[
https://issues.apache.org/jira/browse/FLINK-26176?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Sebastian Mattheis updated FLINK-26176:
---------------------------------------
Description:
The following Scala integration test cases in flink-tests:
*
flink-tests/src/test/scala/org/apache/flink/api/scala/migration/StatefulJobSavepointMigrationITCase.scala
*
flink-tests/src/test/scala/org/apache/flink/api/scala/migration/StatefulJobWBroadcastStateMigrationITCase.scala
are supposed and configured to test migration of savepoints of the RocksDB,
Memory, and HashMap state backend between minor version upgrades (1.x -> 1.y).
However, the test methods (in both test cases), which are
* testCreateSavepoint and testRestoreSavepoint
* testCreateSavepointWithBroadcastState and testRestoreSavepointWithBroadcast
do set the state backend to use only MemoryStateBackend and overwrite any
previous (correct according to test configuration) setting of the state backend.
The consequence is:
* tests run only with MemoryStateBackend
* previous test resources, i.e., savepoints for RocksDB/HashMap states of
versions (1.3-1.14), are corrputed
Proposed solution;
* fix the setting of state backend
* delete corrupted savepoints
* recreate savepoints for older versions OR omit test runs for older versions
was:
The following Scala integration test cases in flink-tests:
*
flink-tests/src/test/scala/org/apache/flink/api/scala/migration/StatefulJobSavepointMigrationITCase.scala
*
flink-tests/src/test/scala/org/apache/flink/api/scala/migration/StatefulJobWBroadcastStateMigrationITCase.scala
are supposed and configured to test migration of savepoints of the RocksDB,
Memory, and HashMap state backend between minor version upgrades (1.x -> 1.y).
However, the test methods (in both test cases), which are
* testCreateSavepoint and testRestoreSavepoint
* testCreateSavepointWithBroadcastState and testRestoreSavepointWithBroadcast
do set the state backend to use only MemoryStateBackend and overwrite any
previous (correct according to test configuration) setting of the state backend.
The consequence is:
* tests run only with MemoryStateBackend
* previous test resources, i.e., savepoints for RocksDB/HashMap states of
versions (1.3-1.14), are corrputed
Proposed solution;
* fix the configuration
* delete corrupted savepoints
* recreate savepoints for older versions OR omit test runs for older versions
> Scala savepoint migration tests don't run as configured (not testing RocksDB
> savepoints)
> ----------------------------------------------------------------------------------------
>
> Key: FLINK-26176
> URL: https://issues.apache.org/jira/browse/FLINK-26176
> Project: Flink
> Issue Type: Bug
> Components: Tests
> Affects Versions: 1.9.3, 1.10.3, 1.11.6, 1.12.7, 1.13.5, 1.14.3
> Reporter: Sebastian Mattheis
> Priority: Major
>
> The following Scala integration test cases in flink-tests:
> *
> flink-tests/src/test/scala/org/apache/flink/api/scala/migration/StatefulJobSavepointMigrationITCase.scala
> *
> flink-tests/src/test/scala/org/apache/flink/api/scala/migration/StatefulJobWBroadcastStateMigrationITCase.scala
> are supposed and configured to test migration of savepoints of the RocksDB,
> Memory, and HashMap state backend between minor version upgrades (1.x ->
> 1.y). However, the test methods (in both test cases), which are
> * testCreateSavepoint and testRestoreSavepoint
> * testCreateSavepointWithBroadcastState and testRestoreSavepointWithBroadcast
> do set the state backend to use only MemoryStateBackend and overwrite any
> previous (correct according to test configuration) setting of the state
> backend.
> The consequence is:
> * tests run only with MemoryStateBackend
> * previous test resources, i.e., savepoints for RocksDB/HashMap states of
> versions (1.3-1.14), are corrputed
> Proposed solution;
> * fix the setting of state backend
> * delete corrupted savepoints
> * recreate savepoints for older versions OR omit test runs for older versions
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
