[ZOOKEEPER-1282] Learner.java not following Zab 1.0 protocol - setCurrentEpoch should be done upon receipt of NEWLEADER (before acking it) and not upon receipt of UPTODATE

• Type: Sub-task
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 3.4.0
• Fix Version/s: 3.4.0, 3.5.0
• Component/s: server
• Release Note: Revision 1198053
• Assignee: Benjamin Reed
• Reporter: Alexander Shraer
• Created: 02/Nov/11 22:33
• Updated: 23/Nov/11 19:22
• Resolved: 05/Nov/11 20:58

Description

according to https://cwiki.apache.org/confluence/display/ZOOKEEPER/Zab1.0
phase 2 part 2, "Once it receives NEWLEADER(e) it atomically applies
the new state and sets f.currentEpoch =e. "

In Learner.java self.setCurrentEpoch(newEpoch) is done after receiving
UPTODATE and not before acking the NEWLEADER message as should be.

case Leader.UPTODATE:
if (!snapshotTaken)

{ zk.takeSnapshot(); }

self.cnxnFactory.setZooKeeperServer(zk);
break outerLoop;
case Leader.NEWLEADER: // it will be NEWLEADER in v1.0
zk.takeSnapshot();
snapshotTaken = true;
writePacket(new QuorumPacket(Leader.ACK,
newLeaderZxid, null, null), true);
break;
}
}
}
long newEpoch = ZxidUtils.getEpochFromZxid(newLeaderZxid);
self.setCurrentEpoch(newEpoch);

Issue Links

• is part of: ZOOKEEPER-1264 FollowerResyncConcurrencyTest failing intermittently