hadoop-hdfs-issues mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Damien Hardy (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (HDFS-4101) ZKFC should implement zookeeper.recovery.retry like HBase to connect to ZooKeeper
Date Tue, 23 Oct 2012 08:23:12 GMT

    [ https://issues.apache.org/jira/browse/HDFS-4101?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13482194#comment-13482194
] 

Damien Hardy commented on HDFS-4101:
------------------------------------

Actually my patch does not provide a conf entry to set number of retries like HBase does (and
the issue description let thinking)
It just provide the retries policy as soon as the service starts.
Seams to be more a bugfix in fact ?

I am not happy with the tests that duplicate a lot of code and don't tests the limit out of
bound (fails more than 3 times in fact).
But I didn't succeed in providing a polymorph method with optional parameter setting the number
of fails in getNewZooKeeper()
(I suppose I don't get well with variable scope in Mocked classe)
                
> ZKFC should implement zookeeper.recovery.retry like HBase to connect to ZooKeeper
> ---------------------------------------------------------------------------------
>
>                 Key: HDFS-4101
>                 URL: https://issues.apache.org/jira/browse/HDFS-4101
>             Project: Hadoop HDFS
>          Issue Type: Improvement
>          Components: auto-failover, ha
>    Affects Versions: 2.0.0-alpha, 3.0.0
>         Environment: running CDH4.1.1
>            Reporter: Damien Hardy
>            Assignee: Damien Hardy
>            Priority: Minor
>              Labels: newbie
>         Attachments: HDFS-4101-1.patch
>
>
> When zkfc start and zookeeper is not yet started ZKFC fails and stop directly.
> Maybe ZKFC should allow some retries on Zookeeper services like does HBase with zookeeper.recovery.retry
> This particularly appends when I start my whole cluster on VirtualBox for example (every
components nearly at the same time) ZKFC is the only that fail and stop ... 
> Every others can wait each-others some time independently of the start order like NameNode/DataNode/JournalNode/Zookeeper/HBaseMaster/HBaseRS
so that the system can be set and stable in few seconds

--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira

Mime
View raw message