Bytecode Compiler

This version includes new infrastructure for bytecode compilation, and lays the groundwork for future improvements. There should be no impact in performance or correctness with the new code, but we have added a flag to revert to the old implementation in case of issues. To do so, add compiler.new-bytecode-generator-enabled=false to etc/config.properties in the coordinator and workers.

Hive Storage Format

The storage format to use when writing data to Hive can now be configured via the hive.storage-format option in your Hive catalog properties file. Valid options are RCBINARY, RCTEXT, SEQUENCEFILE and TEXTFILE. The default format if the property is not set is RCBINARY.

General Changes

• Show column comments in DESCRIBE
• Add try_cast() which works like cast() but returns null if the cast fails
• nullif now correctly returns a value with the type of the first argument
• Fix an issue with timezone_hour() returning results in milliseconds instead of hours
• Show a proper error message when analyzing queries with non-equijoin clauses
• Improve “too many failures” error message when coordinator can’t talk to workers
• Minor optimization of json_size() function
• Improve feature normalization algorithm for machine learning functions
• Add exponential back-off to the S3 FileSystem retry logic
• Improve CPU efficiency of semi-joins