You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@hive.apache.org by "Eugene Koifman (JIRA)" <ji...@apache.org> on 2014/07/01 02:35:24 UTC

[jira] [Updated] (HIVE-7256) HiveTxnManager should be stateless

     [ https://issues.apache.org/jira/browse/HIVE-7256?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Eugene Koifman updated HIVE-7256:
---------------------------------

    Attachment: HIVE-7256.addendum.patch

HIVE-7256.addendum.patch is in addition to HIVE-7256.patch.  Contains fix to TxnManagerFactory so that both types of managers can coexist.  
Contains a fix to HcatDbTxnManager.reconstructTxnInfo() to make it work for read-only transactions (as designed in HIVE-7256.patch).  Adds some more detailed logging.

> HiveTxnManager should be stateless
> ----------------------------------
>
>                 Key: HIVE-7256
>                 URL: https://issues.apache.org/jira/browse/HIVE-7256
>             Project: Hive
>          Issue Type: Bug
>          Components: Locking
>    Affects Versions: 0.13.1
>            Reporter: Eugene Koifman
>            Assignee: Alan Gates
>         Attachments: HIVE-7256.addendum.patch, HIVE-7256.patch
>
>
> In order to integrate HCat with Hive ACID, we should be able to create an instance of HiveTxnManager and use it to acquire locks, and release locks from a different instance of HiveTxnManager.
> One use case where this shows up is when a job using HCat is retried, since calls to TxnManager are made from the jobs OutputCommitter.
> Another, is HCatReader/Writer.  For example, TestReaderWriter, calls setupJob()  from one instance of OutputCommitterContainer and commitJob() from another instance.  The 2nd case is perhaps better solved by ensuring there is only 1 instance of OutputCommitterContainer.



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