These release notes cover new developer and user-facing incompatibilities, important issues, features, and major improvements.
Ensure that SequenceFileOutputFormat isn’t tied to Writables and can be used with other Serialization frameworks.
HADOOP-4433 | Major | Improve data loader for collecting metrics and log files from hadoop and system
Added startup and shutdown script
Changed reporting in the NameNode Web UI to more closely reflect the behavior of the re-balancer. Removed no longer used config parameter dfs.datanode.du.pct from hadoop-default.xml.
Made Configuration Writable and rename the old write method to writeXml.
Changed command “hadoop dfsadmin -report” to be consistent with Web UI for both Namenode and Datanode reports. “Total raw bytes” is changed to “Configured Capacity”. “Present Capacity” is newly added to indicate the present capacity of the DFS. “Remaining raw bytes” is changed to “DFS Remaining”. “Used raw bytes” is changed to “DFS Used”. “% used” is changed to “DFS Used%”. Applications that parse command output should be reviewed.
Removed the deprecated class org.apache.hadoop.fs.ShellCommand.
Improved Hive metastore and ql to use the refactored SerDe library.
Changed job history format to add a dot at end of each line.
Implemented getFileChecksum(Path) in HftpFileSystemfor distcp support.
Introduced new SerDe library for src/contrib/hive.
Changed scripts to pass initialization script for EC2 instances at boot time (as EC2 user data) rather than embedding initialization information in the EC2 image. This change makes it easy to customize the hadoop-site.xml file for your cluster before launch, by editing the hadoop-ec2-init-remote.sh script, or by setting the environment variable USER_DATA_FILE in hadoop-ec2-env.sh to run a script of your choice.
Changed DataNode protocol version without impact to clients other than to compel use of current version of client application.
Added time, permission and user attribute support to libhdfs.
Added LIMIT to Hive query language.
Introduced “EXPLAIN” plan for Hive.
Introduced new configuration parameter mapred.max.tasks.per.job to specifie the maximum number of tasks per job.
Added a synthetic load generation facility to the test directory.
Implemented MD5-of-xxxMD5-of-yyyCRC32 which is a distributed file checksum algorithm for HDFS, where xxx is the number of CRCs per block and yyy is the number of bytes per CRC.
Changed DistCp to use file checksum for comparing files if both source and destination FileSystem(s) support getFileChecksum(…).
Added getEscapedCompactString() and fromEscapedCompactString() to Counters.java to represent counters as Strings and to reconstruct the counters from the Strings.
Modified libhdfs to return NULL or error code when unrecoverable error occurs rather than exiting itself.
Added new FileSystem APIs: FileChecksum and FileSystem.getFileChecksum(Path).
Added a new option -delete to DistCp so that if the files/directories exist in dst but not in src will be deleted. It uses FsShell to do delete, so that it will use trash if the trash is enable.
Introducted byte space quotas for directories. The count shell command modified to report both name and byte quotas.
Changed TaskScheduler to expose API for Web UI and Command Line Tool.
Added a check to fsck options to make sure -files is not the first option so as to resolve conflicts with GenericOptionsParser.
Improved Fuse-dfs better error message if llibhdfs.so doesn’t exist.
Changed DistCp error messages when there is a RemoteException. Changed the corresponding return value from -999 to -3.
Added two new options -filelimit <n> and -sizelimit <n> to DistCp for limiting the total number of files and the total size in bytes, respectively.
Added a configuration property hadoop.http.filter.initializers and a class org.apache.hadoop.http.FilterInitializer for supporting servlet filter. Cluster administrator could possibly configure customized filters for their web site.
Changed streaming tasks to adhere to task timeout value specified in the job configuration.
Introduced new config parameter org.apache.hadoop.mapred.SkipBadRecords.setMapperMaxSkipRecords to set range of records to be skipped in the neighborhood of a failed record.
Skipped records can optionally be written to the HDFS. Refer org.apache.hadoop.mapred.SkipBadRecords.setSkipOutputPath for setting the output path.
Changed Fuse configuration to use mount options.
Changed exit code from hadoop.fs.FsShell -test to match the usual Unix convention.
Introduced Fair Scheduler.
Added a JobConf constructor that disables loading default configurations so as to take all default values from the JobTracker’s configuration.
Changed streaming StreamJob and Submitter to implement Tool and Configurable, and to use GenericOptionsParser arguments -fs, -jt, -conf, -D, -libjars, -files, and -archives. Deprecated -jobconf, -cacheArchive, -dfs, -cacheArchive, -additionalconfspec, from streaming and pipes in favor of the generic options. Removed from streaming -config, -mapred.job.tracker, and -cluster.
Introduced Chukwa data collection and analysis framework.
Adds a new contrib, bash-tab-completion, which enables bash tab completion for the bin/hadoop script. See the README file in the contrib directory for the installation.
Introduced ChainMapper and the ChainReducer classes to allow composing chains of Maps and Reduces in a single Map/Reduce job, something like MAP+ REDUCE MAP*.
Added an ability in HOD to start multiple workers (TaskTrackers and/or DataNodes) per node to assist testing and simulation of scale. A configuration variable ringmaster.workers_per_ring was added to specify the number of workers to start.
Allowed user to overwrite clone function in a subclass of TaggedMapOutput class.
Removed the following deprecated methods from JobConf: addInputPath(Path) getInputPaths() getMapOutputCompressionType() getOutputPath() getSystemDir() setInputPath(Path) setMapOutputCompressionType(CompressionType style) setOutputPath(Path)
Removed deprecated org.apache.hadoop.mapred.OutputFormatBase.
Introduced support for bzip2 compressed files.
Introduced Hive Data Warehouse built on top of Hadoop that enables structuring Hadoop files as tables and partitions and allows users to query this data through a SQL like language using a command line interface.
Removed deprecated methods for mapred.combine.once functionality.
Added FailMon as a contrib project for hardware failure monitoring and analysis, under /src/contrib/failmon. Created User Manual and Quick Start Guide.
Added the ability to kill process trees transgressing memory limits. TaskTracker uses the configuration parameters introduced in HADOOP-3759. In addition, mapred.tasktracker.taskmemorymanager.monitoring-interval specifies the interval for which TT waits between cycles of monitoring tasks’ memory usage, and mapred.tasktracker.procfsbasedprocesstree.sleeptime-before-sigkill specifies the time TT waits for sending a SIGKILL to a process-tree that has overrun memory limits, after it has been sent a SIGTERM.
Improved error reporting for libhdfs so permission problems now return EACCES.
Extended file globbing alternation to cross path components. For example, {/a/b,/c/d} expands to a path that matches the files /a/b and /c/d.
Introduce write support for Fuse; requires Linux kernel 2.6.15 or better.
Changed reducers to fetch maps in the same order for a given host to speed up identification of the faulty maps; reducers still randomize the host selection to distribute load.
Introduced Capacity Task Scheduler.
Added the ability to chose between many schedulers, and to limit the number of running tasks per job.
Introduced recovery of jobs when JobTracker restarts. This facility is off by default. Introduced config parameters mapred.jobtracker.restart.recover, mapred.jobtracker.job.history.block.size, and mapred.jobtracker.job.history.buffer.size.
Moved task file promotion to the Task. When the task has finished, it will do a commit and is declared SUCCEDED. Job cleanup is done by a separate task. Job is declared SUCCEDED/FAILED after the cleanup task has finished. Added public classes org.apache.hadoop.mapred.JobContext, TaskAttemptContext, OutputCommitter and FileOutputCommiitter. Added public APIs: public OutputCommitter getOutputCommitter() and public void setOutputCommitter(Class<? extends OutputCommitter> theClass) in org.apache.hadoop.mapred.JobConf
Introduced MultipleOutputs class so Map/Reduce jobs can write data to different output files. Each output can use a different OutputFormat. Outpufiles are created within the job output directory. FileOutputFormat.getPathForCustomFile() creates a filename under the outputdir that is named with the task ID and task type (i.e. myfile-r-00001).
Introduced additional log records for data transfers.
Added a partitioner that effects a total order of output data, and an input sampler for generating the partition keyset for TotalOrderPartitioner for when the map’s input keytype and distribution approximates its output.
Improved space reporting for NameNode Web UI. Applications that parse the Web UI output should be reviewed.
Introduced LZOP codec.
Added support for c1.* instance types and associated kernels for EC2.
Hadoop now requires Java 6.
Introduced numerical key comparison for streaming.
Added HDFS file access times. By default, access times will be precise to the most recent hour boundary. A configuration parameter dfs.access.time.precision (milliseconds) is used to control this precision. Setting a value of 0 will disable persisting access times for HDFS files.
bzip2 provided as codec in 0.19.0 https://issues.apache.org/jira/browse/HADOOP-3646
Introduced append operation for HDFS files.
Enabled task JVMs to be reused via the job config mapred.job.reuse.jvm.num.tasks.
Introduced record skipping where tasks fail on certain records. (org.apache.hadoop.mapred.SkipBadRecords)