Release 0.77

    Streaming Index Joins

    Index joins will now switch to use a key-by-key streaming join if index results fail to fit in the allocated index memory space.

    Joins where both tables are distributed are now supported. This allows larger tables to be joined, and can be enabled with the flag. It may perform worse than the existing broadcast join implementation because it requires redistributing both tables. This feature is still experimental, and should be used with caution.

    Hive Changes

    • Handle spurious when closing S3 input streams

    • Add support for ORC, DWRF and Parquet in Hive

    • Fix performance regression in Hive when reading columns

    • Fix Kafka handling of default port

    General Changes

    • Fix race condition in scheduler that could cause queries to hang

    • Add support for string partition keys in Cassandra

    • Add support for variable arity functions

    • Add support for for all types