[ 
https://issues.apache.org/jira/browse/ZOOKEEPER-29?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12700082#action_12700082
 ] 

Flavio Paiva Junqueira commented on ZOOKEEPER-29:
-------------------------------------------------

Thanks for your comments, Mahadev. Here are my response based on our discussion:

1- Using this new feature with FastLeaderElection is easier because the 
protocol already exchanges server ids. It will require more modifications to 
move LeaderElection and AuthFastLeaderElection, so I propose to work on it in a 
different jira. If a user decides to use one of these implementations, then 
only majority quorums will be available;
2- Mahadev suggested to change the LASTZXID message to FOLLOWERINFO and send 
both the lastzxid and the server id in the same message. The main reason for 
doing this is to avoid creating a new message every time we have to send a new 
piece of information. I think it is a good suggestion, so I'll implement it in 
the next patch;
3- The SID case is in the loop just in case we receive a stray message. It 
shouldn't happen with this patch, so I'm ok with removing it. In fact, with 
modification of 2, that case will have to go anyway;
4- We agreed that the problem pointed out in Leader.java:297 is not really a 
problem because SID is the first message sent. The same will happen with the 
modification of 2;
5- Mahadev suggests that instead of falling back to majority if there is a 
problem with the group configuration, we should throw an exception. I'm ok with 
that suggestion;
6- Although it may seem that using quorums of size 4 with 9 replicas may cause 
replicas to diverge, it is simple exercise to verify that all quorums 
constructed hierarchically in the way proposed are guaranteed to intersect. All 
we need to guarantee that replicas won't diverge is that quorums intersect. 
Majority quorums trivially intersect;
7-  I don't think the failure in the test is due to this patch. I looked over 
the results and it is due to testRecovery failing due to "unreasonable length". 
Isn't it the issue of ZOOKEEPER-367?

> Flexible quorums
> ----------------
>
>                 Key: ZOOKEEPER-29
>                 URL: https://issues.apache.org/jira/browse/ZOOKEEPER-29
>             Project: Zookeeper
>          Issue Type: New Feature
>          Components: server
>            Reporter: Patrick Hunt
>            Assignee: Flavio Paiva Junqueira
>             Fix For: 3.2.0
>
>         Attachments: ZOOKEEPER-29.patch, ZOOKEEPER-29.patch, 
> ZOOKEEPER-29.patch, ZOOKEEPER-29.patch, ZOOKEEPER-29.patch, 
> ZOOKEEPER-29.patch, ZOOKEEPER-29.patch
>
>
> Moved from SourceForge to Apache.
> http://sourceforge.net/tracker/index.php?func=detail&aid=1938782&group_id=209147&atid=1008547

-- 
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