[
https://issues.apache.org/jira/browse/HDFS-3716?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14570333#comment-14570333
]
Vinayakumar B commented on HDFS-3716:
-------------------------------------
Latest patch and test looks fine.
Just small nits in test.
Here {{dfsDir}} is not actually referring to directory. Instead you might want
to rename this variable to 'staleCkptFile' ?
{code}// Create a stale fsimage.ckpt file
File dfsDir = new File(currentDir.getPath() +
"/fsimage.ckpt_0000000000000000002");{code}
Add null check before shutting down for both cluster and snn. Otherwise, since
both snn and cluster are started inside try block, if any actual problem in
starting up cluster/secondarynamenode, that will be masked by NPE from finally
block.
{code} } finally {
secondary.shutdown();
secondary = null;
cluster.shutdown();
cluster = null;
}{code}
> Purger should remove stale fsimage ckpt files
> ---------------------------------------------
>
> Key: HDFS-3716
> URL: https://issues.apache.org/jira/browse/HDFS-3716
> Project: Hadoop HDFS
> Issue Type: Improvement
> Components: namenode
> Affects Versions: 2.0.0-alpha
> Reporter: suja s
> Assignee: J.Andreina
> Priority: Minor
> Attachments: HDFS-3716.1.patch, HDFS-3716.2.patch
>
>
> NN got killed while checkpointing in progress before renaming the ckpt file
> to actual file.
> Since the checkpointing process is not completed, on next NN startup it will
> load previous fsimage and apply rest of the edits.
> Functionally there's no harm but this ckpt file will be retained as is.
> Purger will not remove the ckpt file though other old fsimage files will be
> taken care.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
