Skip to main content

Release 3.1 (2026-9-30)

MR3​

Compression​

  • MR3 no longer uses the Hadoop compression library for intermediate data compression. The usage of compression remains unchanged: tez.runtime.compress enables compression, and tez.runtime.compress.codec selects the compression codec.

  • MR3 supports the following compression codecs:

    • org.apache.hadoop.io.compress.SnappyCodec
    • org.apache.hadoop.io.compress.ZStandardCodec
    • org.apache.hadoop.io.compress.Lz4Codec

  • The native libraries for these codecs are included in the MR3 JAR files, so they do not need to be installed separately on worker nodes.

MR3-UI​

  • MR3-UI is now served by the HiveServer2 Web UI.

Hive on MR3​

Bug fixes​

  • Tez runtime configuration keys set from Beeline are now correctly delivered to MR3.

Query results and temporary files​

  • MR3 workers can return top-level query results as DAG output instead of writing temporary FileSink files on a distributed file system. HiveServer2 materializes the DAG output in local scratch storage.
  • Dummy table scans no longer create dummy files in scratch storage.
  • Hive on MR3 avoids obsolete scratch-directory creation while initializing map vertices and skips temporary-directory creation for DAG-output sinks.

Scratch directories​

  • Hive on MR3 no longer uses scratch directories on a distributed file system for query execution.
  • MR3 still requires the staging directory specified by mr3.am.staging-dir to support DAG recovery in DAGAppMaster.

Miscellaneous​

  • Hive on MR3 no longer uses the internal Hive configuration variable hive.resource.use.hdfs.location.