[IMPALA-759] Consider switching to dictionary encoding for all NULL pages for parquet writer.

• Type: Task
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: Impala 1.2.1, Impala 1.2.3
• Fix Version/s: Impala 1.3
• Assignee: Nong Li
• Reporter: Nong Li
• Created: 22/Jan/14 02:15
• Updated: 05/Mar/14 16:37
• Resolved: 05/Mar/14 16:37

Description

If the page is all NULL, the data encoding doesn't matter since there are no data values (everything is stored in the definition levels).

Impala defaults to dictionary, so the page is written as dictionary encoded. This is unreadable by the the version of parquet in CDH4. We could
write the page as being PLAIN encoded which would be compatible. There is no compat concern on Impala's side.

More details:
https://groups.google.com/forum/#!topic/parquet-dev/Q-TPuA2rGk0