Mailing-List: contact commits-help@cassandra.apache.org; run by ezmlm
Precedence: bulk
Reply-To: dev@cassandra.apache.org
Date: Wed, 15 Jul 2015 11:28:04 +0000 (UTC)
From: "Andreas Schnitzerling (JIRA)" <jira@apache.org>
To: commits@cassandra.apache.org
Message-ID: <JIRA.12845237.1436959683000.193880.1436959684526@Atlassian.JIRA>
In-Reply-To: <JIRA.12845237.1436959683000@Atlassian.JIRA>
References: <JIRA.12845237.1436959683000@Atlassian.JIRA>
 <JIRA.12845237.1436959683911@arcas>
Subject: [jira] [Created] (CASSANDRA-9812) Handle corrupted files during
 startup.
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit

Andreas Schnitzerling created CASSANDRA-9812:
------------------------------------------------

             Summary: Handle corrupted files during startup.
                 Key: CASSANDRA-9812
                 URL: https://issues.apache.org/jira/browse/CASSANDRA-9812
             Project: Cassandra
          Issue Type: New Feature
         Environment: Windows-7-32 bit, 3.2GB RAM, Java 1.7.0_55
            Reporter: Andreas Schnitzerling


This ticket is relying to CASSANDRA-9686 (refer for details). Here the conclusion: In our company we cannot avoid power-cut of the nodes (unexpected and for tests). We need a behavior, which keeps the nodes online even on finding corrupted files during startup. One idea was copy and scrub corrupted files. [~Stefania] wrote:
{code}
Yes a disk corruption due to a power cut could explain it. I don't think we should delete corrupt sstables though, but we could maybe move them somewhere else - where they wouldn't be automatically loaded. Then the scrub tool could copy the fixed version back into the right folder, but this is kind of opposite of what it does at the moment (save a backup and then fix the original).
{code}
This could avoid stopping the nodes and keep the cluster running. We need that behavior only on startup of the nodes, not during runtime. The only cause seems to be power-cut. The nodes are configured to start C* as a service.


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