[
https://issues.apache.org/jira/browse/ZOOKEEPER-3049?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16486540#comment-16486540
]
Bogdan Kanivets commented on ZOOKEEPER-3049:
--------------------------------------------
IMHO, there is a possibility.
Ops within mutli are applied in a loop without any global lock. Locking happens
only on the node within individual setData ops.
https://github.com/apache/zookeeper/blob/branch-3.5/src/java/main/org/apache/zookeeper/server/DataTree.java#L864
One way to protect is to check Stat.mzxid when reading znode1 and znode2. It
should be the same. Multi uses same transaction id for all ops.
> would zookeeper transaction(multi) block the concurrent read?
> -------------------------------------------------------------
>
> Key: ZOOKEEPER-3049
> URL: https://issues.apache.org/jira/browse/ZOOKEEPER-3049
> Project: ZooKeeper
> Issue Type: Wish
> Components: documentation
> Reporter: wayne
> Priority: Blocker
>
> For instance, the original data for znode1 and znode2 are 2 and 4
> respectively. I want to perform increment operations over them. Finally, I
> would get (3, 5) for znode1 and znode2. In order to keep atomicity, I used
> multi() api. Is there any possibility that any clients could read (3, 4)
> concurrently? That is, the read happened after znode1++ and before znode2++?
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
