[
https://issues.apache.org/jira/browse/HDFS-16847?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17636189#comment-17636189
]
ASF GitHub Bot commented on HDFS-16847:
---------------------------------------
simbadzina commented on code in PR #5145:
URL: https://github.com/apache/hadoop/pull/5145#discussion_r1027110257
##########
hadoop-hdfs-project/hadoop-hdfs-rbf/src/main/java/org/apache/hadoop/hdfs/server/federation/store/driver/impl/StateStoreFileSystemImpl.java:
##########
@@ -82,17 +82,8 @@ protected boolean mkdir(String path) {
@Override
protected boolean rename(String src, String dst) {
try {
- if (fs instanceof DistributedFileSystem) {
- DistributedFileSystem dfs = (DistributedFileSystem)fs;
- dfs.rename(new Path(src), new Path(dst), Options.Rename.OVERWRITE);
- return true;
- } else {
- // Replace should be atomic but not available
- if (fs.exists(new Path(dst))) {
- fs.delete(new Path(dst), true);
- }
- return fs.rename(new Path(src), new Path(dst));
- }
+ FileUtil.rename(fs, new Path(src), new Path(dst),
Options.Rename.OVERWRITE);
Review Comment:
Hi @tomscut , if we aren't using a DistributedFileSystem, the old code does
the replace itself by using a delete (if necessary) and then rename. This code
is already available in Filesystem.java.
More importantly, if there is a non DistributedFileSystem that has an atomic
rename, the old code prevents us from using the available atomic rename.
My change delegates the implementation of rename to the underlying
Filesystem.
https://github.com/apache/hadoop/blob/c71a68ca8079ad8c87ddca37e51dfdc3bc2dbfc1/hadoop-common-project/hadoop-common/src/main/java/org/apache/hadoop/fs/FileSystem.java#L1648-L1650
> RBF: StateStore writer should not commit tmp fail if there was an error in
> writing the file.
> --------------------------------------------------------------------------------------------
>
> Key: HDFS-16847
> URL: https://issues.apache.org/jira/browse/HDFS-16847
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: hdfs, rbf
> Reporter: Simbarashe Dzinamarira
> Assignee: Simbarashe Dzinamarira
> Priority: Critical
> Labels: pull-request-available
>
> The file based implementation of the RBF state store has a commit step that
> moves a temporary file to a permanent location.
> There is a check to see if the write of the temp file was successfully,
> however, the code to commit doesn't check the success flag.
> This is the relevant code:
> [https://github.com/apache/hadoop/blob/7d39abd799a5f801a9fd07868a193205ab500bfa/hadoop-hdfs-project/hadoop-hdfs-rbf/src/main/java/org/apache/hadoop/hdfs/server/federation/store/driver/impl/StateStoreFileBaseImpl.java#L369]
>
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
