Re: Clustering a DDD/CQRS/ES app
Greg Young <[email protected]>
| Newsgroups | gmane.comp.programming.domain-driven-design |
|---|---|
| Message-ID | <CAC9RQtifVXUQKbAGoHPBDQVfi4UzBg5vUKUj=HaQ=iBkrY+eCA@mail.gmail.com> |
Deadlocking comes in when you write on two servers to the same stream. On Mon, Apr 8, 2013 at 12:11 PM, Rickard Öberg <[email protected]>wrote: > On 4/8/13 16:40 , Greg Young wrote: > >> The model is most similar to >> http://readwrite.com/2011/02/**10/megastore-googles-answer-**to-no<http://readwrite.com/2011/02/10/megastore-googles-answer-to-no>. There >> are some pros and cons to the model. I think if you go more than 3 >> servers you will eventually build a paxos state machine. >> > > Since I'm using Hazelcast I think it does that under the hood. > > BTW, I've spent the last year building a Paxos impl. Challenging as hell. > Distributed state machines with asynch messages; very different from > anything else. > > > Pros: >> scalability >> consistent >> allows writing to local node >> > > Which is teh awesome. > > > Cons: >> can not write very many events/second to any given stream (they will >> start deadlocking). You could get around that by mapping messages for a >> given stream to a single node but that removes #2 above and limits #1. >> > > For apps needing clustering but not necessarily massive amounts of > instances, it should be pretty ok. Which is most of us, I guess. > > Where does the deadlocking come in? > > /Roclard > > -- Le doute n'est pas une condition agréable, mais la certitude est absurde.