Thanks to aphyr for doing this sort of testing. It is important that we not only verify etcd with our own internal testing but also have dedicated third party feedback. From the beginning we have wanted something that was simple and worked correctly.<p>It is also great to see etcd showing up in lots of interesting projects like skydns and kubernetes. I think we have built something that is not just a great building block for CoreOS but the OSS community at large.<p>Thanks to everyone who has helped get the project to where it is today; there is a bright future ahead.
I think the most surprising fact, despite all the amazing work here, is that Comcast helps pay for this kind of research!<p>Other then that, it's great that we are getting some good results after serious vetting for more modern replacements to ZK. I may not like ZK's crufty-ness, but I could always trust it. Now with these results, I am seriously going to consider Consul as a replacement.
This sounds like a report of a bug, but I believe this is not the actual story. It is more a report of a design tradeoff: the authors of those CP systems completely understand what happens, but were not happy to pay this performance price for reads. One thing is to have a data store that has a very limited performance in write operations but is very fast when you need to read, another thing is a data store where both writes and reads are very slow. However once you read potentially stale data from nodes, many of the advantages of having a CP system are gone. IMHO to revert those systems to a default where reads are applied to the state machine like writes is the sanest thing to do, even if options to potentially read stale reads are also useful in some context.
Wow. I thought of etcd as taking consistency very seriously. Aphyr's discoveries are indeed quite surprising: "The very first test I ran with reported a linearizability failure. I was so surprised I spent another week double-checking Knossos and Jepsen, then writing my own etcd client, to make sure I hadn’t made a mistake. Sure enough, etcd’s registers are not linearizable."
Small pet peeve, please for the love of god and people stop with the "call me maybe" titles... they aren't cute, funny, or anything but a waste of characters. By the title alone, I assume the content to be juvenile (despite already knowing the content of _that_ blog is anything but).