Owen O'Malley
|
56f4f967cb
The old core branches are now common branches.
|
16 år sedan |
Doug Cutting
|
b52540f262
Moving branches to new location.
|
17 år sedan |
Doug Cutting
|
0077fb7333
Merge -r 390737:392453 from trunk, preparing for 0.1.1 release.
|
19 år sedan |
Doug Cutting
|
f0dd4ebc63
Update developer build version in branch.
|
19 år sedan |
Doug Cutting
|
bc897b8c84
Starting a branch for 0.1 releases.
|
19 år sedan |
Doug Cutting
|
0caa6bc1b1
More fixes to get working directory to work on Windows. Long-term we should stop using java.io.File for our abstract file paths. On Windows, 'new File(/foo/bar).isAbsolute()' returns false, which caused lots of problems. So I have added a FileSystem.isAbsolute() method that we use internally. I also added a few more .getAbsoluteFile() calls to convert paths into absolute paths so that LocalFileSystem works correctly. Finally I took advantage of file status information cached in DFSFile, eliminating some namenode RPCs.
|
19 år sedan |
Doug Cutting
|
f465056eb6
Fix wiki url.
|
19 år sedan |
Doug Cutting
|
09d93f190f
If close() fails, then abandon the file, so that any leases are
|
19 år sedan |
Doug Cutting
|
119226c1cb
Fix a bug where writing zero-length files would cause things to hang.
|
19 år sedan |
Doug Cutting
|
1740f0b0d4
DFS re-format should fully delete old namenode data.
|
19 år sedan |
Doug Cutting
|
3f5b4c0626
Fix for HADOOP-102. Contributed by Konstantin.
|
19 år sedan |
Doug Cutting
|
d98cef1ab3
Fix HADOOP-100. Be more consistent about synchronization of access to taskTracker collection. Contributed by Owen O'Malley.
|
19 år sedan |
Doug Cutting
|
03c953c0dc
Fix for HADOOP-107. As they were written, dfs blocks were both trickled to a datanode and tee'd to a temp file (in case the connection to the datanode failed). Now they're only written to the temp file, with no connection to the datanode made until the block is complete. This reduces the number of long-lived mostly-idle connections to datanodes, which was causing problems. It also simplifies the DFSClient code significantly.
|
19 år sedan |
Doug Cutting
|
294aa869c5
Fix for HADOOP-103, part II: I forgot to add this file the first time!
|
19 år sedan |
Doug Cutting
|
eeea2a252d
Fix for HADOOP-103. Add a base class for Mapper and Reducer implementations that implements Closeable and JobConfigurable. Use it in supplied Mappers & Reducers. Also some minor improvements to demos. Contributed by Owen O'Malley.
|
19 år sedan |
Doug Cutting
|
7d0429d4d2
Fix for HADOOP-110. Reuse keys and values when mapping. Contributed by Owen O'Malley.
|
19 år sedan |
Doug Cutting
|
ec87d7e0e3
Fix for HADOOP-2. The combiner now clones keys and values, so mappers may now safely reuse emitted keys and values. Contributed by Owen O'Malley.
|
19 år sedan |
Doug Cutting
|
ecf45fd810
Fix for HADOOP-112. All operations on local files are now performed
|
19 år sedan |
Doug Cutting
|
7018745fac
Fix for HADOOP-84. Improve error and log messages when block cannot be obtained by including file name and offset. Also removed a few unused variables. Contributed by Konstantin Shvachko.
|
19 år sedan |
Doug Cutting
|
95091af3ce
Fix HADOOP-33. Avoid calling df too frequently by caching values internally. Contributed by Konstantin Shvachko.
|
19 år sedan |
Doug Cutting
|
4eccd587cc
Fix HADOOP-67. Add a public API for dfs statistics. Also switch to use the public API for reporting in DFSShell.
|
19 år sedan |
Andrzej Bialecki
|
2948bf60de
Add a tool for checking DFS consistency (HADOOP-101).
|
19 år sedan |
Doug Cutting
|
9f4fa16216
Fix for file names with spaces.
|
19 år sedan |
Doug Cutting
|
27a97ef6d3
Always return an absolute pathname for local files. This fixes
|
19 år sedan |
Doug Cutting
|
24465c6af4
Fix unit tests on Windows. Don't assume that, just because a pathname begins with a slash that it also returns true for File.isAbsolute(). Instead use getAbsoluteFile() to force such things to be absolute.
|
19 år sedan |
Doug Cutting
|
943ba3132a
Document the new format command.
|
19 år sedan |
Doug Cutting
|
73be419f00
Fix for HADOOP-19. A namenode must now be formatted before it may be used. Attempts to start a namenode in an unformatted directory will fail, rather than automatically creating a new, empty filesystem, causing existing datanodes to delete all blocks. Thus a mis-configured dfs.data.dir should no longer cause data loss.
|
19 år sedan |
Doug Cutting
|
82236adb3c
Move checking of output directory existence from JobClient to OutputFormat, so that it can be overridden. Add a base class for OutputFormat that implements this new method.
|
19 år sedan |
Doug Cutting
|
40644a69e6
Fix for HADOOP-46. Jobs can now be named. Contributed by Owen O'Malley.
|
19 år sedan |
Doug Cutting
|
2487ad9876
Fix for HADOOP-98. Keep more accurate task counts. Contributed by Owen O'Malley.
|
19 år sedan |