Release 0.5.0

API change

• Remove module-level api like dpark.textFile.
• Support Streaming shuffle and Disk shuffle (Experimental, compatible).

Fixes

• Bug when parsing mfs chunk info.

Improvement

• Better broadcast impl using shared memory for tasks on the same slave to reduce memory cost.
• Better offer-matching logic for MesosScheduler which remember bad slaves.
• Refactor: style and layout.

New Feature

• Multi segment dump to save memory.
• Gather statics for stage.
• Support run tests/test_rdd on mesos.
• Add colorful progress bar for dpark.
• Support mesos role.
• Support multi named mesos master in conf.
• Loghub for admin.

Release 0.4.2

• Support Python3 & PyPy
• Support MooseFS 3.x & refactor on file-system interface

Release 0.4.1

• Enhancement for the containerizer in DPark
• Use broadcast when parallelize big dataset
• Fix missing line bug for bzip2 files
• Add TopByKey in RDD
• Other minor bugs

Release 0.4.0

• Bugfix: deserialize error of old-style class.
• Refactor beansdb RDD
• Web UI support for dpark
• Use pymesos >= 0.2.0
• Eager serialize values of ParallelCollection