curator-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Mike Drob (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (CURATOR-50) CreateBuilderImpl fails on forPath if connectString features a chroot that does not already exist.
Date Mon, 04 Aug 2014 17:36:11 GMT

    [ https://issues.apache.org/jira/browse/CURATOR-50?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14084942#comment-14084942
] 

Mike Drob commented on CURATOR-50:
----------------------------------

I tried this earlier today, and attempting to create "/" under a non-existent multi-node chroot
did fail for me using raw zk handles.

> CreateBuilderImpl fails on forPath if connectString features a chroot that does not already
exist.
> --------------------------------------------------------------------------------------------------
>
>                 Key: CURATOR-50
>                 URL: https://issues.apache.org/jira/browse/CURATOR-50
>             Project: Apache Curator
>          Issue Type: Bug
>          Components: Framework
>    Affects Versions: 2.1.0-incubating
>            Reporter: Sean-Michael Lewis
>            Priority: Minor
>             Fix For: TBD
>
>
> When a CuratorFramework is initialized with a connectString featuring a chroot that does
not already exist in the ensemble, CreateBuilderImpl will fail to create new nodes even if
createParentsIfNeeded is true.
> For example, the following code will result in a org.apache.zookeeper.KeeperException$NoNodeException.
> {code}
> CuratorFramework client = CuratorFramework.builder().retryPolicy(myPolicy).connectString("myServer1:2181,myServer2:2181/chroot).build();
> client.create().createParentsIfNeeded().forPath("test");
> {code}
> This can be worked around by using a namespace in lieu of the chroot or by calling {code}client.create().forPath("/"){code}
before attempting to create any other nodes. 
> While using namespaces is likely the best practice, the framework does initialize with
the chroot connectString. There are also reasons why one might want to use both chroot connectStrings
as well as namespaces (application environments for the former, application for the latter).
> My proposed fix is to alter {code}ZkPaths.mkdirs{code} to not skip "/" when it walks
the tree. In cases where no chroot configured or a chroot that already exists, the node will
be found and skipped. Otherwise, it will be created.



--
This message was sent by Atlassian JIRA
(v6.2#6252)

Mime
View raw message