understanding there is a main loop (only 1?) that can run several contexts (what is a context?) and each context is related to several sources which in their turn have callbacks that are like handlers.
For simplicity I think its safe to consider contexts and loops the same thing and there can be multiple of them. So in order to be thread-safe the task must know which context the result is returned to.
Firstly i would like to apologize if i could not find anything about what i would like to describe that really solved my problems. This does not mean that i fully searched in the site. Although i have been spending too much time (days). I am also new on here (in the sense that i never wrote/replied to SO users). And i am sorry for my possible english errors.
I have to say i am new to Java EE.
I am working on WildFly 14, using MySQL.
I am now focusing on a JPA problem.
I have a uniqueness constraint. I am doing tests and while performing the uniqueness violation test, from the data source level i get a MySQLIntegrityConstraintViolationException, and that's ok. I have the problem in that the persist() method does not let me catch the exception (i even put Throwable in the clause, but nothing..). I strongly, strictly, need to catch that, in order to manage a crucial procedure (that, indirectly contains the call to .remove()) in my work's code.
By the way, trying to write that exception, the system does not show me the window of the suggested classes/annotations/etc, suggesting me just to create the class "MySQLIntegrityConstraintViolationException". Doesn't working on WildFly, using MySQL, suffice, for having the suggestions?
Not finding the solution, i decided to change: instead of using persist(), i decided to use .createNativeQuery() in which i put as parameter a String describing an insertion in the db. It seems working. Indeed it works (signals uniqueness violation (ok!), does not execute the TRY block code (ok!) and goes into CATCH block (ok!)). But, again, the exception/error is not clear.
Also, when in the code i enter the piece of code that is in charge of managing the catching and then executing what's inside (and i have a .remove(), inside), it raises the exception:
"Transaction is required to perform this operation (either use a transaction or extended persistence context)" --> this referring to my entityManager.remove() execution..
Now i cannot understand.. should not JPA/JTA manage automatically the transactions?
Moreover, trying, later, to put entityManager.getTransaction().begin() (and commit()), it gives me the problem of having tried to manage manually transactions when instead i couldn't.. seems an endless loop..
[edit]: i am working in CMT context, so i am allowed to work with just EntityManager and EntityManagerFactory. I have tried with entityManager.getTransaction().begin() and entityManager.getTransaction().commit() and it hasn't worked.
[edit']: .getTransaction (EntityTransaction object) cannot be used in CMT context, for this reason that didn't work.
[edit'']: i have solved the transaction issue, by means of the transaction management suited for the CMT context: JTA + CMT requires us to manage the transactions with a TRY-CATCH-FINALLY block, in whose TRY body it is needed to put the operation we want to perform on the database and in whose FINALLY body it is needed to put the EntityManager object closing (em.close()). Though, as explained above, i have used em.createNativeQuery(), that, when failing, throws catchable (catchable in my app) exceptions; i would really need to do a roll-back (usage of .createNativeQuery() is temporary) in my work code and use the .persist() method, so i need to know what to do in order to be able to catch that MySQLIntegrityConstraintViolationException.
Thanks so much!
IT SEEMS i have solved the problem.
Rolling back to the use of .persist() (so, discarding createNativeQuery()), putting em.flush() JUST AFTER em.persist(my_entity_object) has helped me, in that, once the uniqueness constraint is violated (see above), the raised exception is now catchable. With the catchable exception, I can now do as described at the beginning of the post.
WARNING: I remind you of the fact that i am new to JavaEE-JPA-JTA. I have been "lucky" because, since my lack of knowledge, i put that instruction (em.flush()) by taking a guess (i don't know how i could think of that). Hence, I would not be able to explain the behaviour; I would appreciate, though, any explanation of what could have happen, of how and when the method flush() is used, and so on and so forth..
Thanks!