Ayende @ Rahien

Udi Dahan commented on A Messaging Saga

Wed, 09 Apr 2008 06:46:03 GMT

Andreas, Time is a business concept, as described in the rule: "verification process should take at most 14 days" and as such needs to be treated explicitly. The technological timeout that you were referring to is handled automatically by the infrastructure: You try to connect to the DB in your code, it timeouts, throws an exception, that bubbles up and breaks the message-level transaction causing the message to be returned to the queue and all other work rolling back. Once the message is in the queue again, some thread will pick it up and try to handle it - effectively retrying. Hope that makes sense.

Ayende Rahien commented on A Messaging Saga

Tue, 08 Apr 2008 15:10:29 GMT

Let us say that we have specified that a verification process should take 14 days. At the end of those 14 days, we have to decide what to do. In this case, I decided that I would assume the candidate is valid if at least one reference or certification came out okay, and no certification/reference came out invalid. Retrying would mean waiting another 14 days, and then what?

Ken Egozi commented on A Messaging Saga

Tue, 08 Apr 2008 14:51:38 GMT

I'd say, Go Udi !

Ayende Rahien commented on A Messaging Saga

Tue, 08 Apr 2008 14:38:36 GMT

Andrew, Using two containers sucks and should be avoided at all costs, period. NSB is using spring as a default container, but is not tied to the idea. You can replace that (you would need to write an adapter for the IBuilder interface, but it is very trivial thing, basically). When you do that, Windsor is responsible for handling IoC

Ayende Rahien commented on A Messaging Saga

Tue, 08 Apr 2008 14:36:16 GMT

Nick, Yes, I forgot to add it to the code. It is an auto property.

Ayende Rahien commented on A Messaging Saga

Tue, 08 Apr 2008 14:35:32 GMT

Yes, it is. How it is guaranteed is a bit complex, but it is.

Andreas commented on A Messaging Saga

Tue, 08 Apr 2008 14:35:15 GMT

Can you explain the Timeout concept a bit. I fail to figure out why you call Complete() in your timeout handler? Shouln't you do some kind of "retry" in case of a timeout?

Andrew commented on A Messaging Saga

Tue, 08 Apr 2008 10:49:58 GMT

Ayende, in this example, what mechanism do you use to inject the dependencies (IRepository) when writing handlers for nServiceBus. I have been using Windsor for over a year now on other projects and just picked up nServiceBus. I am trying to determine whether I swallow the pill and learn how to use Spring or somehow try to introduce WIndsor into my nServiceBus application. What do you think? Seems like having to maintain two separate IoC containers within the same app might lead to confusion and limitations. However, from my cursory review of Spring I am finding that I miss Windsor already. In this example, I assume that the nServiceBus plumbing is brokering the CandidateVerificationSaga, so I guess I would assume that it also injecting the IRepository dependency, is that correct?

Nick commented on A Messaging Saga

Tue, 08 Apr 2008 09:44:33 GMT

Hi Ayende, Looks like a nice alternative to WWF, as Ruurd suggested - much better than attempting to integrate WWF with a messaging technology, the two concepts seem to belong together. Just a quick question on the code - I could be missing it but there appears to be a Completed property that has no origin? Nick

Ken Egozi commented on A Messaging Saga

Tue, 08 Apr 2008 09:39:56 GMT

"The saga will handle only a single message at a point in time" is that guaranteed? otherwise stuff like answeredReferences+=1 might impose some problems

Ruurd Boeke commented on A Messaging Saga

Tue, 08 Apr 2008 08:23:24 GMT

loving it. It is what WF does, but WF persists a blob to the database, which quite frankly s*cks.