[HADOOP-18988] Snapshotdiff does not detect rename if parent dir removed

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 3.3.6
• Component/s: fs
• Assignee: Unassigned
• Reporter: Dieter De Paepe
• Created: 23/Nov/23 12:39
• Updated: 23/Nov/23 12:39

Description

In the example case below, the diff reports the moved folder as a CREATE, where it should be a RENAME.

This means that applications using the diff report (such as DistCP) will have to transfer data where a simple move would suffice.

DistributedFileSystem fs = (DistributedFileSystem) this.fs;

createHdfsFile(new Path("/level1/level2/file"), "content");
fs.allowSnapshot(new Path("/"));
fs.createSnapshot(new Path("/"), "snapshot1");

fs.rename(new Path("/level1/level2/"), new Path("/level2-root"));
fs.delete(new Path("/level1"), true);
fs.createSnapshot(new Path("/"), "snapshot2");

SnapshotDiffReport snapshotDiff =
        fs.getSnapshotDiffReport(new Path("/"), "snapshot1", "snapshot2");
System.out.println(snapshotDiff);
//        Difference between snapshot snapshot1 and snapshot snapshot2 under directory /:
//        M .
//        + ./level2-root
//        - ./level1

// If the delete is commented out:
//        M .
//        M ./level1
//        R ./level1/level2 -> ./level2-root