Rahul Kumar created PHOENIX-8065:
------------------------------------
Summary: Provide option to ignore cell timestamps in
PhoenixSyncTableTool, and sync SYSTEM.SEQUENCE
Key: PHOENIX-8065
URL: https://issues.apache.org/jira/browse/PHOENIX-8065
Project: Phoenix
Issue Type: Improvement
Affects Versions: 5.2.3, 5.4.0, 5.3.3
Reporter: Rahul Kumar
Assignee: Rahul Kumar
Fix For: 5.2.3, 5.4.0, 5.3.3
The chunk checksum includes the cell timestamp. Tables such as
{{SYSTEM.SEQUENCE}} are rewritten on each cluster with the same values at
different timestamps, so a correct copy is reported as a mismatch.
{{SYSTEM.SEQUENCE}} also cannot be synced today: the generated {{SELECT}} does
not quote the table name, and {{SEQUENCE}} is a reserved word, so query
planning fails.
This is needed to validate {{SYSTEM.SEQUENCE}} during an FKP migration.
Change
Add {{--exclude-cell-timestamp-from-checksum}} (off by default). When set, the
region scanner omits the cell timestamp from the SHA-256 digest. The hash still
covers the row key, family, qualifier, type, value, and tags. Source and target
rows with the same data and different timestamps then compare as equal. A real
value difference still mismatches.
Quote the schema and table in the input {{SELECT}}
({{{}SchemaUtil.getEscapedTableName{}}}) so a name that is a Phoenix reserved
word, including {{{}SYSTEM.SEQUENCE{}}}, parses. The job completion log
includes the table name.
Tests
* Timestamp-only differences verify with the flag on, and value differences
still mismatch.
* {{testSyncTableOnSystemSequence}} creates sequences on both clusters,
advances one on the source, and checks that the value drift is reported.
* Unit tests for the flag default, parsing, and digest behavior with and
without the timestamp.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)