[HBASE-22646] boundaries errors in the overlap after using hbck fix

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: 1.3.1, 2.1.0
• Component/s: hbck
• Assignee: Unassigned
• Reporter: Bo Cui
• Created: 02/Jul/19 07:39
• Updated: 03/Jul/19 16:01
• Resolved: 03/Jul/19 01:42

Description

When I use hbck fix overlap, there is data in the region that doesn't belong to it.

reproduction steps:

1、create t1 and set COMPACTION_ENABLED=false
2、put r001 r002 r003 r004
3、flush t1 and flush meta
4、 major_compact meta, and get new meta hfile1
5、 split t1, r0022
6、 flush meta, and get new meta hfile2
7、 stop cluster，delet all WALs，, delete hfile2
8、 restart hbase
9、 split t1,r0022，and hdfs has 5 regions(1 parent, 2 old daughter, 2 new daughter), meta has 3 region(1 parent, 2 new daughter)
10、use hbck to repair until t1 is healthy

the result： 514476289cd49db63ab7ef27e944185b has no data.

solution:

Issue Links

• duplicates: HBASE-21920 Ignoring 'empty' end_key while calculating end_key for new region in HBCK -fixHdfsOverlaps command can cause data loss