David Anderson created FLINK-40138:
--------------------------------------
Summary: Migration test snapshot generator does not drive the
JUnit 5 lifecycle, breaking snapshot generation for JUnit-5-migrated tests in
flink-tests
Key: FLINK-40138
URL: https://issues.apache.org/jira/browse/FLINK-40138
Project: Flink
Issue Type: Bug
Components: Test Infrastructure, Tests
Affects Versions: 2.3.0, 2.4.0
Reporter: David Anderson
h3. Problem
Running the snapshot generator for \{{flink-tests}} as documented in
\{{flink-migration-test-utils/README.md}}:
{code:bash}
mvn clean package -Pgenerate-migration-test-data -Dgenerate.version=2.3 -nsu
-Dfast -DskipTests
{code}
fails on the \{{flink-tests}} module. The Maven output only shows this error:
{code}
[ERROR] Failed to execute goal
org.apache.maven.plugins:maven-antrun-plugin:3.1.0:run
(generate-migration-test-data) on project flink-tests: An Ant BuildException
has occured:
Java returned: 1
{code}
The forked JVM actually fails with:
{code}
Caused by: java.lang.NullPointerException
at org.apache.flink.util.Preconditions.checkNotNull(Preconditions.java:59)
at
org.apache.flink.util.concurrent.ScheduledExecutorServiceAdapter.<init>(ScheduledExecutorServiceAdapter.java:37)
at
org.apache.flink.test.state.operator.restore.AbstractOperatorRestoreTestBase.<init>(AbstractOperatorRestoreTestBase.java:103)
...
{code}
h3. Root cause (as determined by Claude)
The JUnit 5 migration ([FLINK-39124]) converted the migration tests' JUnit 4
\{{@ClassRule}}/\{{@Rule}} \{{ExternalResource}}s into JUnit 5
\{{@RegisterExtension}} extensions and \{{@BeforeEach}} methods, but
\{{MigrationTestsSnapshotGenerator}}/\{{SnapshotGeneratorUtils}} were not
updated. The generator instantiates test classes by reflection and only drives
JUnit 4 rules (\{{RunRules}}), so the JUnit 5 lifecycle never runs. This
surfaces as two distinct failures:
Operator-restore tests (\{{AbstractOperatorRestoreTestBase}} and subclasses):
the instance field initializer \{{scheduledExecutor = new
ScheduledExecutorServiceAdapter(EXECUTOR_RESOURCE.getExecutor())}} throws NPE,
because the \{{@RegisterExtension TestExecutorExtension}}'s \{{beforeAll}}
(which creates the executor) is never called. The \{{MiniClusterExtension}} and
the \{{@BeforeEach setClusterClient(@InjectClusterClient …)}} method are
likewise never driven.
{\{StatefulJob*MigrationITCase}}: silently skipped. [FLINK-39124] also changed
\{{generateSnapshots}} from \{{public}} to \{{private}}, and the generator
discovers generator methods via \{{Class.getMethods()}} (public only), so the
\{{@ParameterizedSnapshotsGenerator}} method (and its \{{private static}}
parameters method, looked up via \{{Class.getMethod}}) are no longer found.
This is the same class of silent failure reported in [FLINK-34972].
The regression is present on \{{master}} and \{{release-2.3}}. It was not
caught earlier because the last reference-data generation ([FLINK-38775], 2.2)
predates the JUnit 5 migration.
h3. Related
- Blocks [FLINK-40020] (Flink 2.3: update reference data for migration tests)
- Regression from [FLINK-39124] (JUnit 5 migration of flink-tests)
- Overlaps with [FLINK-34972] (generator silently produces no files for some
subclasses)
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
