[HIVE-17006] LLAP: Parquet caching v1

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 3.0.0
• Assignee: Sergey Shelukhin
• Reporter: Sergey Shelukhin
• Created: 30/Jun/17 21:36
• Updated: 22/May/18 23:59
• Resolved: 31/Aug/17 21:27

Description

There are multiple options to do Parquet caching in LLAP:
1) Full elevator (too intrusive for now).
2) Page based cache like ORC (requires some changes to Parquet or copy-pasted).
3) Cache disk data on column chunk level as is.

Given that Parquet reads at column chunk granularity, (2) is not as useful as for ORC, but still a good idea. I messaged the dev list about it but didn't get a response, we may follow up later.

For now, do (3).

Issue Links

• is depended upon by: HIVE-18453 ACID: Add "CREATE TRANSACTIONAL TABLE" syntax to unify ACID ORC & Parquet support
• is related to: HIVE-17423 LLAP Parquet caching - support file ID in splits
• links to: RB