Details
-
Bug
-
Status: Closed
-
Major
-
Resolution: Fixed
-
0.10.1
-
None
-
None
Description
dfs "du" is implemented by sending a listPaths request to the namenode to get the size of each file/subdir under the directory. At the namenode side, the size of subdir was calculated by recursively going through the whole subtree with the subdir as the root. But starting from the release 0.10.0, the size of subdir is no longer gets calculated. So dfs "du" shows its size as 0.
The problem is that both "du" and "list" send the same request "listPaths" to the namenode. The previous implmentation made list very expensive, but the current implementation makes du not working.
Attachments
Attachments
Issue Links
- is related to
-
HADOOP-713 dfs list operation is too expensive
- Closed