Hacker News Reader

Comment by lemevi | Hacker News Reader

Seems like the paper includes a pretty straightforward observation that CA just doesn't make much sense. How can you have consistency and availability if there's a partition? That's when you get the strong versus weak consistency or eventual consistency distinctions. If you look at the call me post by Aphyr consistency is a problem when there's a partition in a lot of real world software.

View on HN

That's actually the exact opposite of what he says. He says that CA can be vacuously satisfied by a system by simply going unavailable. It makes totally sense, but is a pathological behavior. It's a confusion that many people have about CAP, which I tried to clarify in a blog post[0] a couple months ago, in response to some other articles were going around at the time.

[0]: http://computationallyendowed.com/blog/2015/07/09/cap-theore...

Yep. A valid partition-intolerant system is one where nodes on both sides of a partition will automatically do something drastic, like unconditionally terminating, so that no subset of the nodes remains for clients to talk to.

The interesting thing about CA systems is that their behavior strongly resembles that of non-distributed systems. It's just like having one SPOF machine!

"CA can be vacuously satisfied by a system by simply going unavailable"

That kind of takes the A out of CA.

CA means you are giving up both consistency and availability if a partition happens (otherwise you'd be CP or AP). How is that at all useful in a distributed system?

>How can you have consistency and availability if there's a partition?

I don't get why people seem to have such a problem with the CA trade off, but seem to understand the CP or AP tradeoff. If you trade partition tolerance then have a partition your clients need to handle that just like they need to handle inconsistency, or unavailability if that was the trade off selected.

Can you give an example of a client handling a partition?

Reply on news.ycombinator.com