[
https://issues.apache.org/jira/browse/HDFS-1839?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13020126#comment-13020126
]
Konstantin Shvachko commented on HDFS-1839:
-------------------------------------------
{{moveLastCheckpoint()}} first
- removes {{previous.checkpoint}}, then
- renames {{lastcheckpoint.tmp}} to {{previous.checkpoint}}, if
{{lastcheckpoint.tmp}} exists.
So if you call {{moveLastCheckpoint()}} twice in one session, then the first
call will correctly rename {{lastcheckpoint.tmp}} to {{previous.checkpoint}},
but the second will only remove {{previous.checkpoint}}, because
{{lastcheckpoint.tmp}} does not exist anymore.
The two calls are done in {{SNN.doCheckpoint()}} - one comes from
{{saveNamespace()}}, which is called by {{doMerge()}}, and another later from
{{endCheckpoint()}}.
The solution is to remove {{endCheckpoint()}}, which seems to be a redundant
operation.
> SecondaryNameNode should not discard "previous.checkpoint"
> ----------------------------------------------------------
>
> Key: HDFS-1839
> URL: https://issues.apache.org/jira/browse/HDFS-1839
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: name-node
> Affects Versions: 0.20.1
> Reporter: Konstantin Shvachko
> Fix For: 0.22.0
>
>
> While creating a checkpoint SNN calls {{moveLastCheckpoint()}} twice. As a
> result directory "previous.checkpoint" is getting removed.
> "previous.checkpoint" directory holds the checkpoint that was created before
> the current one. It is useful for manual recovery, but nothing else.
--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira
