Re: Clustering a DDD/CQRS/ES app

Can I get a dirty read in the middle?

On Monday, April 8, 2013, Rickard Öberg wrote:

> On 4/8/13 16:04 , Greg Young wrote:
>
>> Hmm how do you handle commit protocol?
>>
>> Ex I talk to server 1 and say write aggregate a event 12
>>
>> You at same time are talking to server 3 and say same thing one of us
>> will get to server 2 (makes quorum) first.
>>
>
> When the events get into the Hazelcast atomic broadcast, which provides
> global ordering for consumers (i.e. the event stores on the instances in
> the cluster), they check if the new version of the aggregate already exists
> in the local eventstore. If yes we know there was a concurrent update, and
> the event is dropped. The originating server is notified, and the aggregate
> is dropped from the inmemory repository and re-read on next command.
>
> It's teh awesome. :-) Or am I missing something?
>
> /Rickard
>
>

-- 
Le doute n'est pas une condition agréable, mais la certitude est absurde.