jepsen

We have confirmed this both theoretically as well as by testing the overall system using the Jepsen test framework.

So, this issue is unrelated to the Jepsen network partitioning tests that is the subject of the original article.

From reading the code in Jepsen it looks like kill -9 is all that's being used to start failures.

Are there any plans to run @aphyr's jepsen CAP testing system on influx?

We've also been working with Kyle to understand the Jepsen codebase better.

RethinkDB does really well on the Jepsen tests in our internal testing.

Have you been testing this with Jepsen internally, out of curiosity?

Or is Jepsen doing something more sophisticated, like running each node in a VM with its own block device instead of sharing the host's filesystem?

Just read some of the posts in the Jepsen series.

When MemSQL is configured for synchronous durability and all databases are configured with synchronous replication, Jepsen confirms that writes to MemSQL are durable and acknowledged appropriately.

I wonder if this was sparked by the Jepsen thing on read commits.

Maybe it's not quite the same as how modern KV stores work, but it's still guaranteeing consistency across a network and is therefore a candidate for jepsen testing.

If you were to run this whole system under a tool like Jepsen, this would show up as "acknowledged but lost writes".

Fortunately, the author of Jepsen, Kyle Kingsbury/"Aphyr", has done the maths to back this up, demonstrating that RabbitMQ's clustering is both theoretically and practically unsound [1].

I thought it was a "Jepsen" device at first.

Remember, Jepsen isn't about proving that a database can't violate CAP, nor is it trying to say "all databases are crap." It's about verifying the marketing claims and then determining whether there are any mitigating strategies.

Jepsen is just doing a kill -9 on the java process.