steveloughran commented on a change in pull request #23: CRUNCH-683 Avoid
unnecessary listStatus calls from getPathSize computation
URL: https://github.com/apache/crunch/pull/23#discussion_r278042702
##########
File path:
crunch-core/src/main/java/org/apache/crunch/io/SourceTargetHelper.java
##########
@@ -41,17 +41,23 @@ public static long getPathSize(FileSystem fs, Path path)
throws IOException {
}
long size = 0;
for (FileStatus status : stati) {
- if (status.isDir()) {
- for (FileStatus st : fs.listStatus(status.getPath())) {
- size += getPathSize(fs, st.getPath());
- }
- } else {
- size += status.getLen();
- }
+ size += getPathSize(fs, status);
}
return size;
}
-
+
+ private static long getPathSize(final FileSystem fs, final FileStatus
status) throws IOException {
Review comment:
This is still doing a recursive treewalk. IF you do
`FileSystem.listFiles(path, true)` you get a deep listing of files only from
the store. For an object store with an optimised implementation (as s3a does),
only one HTTP request is made per 1000 objects, irrespective of the depth of
the tree. For HDFS &c there's still efficiencies, especially when you have a
directory with many millions of files in: in listStatus() all the results have
to get serialized and marshalled over as one, rather than paged over to the
client
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
