Mailing-List: contact issues-help@hbase.apache.org; run by ezmlm
Precedence: bulk
Date: Thu, 25 May 2017 23:58:04 +0000 (UTC)
From: "Vladimir Rodionov (JIRA)" <jira@apache.org>
To: issues@hbase.apache.org
Message-ID: <JIRA.12996438.1470867398000.298432.1495756684416@Atlassian.JIRA>
In-Reply-To: <JIRA.12996438.1470867398000@Atlassian.JIRA>
References: <JIRA.12996438.1470867398000@Atlassian.JIRA> <JIRA.12996438.1470867398986@jira-lw-us.apache.org>
Subject: [jira] [Commented] (HBASE-16392) Backup delete fault tolerance
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit
archived-at: Thu, 25 May 2017 23:58:09 -0000


    [ https://issues.apache.org/jira/browse/HBASE-16392?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16025575#comment-16025575 ] 

Vladimir Rodionov commented on HBASE-16392:
-------------------------------------------

{quote}
If deleteSnapshot() fails in the middle, BackupSystemTable.restoreFromSnapshot(conn) call in catch block would fail as well, right ?
{quote}

Good catch. Will need to fix this.

> Backup delete fault tolerance
> -----------------------------
>
>                 Key: HBASE-16392
>                 URL: https://issues.apache.org/jira/browse/HBASE-16392
>             Project: HBase
>          Issue Type: Sub-task
>            Reporter: Vladimir Rodionov
>            Assignee: Vladimir Rodionov
>              Labels: backup
>             Fix For: 2.0.0
>
>         Attachments: HBASE-16392-v1.patch
>
>
> Backup delete modified file system and backup system table. We have to make sure that operation is atomic, durable and isolated.
> Delete operation:
> # Start backup session (this guarantees) that system will be blocked for all backup commands during delete operation
> # Save list of tables being deleted to system table
> # Before delete operation we take backup system table snapshot  
> # During delete operation we detect any failures and restore backup system table from snapshot, then finish backup session
> # To guarantee consistency of the data, delete operation MUST be repeated
> # We guarantee that all file delete operations are idempotent, can be repeated multiple times
> # Any backup operations will be blocked until consistency is restored
> # To restore consistency, repair command must be executed.
> # Repair command checks if there is failed delete op in a backup system table, and repeats delete operation


--
This message was sent by Atlassian JIRA
(v6.3.15#6346)