Fix saga and timeout ATs #4176

SzymonPobiega · 2016-09-29T09:23:54Z

Fixing the saga finder test
Fixing the condition in the timeout test

Resolves #4178

danielmarbach · 2016-09-29T10:29:49Z

src/NServiceBus.AcceptanceTests/Routing/SubscriptionBehavior.cs


-    class SubscriptionBehavior<TContext> : IBehavior<IIncomingPhysicalMessageContext, IIncomingPhysicalMessageContext> where TContext : ScenarioContext
+    class SubscriptionBehavior<TContext> : Behavior<IIncomingPhysicalMessageContext> where TContext : ScenarioContext


Try not to use the base class. Everything internal even the ATTs should use IBehavior, that class is now more of a customer focused convenience class

danielmarbach · 2016-09-29T10:30:06Z

Retargeted against develop

andreasohlund · 2016-09-30T08:38:56Z

src/NServiceBus.AcceptanceTests/Routing/SubscriptionBehavior.cs

+            var retries = 0;
+            var succeeded = false;
+            Exception lastError = null;
+            while (retries < maxRetries && !succeeded)


Isn't this retrying all messages? (shouldn't we only do this if the current message is a subscribe/unsubscribe message)

SzymonPobiega · 2016-10-03T07:03:50Z

@andreasohlund @danielmarbach I've fixed the problems you found and also updated the external timeout manage test removing the dependency on timing.

danielmarbach · 2016-10-03T07:17:45Z

@SzymonPobiega did you forget to save the proj? Build seems to fail

danielmarbach · 2016-10-03T07:18:34Z

src/NServiceBus.AcceptanceTests/Routing/SubscriptionBehavior.cs


-    class SubscriptionBehavior<TContext> : IBehavior<IIncomingPhysicalMessageContext, IIncomingPhysicalMessageContext> where TContext : ScenarioContext
+    class SubscriptionBehavior<TContext> : IBehavior<IIncomingPhysicalMessageContext> where TContext : ScenarioContext


It should be IBehavior<IIncomingPhysicalMessageContext, IIncomingPhysicalMessageContext>

SzymonPobiega · 2016-10-03T07:25:13Z

@danielmarbach nope. But I missed a compiler warning

SzymonPobiega · 2016-10-03T07:28:46Z

OK, not should be good. I used not that IBehavior. BTW, why can't I use Behavior?

danielmarbach · 2016-10-03T08:46:56Z

@SzymonPobiega for ATTs it does not matter that much. For production code internally it does because the behavior base class creates a new closure on each invocation call. We weren't able to refactor it without introducing breaking changes for 6.0

danielmarbach · 2016-10-03T08:48:59Z

@andreasohlund LGTM

timbussmann · 2016-10-03T09:14:57Z

src/NServiceBus.AcceptanceTests/Routing/SubscriptionBehaviorExtensions.cs

-                return new SubscriptionBehavior<TContext>(action, context, MessageIntentEnum.Subscribe);
-            }));
+                return new SubscriptionBehavior<TContext>(action, context, builder.Build<CriticalError>(), MessageIntentEnum.Subscribe);
+            }, DependencyLifecycle.InstancePerCall));


why did you add this? Afaik behaviors are cached anyway, so instance per call will not result in what you would expect when defining it?

timbussmann · 2016-10-03T09:16:20Z

what's the reason behind the retry logic? In what cases is this needed?

rest looks good to me.

SzymonPobiega · 2016-10-03T09:20:06Z

@timbussmann persisters assume race conditions in subscription store are resolved via FLR. I know this is a bad assumption to begin with but we have to live with it till we get rid of these persisters entirely.

andreasohlund · 2016-10-03T09:27:16Z

RavenDB is already doing this

https://github.com/Particular/NServiceBus.RavenDB/blob/develop/src/NServiceBus.RavenDB/Subscriptions/SubscriptionPersister.cs#L31

should we raise issues for both NH and ASP to do the same? (not blocking this PR)

timbussmann · 2016-10-03T10:29:45Z

should we raise issues for both NH and ASP to do the same?

I think this makes sense, otherwise this can hit anyone with disabled retries, so I'd say it's not a testing framework concern. I'm good with keeping it in case there are really tests encountering that race condition, but unless there is a real appearance I'd not pull that in and focus on the the subscription storages.

andreasohlund · 2016-10-04T17:49:26Z

@SzymonPobiega given that ravendb is good, where did you spot this? (inmemory? nh?)

SzymonPobiega · 2016-10-04T18:25:13Z

@andreasohlund NHibernate

andreasohlund · 2016-10-04T18:37:14Z

Should we consider add the inmem retries there instead of here?

SzymonPobiega · 2016-10-05T06:08:36Z

This would assume every persistence has this kind of feature. Should we have this assumption? I am fine with that. We just need to agree on this.

andreasohlund · 2016-10-05T06:10:46Z

I'd say yes, if there is potential race conditions the persister should try to mitigate that without relying on retries to be enabled in the endpoint?

danielmarbach · 2016-10-07T21:41:20Z

This discussion is similar to the discussion we had about ASB relying on retries when the broker connection goes boom. Should we hash this out?

andreasohlund · 2016-10-10T08:38:45Z

So should we fix NHibernate or merge this?

Or merge this, fix NH later, remove this?

danielmarbach · 2016-10-10T09:05:34Z

I'd say fix NHibernate and remove this

andreasohlund · 2016-10-10T09:12:30Z

I'd say fix NHibernate and remove this

@Particular/nhibernate-persistence-maintainers thoughts?

MarcinHoppe · 2016-10-10T09:27:55Z

I'm not sure:

@SzymonPobiega Will implementing the in-memory retries in subscription store alleviate the need for "wait until subscriber endpoint starts" in these ATTs?

At a minimum we need to carve out the fix to #4178 out of this PR before closing it. I'll be happy to do this.

timbussmann · 2016-10-10T10:49:41Z

(marking as wip to prevent it from merging at this point)

timbussmann · 2016-10-17T15:29:02Z

@SzymonPobiega @MarcinHoppe any updates on this?

MarcinHoppe · 2016-10-18T06:18:59Z

@timbussmann We have an issue and a PR in NHibernate to address. I chatted with @Scooletz and ASP does not suffer from this issue (similar to Raven).

/cc @SzymonPobiega

timbussmann · 2016-10-18T06:49:18Z

sounds like we can close this issue and handle this on the NH repo then?

SzymonPobiega · 2016-10-18T06:52:27Z

I think it is still valid to wait till endpoints are started before subscribing. What do you think? Worth a separate PR?

MarcinHoppe · 2016-10-18T06:53:23Z

👍 to what @SzymonPobiega said. There's also issue with the external timeout manager test. I will file a separate PR to fix this single test.

timbussmann · 2016-10-18T08:36:44Z

...ceBus.AcceptanceTests/Routing/MessageDrivenSubscriptions/When_subscribing_to_a_base_event.cs

@@ -18,7 +18,7 @@ public Task Both_base_and_specific_events_should_be_delivered()
                    await session.Publish(new SpecificEvent());
                    await session.Publish<IBaseEvent>();
                }))
-                .WithEndpoint<GeneralSubscriber>(b => b.When(async (session, c) => await session.Subscribe<IBaseEvent>()))
+                .WithEndpoint<GeneralSubscriber>(b => b.When(c => c.EndpointsStarted, async (session, c) => await session.Subscribe<IBaseEvent>()))


I don't think this when condition is necessary, as the ATT framework will execute whens already after endpoints have been started. See ScenarioRunner line 193.

SzymonPobiega · 2016-10-20T07:08:25Z

@timbussmann I removed all the unnecessar changes and left only two:

Fixing the saga finder test
Fixing the condition in the timeout test

timbussmann · 2016-10-20T07:19:06Z

LGTM

danielmarbach · 2016-10-21T06:47:47Z

@SzymonPobiega I moved the comment bullet points from here into the PR description is there anything else we need to update?

SzymonPobiega self-assigned this Sep 29, 2016

danielmarbach changed the base branch from release-6.0.0 to develop September 29, 2016 10:28

danielmarbach reviewed Sep 29, 2016

View reviewed changes

andreasohlund reviewed Sep 30, 2016

View reviewed changes

danielmarbach requested changes Oct 3, 2016

View reviewed changes

danielmarbach approved these changes Oct 3, 2016

View reviewed changes

timbussmann reviewed Oct 3, 2016

View reviewed changes

MarcinHoppe mentioned this pull request Oct 4, 2016

External timeout manager ATT has a race condition #4178

Closed

timbussmann changed the title ~~Fix message driven pub sub tests~~ [WIP] Fix message driven pub sub tests Oct 10, 2016

timbussmann reviewed Oct 18, 2016

View reviewed changes

Fix timeouts and saga finder tests.

7b3ba94

SzymonPobiega force-pushed the fix-md-pub-sub-tests branch from bbb8e59 to 7b3ba94 Compare October 20, 2016 07:07

SzymonPobiega changed the title ~~[WIP] Fix message driven pub sub tests~~ [WIP] Fix saga and timeout ATs Oct 20, 2016

SzymonPobiega changed the title ~~[WIP] Fix saga and timeout ATs~~ Fix saga and timeout ATs Oct 20, 2016

timbussmann approved these changes Oct 20, 2016

View reviewed changes

danielmarbach merged commit eeb28e8 into develop Oct 21, 2016

danielmarbach deleted the fix-md-pub-sub-tests branch October 21, 2016 06:47


		class SubscriptionBehavior<TContext> : IBehavior<IIncomingPhysicalMessageContext, IIncomingPhysicalMessageContext> where TContext : ScenarioContext
		class SubscriptionBehavior<TContext> : Behavior<IIncomingPhysicalMessageContext> where TContext : ScenarioContext

Fix saga and timeout ATs #4176

Fix saga and timeout ATs #4176

Conversation

SzymonPobiega commented Sep 29, 2016 • edited by danielmarbach

danielmarbach Sep 29, 2016

Choose a reason for hiding this comment

danielmarbach commented Sep 29, 2016

andreasohlund Sep 30, 2016

Choose a reason for hiding this comment

SzymonPobiega commented Oct 3, 2016

danielmarbach commented Oct 3, 2016

danielmarbach Oct 3, 2016

Choose a reason for hiding this comment

SzymonPobiega commented Oct 3, 2016

SzymonPobiega commented Oct 3, 2016

danielmarbach commented Oct 3, 2016

danielmarbach commented Oct 3, 2016

timbussmann Oct 3, 2016

Choose a reason for hiding this comment

timbussmann commented Oct 3, 2016

SzymonPobiega commented Oct 3, 2016

andreasohlund commented Oct 3, 2016

timbussmann commented Oct 3, 2016

andreasohlund commented Oct 4, 2016

SzymonPobiega commented Oct 4, 2016

andreasohlund commented Oct 4, 2016

SzymonPobiega commented Oct 5, 2016

andreasohlund commented Oct 5, 2016

danielmarbach commented Oct 7, 2016

andreasohlund commented Oct 10, 2016

danielmarbach commented Oct 10, 2016

andreasohlund commented Oct 10, 2016

MarcinHoppe commented Oct 10, 2016

timbussmann commented Oct 10, 2016

timbussmann commented Oct 17, 2016

MarcinHoppe commented Oct 18, 2016

timbussmann commented Oct 18, 2016

SzymonPobiega commented Oct 18, 2016

MarcinHoppe commented Oct 18, 2016

timbussmann Oct 18, 2016

Choose a reason for hiding this comment

SzymonPobiega commented Oct 20, 2016

timbussmann commented Oct 20, 2016

danielmarbach commented Oct 21, 2016

SzymonPobiega commented Sep 29, 2016 •

edited by danielmarbach