gaoxiao created ZOOKEEPER-1492:
----------------------------------
Summary: leader cannot switch to LOOKING state when lost the
majority
Key: ZOOKEEPER-1492
URL: https://issues.apache.org/jira/browse/ZOOKEEPER-1492
Project: ZooKeeper
Issue Type: Bug
Components: quorum
Affects Versions: 3.4.3
Environment: eclipse linux
Reporter: gaoxiao
Priority: Critical
When a follower leave the cluster, and the cluster cannot achieve a majority,
the leader should get out from Leading stat and get into Looking state, but if
the there are some observers, the leader will not get away and the client
cannot use the cluster.
eg:
The servers config:
server.1=z1:2888:3888
server.2=z2:2888:3888
server.3=z3:2888:3888:observer
At first, 1,2,3 are all started, it's all ok, 2 is the leader, but at this
time, if 1 is stopped, 2 will not leave the Leading state, and client cannot
connect to cluster.
I think the problem is:
(Leader.java method:lead)
Line 388-407
syncedSet.add(self.getId());
synchronized (learners) {
for (LearnerHandler f : learners) {
if (f.synced()) {
syncedCount++;
syncedSet.add(f.getSid());
}
f.ping();
}
}
if (!tickSkip &&
!self.getQuorumVerifier().containsQuorum(syncedSet)) {
//if (!tickSkip && syncedCount < self.quorumPeers.size() / 2) {
// Lost quorum, shutdown
// TODO: message is wrong unless majority quorums used
shutdown("Only " + syncedCount + " followers, need "
+ (self.getVotingView().size() / 2));
// make sure the order is the same!
// the leader goes to looking
return;
}
The code add all learners' ping to syncedSet, and I think at this place, only
followers should be added to syncedSet, so the method 'containsQuorum' can
figure out the majority.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
