[FLINK-29295] Clear RecordWriter slower to avoid causing frequent compaction conflicts

• Type: Bug
• Status: Closed
• Priority: Major
• Resolution: Fixed
• Fix Version/s: table-store-0.3.0, table-store-0.2.1
• Component/s: Table Store
• Labels: pull-request-available
• Assignee: Jingsong Lee
• Reporter: Jingsong Lee
• Created: 14/Sep/22 02:20
• Updated: 14/Sep/22 06:27
• Resolved: 14/Sep/22 06:27

Description

In AbstractTableWrite, the Writer is cleaned up as soon as no new files are generated, which may lead to the changes generated after the compaction have not been committed, but the new data from the next checkpoint comes to create a new writer, which conflicts with the changes generated in the next round of checkpoint and the previous round, resulting in an exception:

Caused by: java.lang.IllegalStateException: Trying to delete file {org.apache.flink.table.data.binary.BinaryRowData@5759f99e, 0, 0, data-7bf2498e-d0a1-42a5-97b7-b3860f10b076-0.orc} which is not previously added. Manifest might be corrupted.

Issue Links

• links to: GitHub Pull Request #294