Ashu Pachauri created HBASE-17041:
-------------------------------------
Summary: VerifyReplication reporting false inconsistencies
Key: HBASE-17041
URL: https://issues.apache.org/jira/browse/HBASE-17041
Project: HBase
Issue Type: Bug
Components: Replication, Scanners
Affects Versions: 1.3.0
Reporter: Ashu Pachauri
Priority: Critical
VerifyReplication often report inconsistencies that are not there. The reported
inconsistencies span all types i.e. CONTENT_DIFFERENT_ROWS,
ONLY_IN_SOURCE_TABLE_ROWS, ONLY_IN_PEER_TABLE_ROWS. I tried an approach
similar to HBASE-16423 to re-compare using a Get on the offending row, and the
inconsistencies turn out to be false.
This may suggest an issue with the VerifyReplication setup (unlikely, as and
immediate Get based comparison succeeds), or scanners losing data.
The configurations that I tried were (modified VerifyReplication to take a row
range as input):
1. Column families with only one max version with KEEP_DELETED_CELLS set to
true and purge time to 7 days. VerifyReplication run was on a small time range
two days ago. Multiple regions: reports inconsistencies, one region: reports
fine (could just be due to smaller volume of data).
2. Column families with Integer.MAX_VALUE number of max versions,
KEEP_DELETED_CELLS set to true, purge time set to 7 days and verification
timerange two days ago. Both one region and multiple region setups report
inconsistencies.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
