Hi ly, Thanks for raising the question. eventQueue mainly contains updateCommitEvent(s) which are for updating the commit index. So that it is okay to process them before checking leadership. Since the entries are committed, it won't cause split-brain.
Tsz-Wo On Thu, May 13, 2021 at 6:26 PM ly <[email protected]> wrote: > Hi, I found ratis leader would not call stepDown when it is receiving > write requests continually even though it lose connection with the majority. > I guess the problem may caused by a bug in EventProcessor.run(). Here is > what the code looks like: > > ===== > > while (running) { > > final StateUpdateEvent event = eventQueue.poll(); > > synchronized(server) { > > if (running) { > > if (event != null) { > > event.execute(); > > } else if (inStagingState()) { > > checkStaging(); > > } else { > > yieldLeaderToHigherPriorityPeer(); > > checkLeadership(); > > } > > } > > } > > } > > ===== > > We can see that if the eventQueue is not empty, then the leader would not > call checkLeadership() to make sure its > > leader state. And this can cause Split-brain. Is that right? If so, I > would like to create a jira and help to fix it. > > > > > Sorry I don't know where is the best place to discuss this kind of > problem, so I may send it to different places. > > Thanks, > ly
