SleeperSmith commented on issue #572: URL: https://github.com/apache/pulsar/issues/572#issuecomment-782609950
@kjaggann split brain is a situation where the minority "server team" accepts write/read command and leads to a situation where the two clusters becomes inconsistent (in terms of data). It does not occur trivially from just 2 servers being cut off (say AWS AZ connection outage) from the rest of the cluster. That's a clear cut case of minority / not holding the quorum. One potential scenario is when servers have partial connections to rest of the cluster. Say servers of ABCDE. C can talk to the whole cluster, but AB can't talk to DE. AB and DE then both elect a leader amongst themselves and function independently thinking they both have majority with C. Even the above scenario does not cause split brain with ZK. ZK is a very robust and well tested software. Those nuances are already well tested and handled. Have a read of this https://aphyr.com/posts/291-jepsen-zookeeper and other related posts to understand more about consistency / quorum. You'll also see why managing all that stuff inside Pulsar ..... might not be the best idea. You are just reinventing the wheel twice, and it's a pretty fucking complex wheel. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected]
