You are viewing a plain text version of this content. The canonical link for it is here.
Posted to hdfs-issues@hadoop.apache.org by "Allen Wittenauer (JIRA)" <ji...@apache.org> on 2009/08/31 19:49:32 UTC

[jira] Created: (HDFS-580) Name node will exit safe mode w/0 blocks even if data nodes are broken

Name node will exit safe mode w/0 blocks even if data nodes are broken
----------------------------------------------------------------------

                 Key: HDFS-580
                 URL: https://issues.apache.org/jira/browse/HDFS-580
             Project: Hadoop HDFS
          Issue Type: Bug
            Reporter: Allen Wittenauer


If one brings up a freshly formatted name node against older data nodes with an incompatible storage id (such that the datanodes fail with Directory /mnt/u001/dfs-data is in an inconsistent state: is incompatible with others.), the name node will still come out of safe mode.  Writes will partially succeed--entries are created, but all are zero length.


-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


[jira] Commented: (HDFS-580) Name node will exit safe mode w/0 blocks even if data nodes are broken

Posted by "Todd Lipcon (JIRA)" <ji...@apache.org>.
    [ https://issues.apache.org/jira/browse/HDFS-580?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12750033#action_12750033 ] 

Todd Lipcon commented on HDFS-580:
----------------------------------

Can this be fixed by HDFS-528?

> Name node will exit safe mode w/0 blocks even if data nodes are broken
> ----------------------------------------------------------------------
>
>                 Key: HDFS-580
>                 URL: https://issues.apache.org/jira/browse/HDFS-580
>             Project: Hadoop HDFS
>          Issue Type: Bug
>            Reporter: Allen Wittenauer
>
> If one brings up a freshly formatted name node against older data nodes with an incompatible storage id (such that the datanodes fail with Directory /mnt/u001/dfs-data is in an inconsistent state: is incompatible with others.), the name node will still come out of safe mode.  Writes will partially succeed--entries are created, but all are zero length.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


[jira] Commented: (HDFS-580) Name node will exit safe mode w/0 blocks even if data nodes are broken

Posted by "Allen Wittenauer (JIRA)" <ji...@apache.org>.
    [ https://issues.apache.org/jira/browse/HDFS-580?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12750037#action_12750037 ] 

Allen Wittenauer commented on HDFS-580:
---------------------------------------

Potentially.  It really comes down to what does the name node consider 'live'.  In our case, the name node actually reported all the DNs as live.  But it makes me wonder if we've just hit a specialized form of HDFS-528.

> Name node will exit safe mode w/0 blocks even if data nodes are broken
> ----------------------------------------------------------------------
>
>                 Key: HDFS-580
>                 URL: https://issues.apache.org/jira/browse/HDFS-580
>             Project: Hadoop HDFS
>          Issue Type: Bug
>            Reporter: Allen Wittenauer
>
> If one brings up a freshly formatted name node against older data nodes with an incompatible storage id (such that the datanodes fail with Directory /mnt/u001/dfs-data is in an inconsistent state: is incompatible with others.), the name node will still come out of safe mode.  Writes will partially succeed--entries are created, but all are zero length.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


[jira] Commented: (HDFS-580) Name node will exit safe mode w/0 blocks even if data nodes are broken

Posted by "Steve Loughran (JIRA)" <ji...@apache.org>.
    [ https://issues.apache.org/jira/browse/HDFS-580?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12750354#action_12750354 ] 

Steve Loughran commented on HDFS-580:
-------------------------------------

Maybe the NN should do a health check by trying to write and then read a file before declaring itself live

> Name node will exit safe mode w/0 blocks even if data nodes are broken
> ----------------------------------------------------------------------
>
>                 Key: HDFS-580
>                 URL: https://issues.apache.org/jira/browse/HDFS-580
>             Project: Hadoop HDFS
>          Issue Type: Bug
>            Reporter: Allen Wittenauer
>
> If one brings up a freshly formatted name node against older data nodes with an incompatible storage id (such that the datanodes fail with Directory /mnt/u001/dfs-data is in an inconsistent state: is incompatible with others.), the name node will still come out of safe mode.  Writes will partially succeed--entries are created, but all are zero length.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.


[jira] Commented: (HDFS-580) Name node will exit safe mode w/0 blocks even if data nodes are broken

Posted by "Allen Wittenauer (JIRA)" <ji...@apache.org>.
    [ https://issues.apache.org/jira/browse/HDFS-580?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12749553#action_12749553 ] 

Allen Wittenauer commented on HDFS-580:
---------------------------------------

My gut tells me that since there are zero blocks assigned, the name node is just happy to get a response not really caring what that response is.

> Name node will exit safe mode w/0 blocks even if data nodes are broken
> ----------------------------------------------------------------------
>
>                 Key: HDFS-580
>                 URL: https://issues.apache.org/jira/browse/HDFS-580
>             Project: Hadoop HDFS
>          Issue Type: Bug
>            Reporter: Allen Wittenauer
>
> If one brings up a freshly formatted name node against older data nodes with an incompatible storage id (such that the datanodes fail with Directory /mnt/u001/dfs-data is in an inconsistent state: is incompatible with others.), the name node will still come out of safe mode.  Writes will partially succeed--entries are created, but all are zero length.

-- 
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.