You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@zookeeper.apache.org by "Hongchao Deng (JIRA)" <ji...@apache.org> on 2014/08/24 21:05:38 UTC
[jira] [Commented] (ZOOKEEPER-2020) Change TRUNC to SNAP in sync
phase for safety guarantee
[ https://issues.apache.org/jira/browse/ZOOKEEPER-2020?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14108497#comment-14108497 ]
Hongchao Deng commented on ZOOKEEPER-2020:
------------------------------------------
[~fpj][~thawan][~fanster.z]
> Change TRUNC to SNAP in sync phase for safety guarantee
> --------------------------------------------------------
>
> Key: ZOOKEEPER-2020
> URL: https://issues.apache.org/jira/browse/ZOOKEEPER-2020
> Project: ZooKeeper
> Issue Type: Sub-task
> Components: quorum
> Affects Versions: 3.5.0
> Reporter: Hongchao Deng
> Fix For: 3.5.1
>
>
> ZOOKEEPER-1549 discusses the problem:
> "When the leader started, it will apply every txn in its txnlog (incl. uncommitted ones) into its in-memory data tree"
> I didn't any solution so far solved this problem in 3.5.x. Since this affects only TRUNC part -- only old leader that needs TRUNC applies uncommitted txns, a simple fix would be change current TRUNC logic to SNAP. This isn't hard to implement, but guarantees safety.
> Ideally, we will solve the whole problem by untangling all compatibility issues and fixing the protocol.
--
This message was sent by Atlassian JIRA
(v6.2#6252)