walrus (High Performance distributed log streaming engine)
7 points by rmoff
7 points by rmoff
How is it compared to kafka?
I don't believe Kafka ships with this https://github.com/nubskr/walrus#correctness
The cited client protocol doesn't seem to support any kind of seeking (server-managed offsets or client-managed offsets as with Kafka), but I'd have to dig further into what their mention of "shared cursor" actually means in practice
Separately, I've heard complaints about both Zookeeper and the JVM contributing to the heartache of running Kafka, which this sidesteps. I personally have experienced the ZK pain, but the JVM and I are good friends, so that one will depend a great deal on if Kafka is the only JVM application in ones stack
Vanlightly (Confluent)'s work seems more than credible.
I'm glad such a thing exists, I did say "ships with," and that leads to a related observation that any external repo almost certainly will have to be separately updated/tested since it's not integrated in apache/kakfka's tooling (that I could easily find, anyway). That external repo is almost like a Jepsen report in that way: valuable, but a point-in-time representation of the capabilities, right?
ZooKeeper is not required (actually not even supported) any more since Kafka 4.0. Kafka nodes themselves (with the "controller" role) take care of coordination in the cluster.