Andrew Kyle Purtell created PHOENIX-7834:
--------------------------------------------
Summary: CDCBaseIT do not force-delete a never-upserted row
Key: PHOENIX-7834
URL: https://issues.apache.org/jira/browse/PHOENIX-7834
Project: Phoenix
Issue Type: Sub-task
Components: test
Reporter: Andrew Kyle Purtell
Assignee: Andrew Kyle Purtell
Fix For: 5.4.0, 5.3.1
{{CDCBaseIT.generateMutations}} has a second-half "force a delete if there was
none so far" branch that flips {{isDelete=true}} without checking whether the
candidate row was ever upserted in this run, so when random batch participation
has skipped a key in every prior batch the forced {{DELETE}} lands on a
non-existent row and produces an HBase tombstone with no matching put. The data
table view sees no change, and a later legitimate upsert for the same key
surfaces as the next CDC change event on a {{forView=true}} CDC while the
expected change list still has the tombstone marked as a delete, tripping
{{CDCBaseIT.verifyChangesViaSCN}} with {{{}ComparisonFailure expected:<delete>
but was:<upsert> in CDCQueryIT.testSelectGeneric[forView=true,
encodingScheme=TWO_BYTE_QUALIFIERS, multitenant=false, tableSaltBuckets=null,
withSchemaName=false, caseSensitiveNames=false]{}}}. The fix gates the
forced-delete branch on {{mutatedRows.contains(rows.get(j))}} so a delete is
only ever forced on a row that has already been upserted.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
