> How much better is ZooKeeper at mitigating split brain than having a 3+ node cluster?
I'm not sure I could quantify the improvement. The goal with the change was to offload the leader election responsibilities from the brokers themselves to something that was better suited for those responsibilities (e.g. ZooKeeper). Leader election is not a simple problem to solve, and it's not really a core competency of a message broker so integrating with external providers makes sense. Furthermore, tying leader election to the broker meant you had to have a quorum of live brokers in a cluster (i.e. 3 or more) even if you just wanted a single, replicated HA pair. With the new design that's no longer the case. > Why would ZooKeeper be better than quorum voting at dealing with network issues that would cause split brain? Quorum voting isn't exactly the issue, per se. The real issue is leader election. ZooKeeper is purpose-built for this kind of work whereas the broker is not. > Are there any plans to drop the traditional Artemis quorum voting mechanism for the pluggable quorum provider configuration? I imagine it will be deprecated and phased out eventually, but there are no plans to do so at this time. > Any idea when it will become officially stable? As far as I understand it is ready to use now. I have no reason to think otherwise. Hope that helps. Justin On Wed, May 25, 2022 at 10:08 PM Steigerwald, Aaron <asteigerw...@brandesassociates.com.invalid> wrote: > Hello again Justin, > > Are there any plans to drop the traditional Artemis quorum voting > mechanism for the pluggable quorum provider configuration? Any idea when it > will become officially stable? > > Thank you, > Aaron Steigerwald > > -----Original Message----- > From: Steigerwald, Aaron > Sent: Friday, April 8, 2022 11:46 AM > To: users@activemq.apache.org > Subject: RE: [EXTERNAL]:Re: Journal corruption caused by split brain? > > Hello Justin, > > Thank you for the thorough explanation. > > How much better is ZooKeeper at mitigating split brain than having a 3+ > node cluster? Why would ZooKeeper be better than quorum voting at dealing > with network issues that would cause split brain? > > Thanks again, > Aaron > > -----Original Message----- > From: Justin Bertram <jbert...@apache.org> > Sent: Thursday, April 7, 2022 11:18 PM > To: users@activemq.apache.org > Subject: [EXTERNAL]:Re: Journal corruption caused by split brain? > > [CAUTION: This email originated from outside of the organization. Do not > click links or open attachments unless you recognize the sender and know > the content is safe.] ________________________________ > > > It's not clear exactly what is meant here by "corruption," but I think > it's probably the wrong word to describe the issues caused by split-brain. > The brokers should be able to read all the data in the journals no problem. > The actual problems are more related to potential duplicate consumption or > missed messages. > > Regarding duplicate consumption, consider 2 JMS consumers listening on a > queue on the master broker. Those consumers would be competing for the same > messages such that each message would only be consumed once (i.e. by either > consumer). However, once split brain occurs you could potentially have a > consumer on the same queue on *each* broker. In that case, they could each > receive the same message since they were no longer competing with each > other. > > Regarding missed messages, consider a non-durable JMS topic subscriber. > While it's connected to the master it receives every message sent to the > topic. However, once split brain occurs the producer might send messages to > the broker where the subscriber isn't connected which means it wouldn't get > those messages. > > In short, the data on each broker should stay 100% in-tact from a > technical stand-point. The problem is more in the realm of "irreconcilable > differences" for the applications connected to the brokers. > > The simplest way to mitigate split-brain is to use a shared-store. > However, if a shared-store is not viable for your use-case the next best > solution is to integrate with ZooKeeper via the pluggable quorum vote > replication configuration [1]. > > > Justin > > [1] > > https://activemq.apache.org/components/artemis/documentation/latest/ha.html#pluggable-quorum-vote-replication-configurations > > > On Thu, Apr 7, 2022 at 9:53 PM Steigerwald, Aaron > <asteigerw...@brandesassociates.com.invalid> wrote: > > > Hello, > > > > My colleague has read that split brain can cause journal corruption in > > a master/slave network replication scenario. Is anyone aware if this > > is possible with current versions of Artemis? > > > > Thank you, > > Aaron Steigerwald > > >
