Martijn Visser created FLINK-40967:
--------------------------------------
Summary: Restore test files of
OVER_AGGREGATE_TIME_BOUNDED_NON_PARTITIONED_ROWS_WITH_OUT_OF_ORDER_RECORDS
cannot be regenerated
Key: FLINK-40967
URL: https://issues.apache.org/jira/browse/FLINK-40967
Project: Flink
Issue Type: Bug
Components: Table SQL / Planner
Reporter: Martijn Visser
Running OverAggregateRestoreTest#generateTestSetupFiles (with the @Disabled
condition deactivated) for the program
over-aggregate-bounded-non-partitioned-rows-with-out-of-order-records on master
never reaches the stop-with-savepoint: the sink observer waits for the output
to equal consumedBeforeRestore (18 rows), but the generator run emits 16 rows.
It never emits +I[Hello, 53, ..., 15, 1, 1, 19] and +I[Hello, 99, ..., 15, 3,
3, 31], and emits +I[Hello World, 100, ..., 7, 3, 3, 19] where the program
lists +I[Hello World, 100, ..., 7, 1, 1, 7]. After five minutes it fails with
"Sink did not produce the expected results". The committed version 1 files
restore fine, only regeneration is broken, so the program's before-restore
expectations no longer match what the generator setup (values source with
runtime-source=NewSource and per-record watermarks) produces for the late rows.
FLINK-40737 had to leave this program out of stream-exec-over-aggregate version
2 for that reason.
Side effect worth fixing in RestoreTestBase: when one program times out, its
job (terminating=false) keeps running, and every later program in the same JVM
sees an empty sink and times out too, which made it look like seven programs
were affected.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
