[HIVE-23671] MSCK repair should handle transactional tables in certain usecases

• Type: Improvement
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Component/s: Metastore
• Labels: pull-request-available
• Assignee: Peter Varga
• Reporter: Peter Varga
• Created: 10/Jun/20 09:37
• Updated: 20/Jul/20 08:37
• Resolved: 20/Jul/20 08:37

Description

The MSCK REPAIR tool does not handle transactional tables too well. It can find and add new partitions the same way as for non-transactional tables, but since the writeId differences are not handled, the data can not read back from the new partitions.

We could handle some usecases when the writeIds in the HMS and the underlying data are not conflicting. If the HMS does not contains allocated writes for the table we can seed the table with the writeIds read from the directory structrure.

Real life use cases could be:

• Copy data files from one cluster to another with different HMS, create the table and call MSCK REPAIR
• If the HMS db is lost, recreate the table and call MSCK REPAIR

 

Issue Links

• links to: GitHub Pull Request #1087