Mailing-List: contact hdfs-issues-help@hadoop.apache.org; run by ezmlm
Precedence: bulk
Date: Mon, 24 Oct 2016 21:20:58 +0000 (UTC)
From: "Arpit Agarwal (JIRA)" <jira@apache.org>
To: hdfs-issues@hadoop.apache.org
Message-ID: <JIRA.13004364.1473701057000.71731.1477344058537@Atlassian.JIRA>
In-Reply-To: <JIRA.13004364.1473701057000@Atlassian.JIRA>
References: <JIRA.13004364.1473701057000@Atlassian.JIRA> <JIRA.13004364.1473701057026@arcas>
Subject: [jira] [Commented] (HDFS-10857) Rolling upgrade can make data
 unavailable when the cluster has many failed volumes
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit
archived-at: Mon, 24 Oct 2016 21:21:00 -0000


    [ https://issues.apache.org/jira/browse/HDFS-10857?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15603256#comment-15603256 ] 

Arpit Agarwal commented on HDFS-10857:
--------------------------------------

Looks like {{checkDiskError}} should get the DataNode object lock for the {{dataDirs}} modification to avoid a potential race with {{refreshVolumes}}.

> Rolling upgrade can make data unavailable when the cluster has many failed volumes
> ----------------------------------------------------------------------------------
>
>                 Key: HDFS-10857
>                 URL: https://issues.apache.org/jira/browse/HDFS-10857
>             Project: Hadoop HDFS
>          Issue Type: Bug
>    Affects Versions: 2.6.4
>            Reporter: Kihwal Lee
>            Priority: Critical
>         Attachments: HDFS-10857.branch-2.6.patch
>
>
> When the marker file or trash dir is created or removed during the heartbeat response processing, an {{IOException}} is thrown if tried on a failed volume.   This stops processing of the rest of storage directories and any DNA commands that were part of the heartbeat response.
> While this is happening, the block token key update does not happen and all read and write requests start to fail, until the upgrade is finalized and the DN receives a new key. All it takes is one failed volume. If there are three such nodes in the cluster, it is very likely that some blocks cannot be read. The NN has no idea unlike the common missing blocks scenarios, although the effect is the same.


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

---------------------------------------------------------------------
To unsubscribe, e-mail: hdfs-issues-unsubscribe@hadoop.apache.org
For additional commands, e-mail: hdfs-issues-help@hadoop.apache.org