Rahul Kumar created PHOENIX-8061:
------------------------------------
Summary: PhoenixSyncTableTool skips rows after an empty or dummy
scanner result
Key: PHOENIX-8061
URL: https://issues.apache.org/jira/browse/PHOENIX-8061
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
{{PhoenixSyncTableTool}} can skip rows when a lower scanner returns an empty
result or a Phoenix dummy row in the middle of a region.
In {{{}PhoenixSyncTableRegionScanner{}}}, that result broke out of the chunk
loop. If no real rows had been accumulated, the scanner returned {{{}false{}}},
which the client treats as the end of the scan. Rows after the empty or dummy
result in that region were never hashed.
On the client, {{ChunkScannerContext.getNextChunk()}} treated an empty
{{Result}} the same way and stopped reading further chunks.
Change
* Log empty rows and dummy rows separately, including the region, rows and
bytes accumulated so far, the current chunk bounds, and whether more rows
remain.
* When the call accumulated no real rows but the lower scanner still has more
rows, yield the RPC ({{{}PhoenixScannerContext.setReturnImmediately{}}}) and
return {{hasMoreRows}} so the client re-invokes the scanner instead of ending
the scan.
* In {{{}getNextChunk(){}}}, skip empty results and keep calling
{{scanner.next()}} until a real chunk or a true end of scan.
Logging also now includes the table name, source and target chunk bounds, row
counts, and source and target rows scanned during a row-level diff, so a
mismatch can be traced back to the scan that produced it.
--
This message was sent by Atlassian Jira
(v10.3.26#10030026)