hadoop/CHANGES.txt

Hadoop Change Log


Trunk (unreleased changes)

 1. HADOOP-352.  Fix shell scripts to use /bin/sh instead of
    /bin/bash, for better portability.
    (Jean-Baptiste Quenot via cutting)

 2. HADOOP-313.  Permit task state to be saved so that single tasks
    may be manually re-executed when debugging.  (omalley via cutting)

 3. HADOOP-339.  Add method to JobClient API listing jobs that are
    not yet complete, i.e., that are queued or running.
    (Mahadev Konar via cutting)

 4. HADOOP-355.  Updates to the streaming contrib module, including
    API fixes, making reduce optional, and adding an input type for
    StreamSequenceRecordReader.  (Michel Tourn via cutting)

 5. HADOOP-358.  Fix a NPE bug in Path.equals().
    (Frédéric Bertin via cutting)

 6. HADOOP-327.  Fix ToolBase to not call System.exit() when
    exceptions are thrown.  (Hairong Kuang via cutting)

 7. HADOOP-359.  Permit map output to be compressed.
    (omalley via cutting)

 8. HADOOP-341.  Permit input URI to CopyFiles to use the HTTP
    protocol.  This lets one, e.g., more easily copy log files into
    DFS.  (Arun C Murthy via cutting)

 9. HADOOP-361.  Remove unix dependencies from streaming contrib
    module tests, making them pure java. (Michel Tourn via cutting)

10. HADOOP-354.  Make public methods to stop DFS daemons.
    (Barry Kaplan via cutting)

11. HADOOP-252.  Add versioning to RPC protocols.
    (Milind Bhandarkar via cutting)

12. HADOOP-356.  Add contrib to "compile" and "test" build targets, so
    that this code is better maintained. (Michel Tourn via cutting)

13. HADOOP-307.  Add smallJobsBenchmark contrib module.  This runs
    lots of small jobs, in order to determine per-task overheads.
    (Sanjay Dahiya via cutting)

14. HADOOP-342.  Add a tool for log analysis: Logalyzer.
    (Arun C Murthy via cutting)

15. HADOOP-347.  Add web-based browsing of DFS content.  The namenode
    redirects browsing requests to datanodes.  Content requests are
    redirected to datanodes where the data is local when possible.
    (Devaraj Das via cutting)

16. HADOOP-351.  Make Hadoop IPC kernel independent of Jetty.
    (Devaraj Das via cutting)

17. HADOOP-237.  Add metric reporting to DFS and MapReduce.  With only
    minor configuration changes, one can now monitor many Hadoop
    system statistics using Ganglia or other monitoring systems.
    (Milind Bhandarkar via cutting)

18. HADOOP-376.  Fix datanode's HTTP server to scan for a free port.
    (omalley via cutting)

19. HADOOP-260.  Add --config option to shell scripts, specifying an
    alternate configuration directory. (Milind Bhandarkar via cutting)

20. HADOOP-381.  Permit developers to save the temporary files for
    tasks whose names match a regular expression, to facilliate
    debugging.  (omalley via cutting)

21. HADOOP-344.  Fix some Windows-related problems with DF.
    (Konstantin Shvachko via cutting)

22. HADOOP-380.  Fix reduce tasks to poll less frequently for map
    outputs. (Mahadev Konar via cutting)

23. HADOOP-321.  Refactor DatanodeInfo, in preparation for
    HADOOP-306.  (Konstantin Shvachko & omalley via cutting)

24. HADOOP-385.  Fix some bugs in record io code generation.
    (Milind Bhandarkar via cutting)

25. HADOOP-302.  Add new Text class to replace UTF8, removing
    limitations of that class.  Also refactor utility methods for
    writing zero-compressed integers (VInts and VLongs).
    (Hairong Kuang via cutting)

26. HADOOP-335.  Refactor DFS namespace/transaction logging in
    namenode.   (Konstantin Shvachko via cutting)

27. HADOOP-375.  Fix handling of the datanode HTTP daemon's port so
    that multiple datanode's can be run on a single host.
    (Devaraj Das via cutting)

28. HADOOP-386.  When removing excess DFS block replicas, remove those
    on nodes with the least free space first.
    (Johan Oskarson via cutting)

29. HADOOP-389.  Fix intermittent failures of mapreduce unit tests.
    Also fix some build dependencies.
    (Mahadev & Konstantin via cutting)

30. HADOOP-362.  Fix a problem where jobs hang when status messages
    are recieved out-of-order.  (omalley via cutting)

31. HADOOP-394.  Change order of DFS shutdown in unit tests to
    minimize errors logged.  (Konstantin Shvachko via cutting)

32. HADOOP-396.  Make DatanodeID implement Writable.
    (Konstantin Shvachko via cutting)

33. HADOOP-377.  Permit one to add URL resources to a Configuration.
    (Jean-Baptiste Quenot via cutting)

34. HADOOP-345.  Permit iteration over Configuration key/value pairs.
    (Michel Tourn via cutting)

35. HADOOP-409.  Streaming contrib module: make configuration
    properties available to commands as environment variables.
    (Michel Tourn via cutting)

36. HADOOP-369.  Add -getmerge option to dfs command that appends all
    files in a directory into a single local file.
    (Johan Oskarson via cutting)

37. HADOOP-410.  Replace some TreeMaps with HashMaps in DFS, for
    a 17% performance improvement. (Milind Bhandarkar via cutting)

38. HADOOP-411.  Add unit tests for command line parser.
    (Hairong Kuang via cutting)

39. HADOOP-412.  Add MapReduce input formats that support filtering
    of SequenceFile data, including sampling and regex matching.
    Also, move JobConf.newInstance() to a new utility class.
    (Hairong Kuang via cutting)


Release 0.4.0 - 2006-06-28

 1. HADOOP-298.  Improved progress reports for CopyFiles utility, the
    distributed file copier.  (omalley via cutting)

 2. HADOOP-299.  Fix the task tracker, permitting multiple jobs to
    more easily execute at the same time.  (omalley via cutting)

 3. HADOOP-250.  Add an HTTP user interface to the namenode, running
    on port 50070. (Devaraj Das via cutting)

 4. HADOOP-123.  Add MapReduce unit tests that run a jobtracker and
    tasktracker, greatly increasing code coverage.
    (Milind Bhandarkar via cutting)

 5. HADOOP-271.  Add links from jobtracker's web ui to tasktracker's
    web ui.  Also attempt to log a thread dump of child processes
    before they're killed.  (omalley via cutting)

 6. HADOOP-210.  Change RPC server to use a selector instead of a
    thread per connection.  This should make it easier to scale to
    larger clusters.  Note that this incompatibly changes the RPC
    protocol: clients and servers must both be upgraded to the new
    version to ensure correct operation.  (Devaraj Das via cutting)

 7. HADOOP-311.  Change DFS client to retry failed reads, so that a
    single read failure will not alone cause failure of a task.
    (omalley via cutting)

 8. HADOOP-314.  Remove the "append" phase when reducing.  Map output
    files are now directly passed to the sorter, without first
    appending them into a single file.  Now, the first third of reduce
    progress is "copy" (transferring map output to reduce nodes), the
    middle third is "sort" (sorting map output) and the last third is
    "reduce" (generating output).  Long-term, the "sort" phase will
    also be removed.  (omalley via cutting)

 9. HADOOP-316.  Fix a potential deadlock in the jobtracker.
    (omalley via cutting)

10. HADOOP-319.  Fix FileSystem.close() to remove the FileSystem
    instance from the cache.  (Hairong Kuang via cutting)

11. HADOOP-135.  Fix potential deadlock in JobTracker by acquiring
    locks in a consistent order.  (omalley via cutting)

12. HADOOP-278.  Check for existence of input directories before
    starting MapReduce jobs, making it easier to debug this common
    error.  (omalley via cutting)

13. HADOOP-304.  Improve error message for
    UnregisterdDatanodeException to include expected node name.
   (Konstantin Shvachko via cutting)

14. HADOOP-305.  Fix TaskTracker to ask for new tasks as soon as a
    task is finished, rather than waiting for the next heartbeat.
    This improves performance when tasks are short.
    (Mahadev Konar via cutting)

15. HADOOP-59.  Add support for generic command line options.  One may
    now specify the filesystem (-fs), the MapReduce jobtracker (-jt),
    a config file (-conf) or any configuration property (-D).  The
    "dfs", "fsck", "job", and "distcp" commands currently support
    this, with more to be added.  (Hairong Kuang via cutting)

16. HADOOP-296.  Permit specification of the amount of reserved space
    on a DFS datanode.  One may specify both the percentage free and
    the number of bytes.  (Johan Oskarson via cutting)

17. HADOOP-325.  Fix a problem initializing RPC parameter classes, and
    remove the workaround used to initialize classes.
    (omalley via cutting)

18. HADOOP-328.  Add an option to the "distcp" command to ignore read
    errors while copying.  (omalley via cutting)

19. HADOOP-27.  Don't allocate tasks to trackers whose local free
    space is too low.  (Johan Oskarson via cutting)

20. HADOOP-318.  Keep slow DFS output from causing task timeouts.
    This incompatibly changes some public interfaces, adding a
    parameter to OutputFormat.getRecordWriter() and the new method
    Reporter.progress(), but it makes lots of tasks succeed that were
    previously failing.  (Milind Bhandarkar via cutting)


Release 0.3.2 - 2006-06-09

 1. HADOOP-275.  Update the streaming contrib module to use log4j for
    its logging.  (Michel Tourn via cutting)

 2. HADOOP-279.  Provide defaults for log4j logging parameters, so
    that things still work reasonably when Hadoop-specific system
    properties are not provided.  (omalley via cutting)

 3. HADOOP-280.  Fix a typo in AllTestDriver which caused the wrong
    test to be run when "DistributedFSCheck" was specified.
   (Konstantin Shvachko via cutting)

 4. HADOOP-240.  DFS's mkdirs() implementation no longer logs a warning
    when the directory already exists. (Hairong Kuang via cutting)

 5. HADOOP-285.  Fix DFS datanodes to be able to re-join the cluster
    after the connection to the namenode is lost.  (omalley via cutting)

 6. HADOOP-277.  Fix a race condition when creating directories.
   (Sameer Paranjpye via cutting)

 7. HADOOP-289.  Improved exception handling in DFS datanode.
    (Konstantin Shvachko via cutting)

 8. HADOOP-292.  Fix client-side logging to go to standard error
    rather than standard output, so that it can be distinguished from
    application output.  (omalley via cutting)

 9. HADOOP-294.  Fixed bug where conditions for retrying after errors
    in the DFS client were reversed.  (omalley via cutting)


Release 0.3.1 - 2006-06-05

 1. HADOOP-272.  Fix a bug in bin/hadoop setting log
    parameters. (omalley & cutting)

 2. HADOOP-274.  Change applications to log to standard output rather
    than to a rolling log file like daemons.  (omalley via cutting)

 3. HADOOP-262.  Fix reduce tasks to report progress while they're
    waiting for map outputs, so that they do not time out.
    (Mahadev Konar via cutting)

 4. HADOOP-245 and HADOOP-246.  Improvements to record io package.  
    (Mahadev Konar via cutting)

 5. HADOOP-276.  Add logging config files to jar file so that they're
    always found.  (omalley via cutting)


Release 0.3.0 - 2006-06-02

 1. HADOOP-208.  Enhance MapReduce web interface, adding new pages
    for failed tasks, and tasktrackers.  (omalley via cutting)

 2. HADOOP-204.  Tweaks to metrics package.  (David Bowen via cutting)

 3. HADOOP-209.  Add a MapReduce-based file copier.  This will
    copy files within or between file systems in parallel.
    (Milind Bhandarkar via cutting)

 4. HADOOP-146.  Fix DFS to check when randomly generating a new block
    id that no existing blocks already have that id.
    (Milind Bhandarkar via cutting)

 5. HADOOP-180. Make a daemon thread that does the actual task clean ups, so
    that the main offerService thread in the taskTracker doesn't get stuck
    and miss his heartbeat window. This was killing many task trackers as
    big jobs finished (300+ tasks / node). (omalley via cutting)

 6. HADOOP-200. Avoid transmitting entire list of map task names to
    reduce tasks.  Instead just transmit the number of map tasks and
    henceforth refer to them by number when collecting map output.
    (omalley via cutting)

 7. HADOOP-219. Fix a NullPointerException when handling a checksum
    exception under SequenceFile.Sorter.sort().  (cutting & stack)

 8. HADOOP-212. Permit alteration of the file block size in DFS.  The
    default block size for new files may now be specified in the
    configuration with the dfs.block.size property.  The block size
    may also be specified when files are opened.
    (omalley via cutting)

 9. HADOOP-218. Avoid accessing configuration while looping through
    tasks in JobTracker.  (Mahadev Konar via cutting)

10. HADOOP-161. Add hashCode() method to DFS's Block.
    (Milind Bhandarkar via cutting)

11. HADOOP-115. Map output types may now be specified.  These are also
    used as reduce input types, thus permitting reduce input types to
    differ from reduce output types.  (Runping Qi via cutting)

12. HADOOP-216. Add task progress to task status page.
    (Bryan Pendelton via cutting)

13. HADOOP-233.  Add web server to task tracker that shows running
    tasks and logs.  Also add log access to job tracker web interface.
    (omalley via cutting)

14. HADOOP-205.  Incorporate pending tasks into tasktracker load
    calculations.  (Mahadev Konar via cutting)

15. HADOOP-247.  Fix sort progress to better handle exceptions.
    (Mahadev Konar via cutting)

16. HADOOP-195.  Improve performance of the transfer of map outputs to
    reduce nodes by performing multiple transfers in parallel, each on
    a separate socket.  (Sameer Paranjpye via cutting)

17. HADOOP-251.  Fix task processes to be tolerant of failed progress
    reports to their parent process.  (omalley via cutting)

18. HADOOP-325.  Improve the FileNotFound exceptions thrown by
    LocalFileSystem to include the name of the file.
    (Benjamin Reed via cutting)

19. HADOOP-254.  Use HTTP to transfer map output data to reduce
    nodes.  This, together with HADOOP-195, greatly improves the
    performance of these transfers.  (omalley via cutting)

20. HADOOP-163.  Cause datanodes that\ are unable to either read or
    write data to exit, so that the namenode will no longer target
    them for new blocks and will replicate their data on other nodes.
    (Hairong Kuang via cutting)

21. HADOOP-222.  Add a -setrep option to the dfs commands that alters
    file replication levels.  (Johan Oskarson via cutting)

22. HADOOP-75.  In DFS, only check for a complete file when the file
    is closed, rather than as each block is written.
    (Milind Bhandarkar via cutting)

23. HADOOP-124. Change DFS so that datanodes are identified by a
    persistent ID rather than by host and port.  This solves a number
    of filesystem integrity problems, when, e.g., datanodes are
    restarted.  (Konstantin Shvachko via cutting)

24. HADOOP-256.  Add a C API for DFS.  (Arun C Murthy via cutting)

25. HADOOP-211.  Switch to use the Jakarta Commons logging internally,
    configured to use log4j by default.  (Arun C Murthy and cutting)

26. HADOOP-265.  Tasktracker now fails to start if it does not have a
    writable local directory for temporary files.  In this case, it
    logs a message to the JobTracker and exits. (Hairong Kuang via cutting)

27. HADOOP-270.  Fix potential deadlock in datanode shutdown.
    (Hairong Kuang via cutting)

Release 0.2.1 - 2006-05-12

 1. HADOOP-199.  Fix reduce progress (broken by HADOOP-182).
    (omalley via cutting)

 2. HADOOP-201.  Fix 'bin/hadoop dfs -report'.  (cutting)

 3. HADOOP-207.  Fix JDK 1.4 incompatibility introduced by HADOOP-96.
    System.getenv() does not work in JDK 1.4.  (Hairong Kuang via cutting)


Release 0.2.0 - 2006-05-05

 1. Fix HADOOP-126. 'bin/hadoop dfs -cp' now correctly copies .crc
    files.  (Konstantin Shvachko via cutting)

 2. Fix HADOOP-51. Change DFS to support per-file replication counts.
    (Konstantin Shvachko via cutting)

 3. Fix HADOOP-131.  Add scripts to start/stop dfs and mapred daemons.
    Use these in start/stop-all scripts.  (Chris Mattmann via cutting)

 4. Stop using ssh options by default that are not yet in widely used
    versions of ssh.  Folks can still enable their use by uncommenting
    a line in conf/hadoop-env.sh. (cutting)

 5. Fix HADOOP-92.  Show information about all attempts to run each
    task in the web ui.  (Mahadev konar via cutting)

 6. Fix HADOOP-128.  Improved DFS error handling. (Owen O'Malley via cutting)

 7. Fix HADOOP-129.  Replace uses of java.io.File with new class named
    Path.  This fixes bugs where java.io.File methods were called
    directly when FileSystem methods were desired, and reduces the
    likelihood of such bugs in the future.  It also makes the handling
    of pathnames more consistent between local and dfs FileSystems and
    between Windows and Unix. java.io.File-based methods are still
    available for back-compatibility, but are deprecated and will be
    removed once 0.2 is released. (cutting)

 8. Change dfs.data.dir and mapred.local.dir to be comma-separated
    lists of directories, no longer be space-separated. This fixes
    several bugs on Windows. (cutting)

 9. Fix HADOOP-144.  Use mapred task id for dfs client id, to
    facilitate debugging.  (omalley via cutting)

10. Fix HADOOP-143.  Do not line-wrap stack-traces in web ui.
    (omalley via cutting)

11. Fix HADOOP-118.  In DFS, improve clean up of abandoned file
    creations.  (omalley via cutting)

12. Fix HADOOP-138.  Stop multiple tasks in a single heartbeat, rather
    than one per heartbeat.  (Stefan via cutting)

13. Fix HADOOP-139.  Remove a potential deadlock in
    LocalFileSystem.lock().  (Igor Bolotin via cutting)

14. Fix HADOOP-134.  Don't hang jobs when the tasktracker is
    misconfigured to use an un-writable local directory.  (omalley via cutting)

15. Fix HADOOP-115.  Correct an error message.  (Stack via cutting)

16. Fix HADOOP-133.  Retry pings from child to parent, in case of
    (local) communcation problems.  Also log exit status, so that one
    can distinguish patricide from other deaths.  (omalley via cutting)

17. Fix HADOOP-142.  Avoid re-running a task on a host where it has
    previously failed.  (omalley via cutting)

18. Fix HADOOP-148.  Maintain a task failure count for each
    tasktracker and display it in the web ui.  (omalley via cutting)

19. Fix HADOOP-151.  Close a potential socket leak, where new IPC
    connection pools were created per configuration instance that RPCs
    use.  Now a global RPC connection pool is used again, as
    originally intended.  (cutting)

20. Fix HADOOP-69.  Don't throw a NullPointerException when getting
    hints for non-existing file split.  (Bryan Pendelton via cutting)

21. Fix HADOOP-157.  When a task that writes dfs files (e.g., a reduce
    task) failed and was retried, it would fail again and again,
    eventually failing the job.  The problem was that dfs did not yet
    know that the failed task had abandoned the files, and would not
    yet let another task create files with the same names.  Dfs now
    retries when creating a file long enough for locks on abandoned
    files to expire.  (omalley via cutting)

22. Fix HADOOP-150.  Improved task names that include job
    names. (omalley via cutting)

23. Fix HADOOP-162.  Fix ConcurrentModificationException when
    releasing file locks. (omalley via cutting)

24. Fix HADOOP-132.  Initial check-in of new Metrics API, including 
    implementations for writing metric data to a file and for sending
    it to Ganglia.  (David Bowen via cutting)

25. Fix HADOOP-160.  Remove some uneeded synchronization around
    time-consuming operations in the TaskTracker.  (omalley via cutting)

26. Fix HADOOP-166.  RPCs failed when passed subclasses of a declared
    parameter type.  This is fixed by changing ObjectWritable to store
    both the declared type and the instance type for Writables.  Note
    that this incompatibly changes the format of ObjectWritable and
    will render unreadable any ObjectWritables stored in files.
    Nutch only uses ObjectWritable in intermediate files, so this
    should not be a problem for Nutch.  (Stefan & cutting)

27. Fix HADOOP-168.  MapReduce RPC protocol methods should all declare
    IOException, so that timeouts are handled appropriately.
    (omalley via cutting)

28. Fix HADOOP-169.  Don't fail a reduce task if a call to the
    jobtracker to locate map outputs fails.  (omalley via cutting)

29. Fix HADOOP-170.  Permit FileSystem clients to examine and modify
    the replication count of individual files.  Also fix a few
    replication-related bugs. (Konstantin Shvachko via cutting)

30. Permit specification of a higher replication levels for job
    submission files (job.xml and job.jar).  This helps with large
    clusters, since these files are read by every node.  (cutting)

31. HADOOP-173.  Optimize allocation of tasks with local data.  (cutting)

32. HADOOP-167.  Reduce number of Configurations and JobConf's
    created.  (omalley via cutting)

33. NUTCH-256.  Change FileSystem#createNewFile() to create a .crc
    file.  The lack of a .crc file was causing warnings.  (cutting)

34. HADOOP-174.  Change JobClient to not abort job until it has failed
    to contact the job tracker for five attempts, not just one as
    before.  (omalley via cutting)

35. HADOOP-177.  Change MapReduce web interface to page through tasks.
    Previously, when jobs had more than a few thousand tasks they
    could crash web browsers.  (Mahadev Konar via cutting)

36. HADOOP-178.  In DFS, piggyback blockwork requests from datanodes
    on heartbeat responses from namenode.  This reduces the volume of
    RPC traffic.  Also move startup delay in blockwork from datanode
    to namenode.  This fixes a problem where restarting the namenode
    triggered a lot of uneeded replication. (Hairong Kuang via cutting)

37. HADOOP-183.  If the DFS namenode is restarted with different
    minimum and/or maximum replication counts, existing files'
    replication counts are now automatically adjusted to be within the
    newly configured bounds. (Hairong Kuang via cutting)

38. HADOOP-186.  Better error handling in TaskTracker's top-level
    loop.  Also improve calculation of time to send next heartbeat.
    (omalley via cutting)

39. HADOOP-187.  Add two MapReduce examples/benchmarks.  One creates
    files containing random data.  The second sorts the output of the
    first.  (omalley via cutting)

40. HADOOP-185.  Fix so that, when a task tracker times out making the
    RPC asking for a new task to run, the job tracker does not think
    that it is actually running the task returned.  (omalley via cutting)

41. HADOOP-190.  If a child process hangs after it has reported
    completion, its output should not be lost.  (Stack via cutting)

42. HADOOP-184. Re-structure some test code to better support testing
    on a cluster.  (Mahadev Konar via cutting)

43. HADOOP-191  Add streaming package, Hadoop's first contrib module.
    This permits folks to easily submit MapReduce jobs whose map and
    reduce functions are implemented by shell commands.  Use
    'bin/hadoop jar build/hadoop-streaming.jar' to get details.
    (Michel Tourn via cutting)

44. HADOOP-189.  Fix MapReduce in standalone configuration to
    correctly handle job jar files that contain a lib directory with
    nested jar files.  (cutting)

45. HADOOP-65.  Initial version of record I/O framework that enables
    the specification of record types and generates marshalling code
    in both Java and C++.  Generated Java code implements
    WritableComparable, but is not yet otherwise used by
    Hadoop. (Milind Bhandarkar via cutting)

46. HADOOP-193.  Add a MapReduce-based FileSystem benchmark.
    (Konstantin Shvachko via cutting)

47. HADOOP-194.  Add a MapReduce-based FileSystem checker.  This reads
    every block in every file in the filesystem.  (Konstantin Shvachko
    via cutting)

48. HADOOP-182.  Fix so that lost task trackers to not change the
    status of reduce tasks or completed jobs.  Also fixes the progress
    meter so that failed tasks are subtracted. (omalley via cutting)

49. HADOOP-96.  Logging improvements.  Log files are now separate from
    standard output and standard error files.  Logs are now rolled.
    Logging of all DFS state changes can be enabled, to facilitate
    debugging.  (Hairong Kuang via cutting)


Release 0.1.1 - 2006-04-08

 1. Added CHANGES.txt, logging all significant changes to Hadoop.  (cutting)

 2. Fix MapReduceBase.close() to throw IOException, as declared in the
    Closeable interface.  This permits subclasses which override this
    method to throw that exception. (cutting)

 3. Fix HADOOP-117.  Pathnames were mistakenly transposed in
    JobConf.getLocalFile() causing many mapred temporary files to not
    be removed.  (Raghavendra Prabhu via cutting)
 
 4. Fix HADOOP-116. Clean up job submission files when jobs complete.
    (cutting)

 5. Fix HADOOP-125. Fix handling of absolute paths on Windows (cutting)

Release 0.1.0 - 2006-04-01

 1. The first release of Hadoop.