Mailing-List: contact dev-help@tephra.incubator.apache.org; run by ezmlm
Precedence: bulk
Reply-To: dev@tephra.incubator.apache.org
Date: Wed, 16 Nov 2016 07:06:58 +0000 (UTC)
From: "Poorna Chandra (JIRA)" <jira@apache.org>
To: dev@tephra.incubator.apache.org
Message-ID: <JIRA.13014103.1477012719000.289811.1479280018319@Atlassian.JIRA>
In-Reply-To: <JIRA.13014103.1477012719000@Atlassian.JIRA>
References: <JIRA.13014103.1477012719000@Atlassian.JIRA> <JIRA.13014103.1477012719156@arcas>
Subject: [jira] [Commented] (TEPHRA-195) Store transaction state in HBase
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit
archived-at: Wed, 16 Nov 2016 07:07:07 -0000


    [ https://issues.apache.org/jira/browse/TEPHRA-195?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15669669#comment-15669669 ] 

Poorna Chandra commented on TEPHRA-195:
---------------------------------------

Storing transaction state in HBase can be done by implementing an HBase table backed storage for {{TransactionStateStorage}} and {{TransactionLogWriter}} interfaces. 

To ensure data consistency, in addition to replicating transaction state we would have to make sure that a committed transaction is only visible in the destination cluster after all the tables that contain a write from that transaction have been replicated.

Another approach for replication would be to build a replication mechanism into Tephra to replicate the transaction logs that transaction manager writes for every operation. We may have more control on determining the replicated state on the destination cluster in this case.

Once we come up with the conditions that need to be met for the replicated data to be consistent, we can make a choice on which approach would satisfy the conditions.


> Store transaction state in HBase
> --------------------------------
>
>                 Key: TEPHRA-195
>                 URL: https://issues.apache.org/jira/browse/TEPHRA-195
>             Project: Tephra
>          Issue Type: Improvement
>            Reporter: James Taylor
>            Assignee: Poorna Chandra
>
> The transactional state (inflight and invalid transaction info) is stored in HDFS for Tephra making it difficult to provide the same disaster recovery guarantees that can be provided if the data was stored in HBase. Instead of writing directly to HDFS we should have an option to write to HBase instead.


--
This message was sent by Atlassian JIRA
(v6.3.4#6332)