Henry Robinson commented on ZOOKEEPER-769:

Sergey - 

In the cfg files for nodes 3 and 5, did you include the following line? 


See http://hadoop.apache.org/zookeeper/docs/r3.3.0/zookeeperObservers.html for 
details. The observer log contains this line:

2010-05-06 22:46:00,876 - INFO  
[QuorumPeer:/0:0:0:0:0:0:0:0:2183:quorump...@642] - FOLLOWING

which is a big red flag because observers should never adopt the FOLLOWING 

If I don't have that line I can reproduce your issue. If I add it, the 
observers work as expected. Can you check your cfg files?


> Leader can treat observers as quorum members
> --------------------------------------------
>                 Key: ZOOKEEPER-769
>                 URL: https://issues.apache.org/jira/browse/ZOOKEEPER-769
>             Project: Zookeeper
>          Issue Type: Bug
>    Affects Versions: 3.3.0
>         Environment: Ubuntu Karmic x64
>            Reporter: Sergey Doroshenko
>             Fix For: 3.3.0
>         Attachments: follower.log, leader.log, observer.log, zoo1.cfg
> In short: it seems leader can treat observers as quorum members.
> Steps to repro:
> 1. Server configuration: 3 voters, 2 observers (attached).
> 2. Bring up 2 voters and one observer. It's enough for quorum.
> 3. Shut down the one from the quorum who is the follower.
> As I understand, expected result is that leader will start a new election 
> round so that to regain quorum.
> But the real situation is that it just says goodbye to that follower, and is 
> still operable. (When I'm shutting down 3rd one -- observer -- leader starts 
> trying to regain a quorum).
> (Expectedly, if on step 3 we shut down the leader, not the follower, 
> remaining follower starta new leader election, as it should be).

This message is automatically generated by JIRA.
You can reply to this email to add a comment to the issue online.

Reply via email to