Details
-
New Feature
-
Status: Closed
-
Major
-
Resolution: Fixed
-
None
-
None
-
None
-
Tested on Linux and Windows
Description
A large patch for streaming. Changes:
Support for on-the-fly merge sort of multiple map input files.
This supposes that the inputs are already sorted.
Support for reducer-NONE side-effects to a single local output with DFS inputs.
This can be used to do an on-the-fly merge-sort of remote sorted files.
(Compare to DFSShell -getmerge which does catenation of remote sorted files)
The single output can be a regular file, a named pipe or a socket.
URI Syntax: -mapsideoutput file:/C:/win
Add an optional JUnit test for on-the-fly merge-sort.
It requires Unix tools. It also works with cygwin.
If it has been more than 10 secs since last time we did this:
call reporter.setStatus() when consuming a stderr line from the Application.
Calling setStatus with reducer-NONE was already done as part of HADOOP-413.
So overall this resolves HADOOP-540.
Reformat streaming code to conform to Hadoop conventions
(indent 2 spaces, opening bracket on same-line)