You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@flink.apache.org by "Savonitar (via GitHub)" <gi...@apache.org> on 2026/10/05 18:22:55 UTC

Re: [PR] [FLINK-40139][Connectors/Kafka] test: Add default JUnit test timeout to fail hung tests fast [flink-connector-kafka]

Savonitar commented on code in PR #288:
URL: https://github.com/apache/flink-connector-kafka/pull/288#discussion_r4159374517


##########
flink-connector-kafka/pom.xml:
##########
@@ -418,6 +418,24 @@ under the License.
 					</execution>
 				</executions>
 			</plugin>
+
+			<plugin>
+				<groupId>org.apache.maven.plugins</groupId>
+				<artifactId>maven-surefire-plugin</artifactId>
+				<configuration>
+					<properties>
+						<!-- Fail a hung test method after 15 minutes, with its stack trace, instead of
+							letting it consume the CI job's timeout. A test can override this with @Timeout.
+							The test body runs in a separate thread so that a method which swallows the
+							interrupt still fails on time and the rest of the suite runs. -->
+						<configurationParameters>
+							junit.jupiter.execution.timeout.testable.method.default = 15 m

Review Comment:
   This setting covers testable methods only, leaving setup and teardown outside the new timeout. Right? 
   For example, KafkaSinkITCase.tearDown() waits on topic deletion through result.all().get(). 
   Could the description explicitly document this scope and the remaining reliance on CI-level timeout handling for lifecycle hangs?



##########
flink-connector-kafka/pom.xml:
##########
@@ -418,6 +418,24 @@ under the License.
 					</execution>
 				</executions>
 			</plugin>
+
+			<plugin>
+				<groupId>org.apache.maven.plugins</groupId>
+				<artifactId>maven-surefire-plugin</artifactId>
+				<configuration>
+					<properties>
+						<!-- Fail a hung test method after 15 minutes, with its stack trace, instead of
+							letting it consume the CI job's timeout. A test can override this with @Timeout.
+							The test body runs in a separate thread so that a method which swallows the
+							interrupt still fails on time and the rest of the suite runs. -->
+						<configurationParameters>

Review Comment:
   The ticket/commit/comment say “without a stack trace”.  Not arguing that stack trace is useful, however, I opened failed CI build from July https://github.com/apache/flink-connector-kafka/actions/runs/29204695279/job/86681985234 
   and it already has thread dump including stack
   
   ```
   "main" #1 prio=5 os_prio=0 cpu=9539.69ms elapsed=2413.56s tid=0x00007f407c029db0 nid=0xa504 waiting on condition  [0x00007f4083511000]
      java.lang.Thread.State: TIMED_WAITING (sleeping)
   	at java.lang.Thread.sleep(java.base@17.0.19/Native Method)
   	at org.apache.flink.streaming.api.operators.collect.CollectResultFetcher.sleepBeforeRetry(CollectResultFetcher.java:249)
   	at org.apache.flink.streaming.api.operators.collect.CollectResultFetcher.next(CollectResultFetcher.java:115)
   	at org.apache.flink.streaming.api.operators.collect.CollectResultIterator.nextResultFromFetcher(CollectResultIterator.java:124)
   	at org.apache.flink.streaming.api.operators.collect.CollectResultIterator.hasNext(CollectResultIterator.java:98)
   	at org.apache.flink.connector.testframe.utils.CollectIteratorAssert.compareWithExactlyOnceSemantic(CollectIteratorAssert.java:130)
   	at org.apache.flink.connector.testframe.utils.CollectIteratorAssert.matchesRecordsFromSource(CollectIteratorAssert.java:85)
   	at org.apache.flink.connector.testframe.testsuites.SourceTestSuiteBase.checkResultWithSemantic(SourceTestSuiteBase.java:749)
   	at org.apache.flink.connector.testframe.testsuites.SourceTestSuiteBase.testIdleReader(SourceTestSuiteBase.java:526)
   	at jdk.internal.reflect.NativeMethodAccessorImpl.invoke0(java.base@17.0.19/Native Method)
   ```
   
   So we already get the dump that the style guide relies on.
   If I'm not missing something, the actual benefit here is earlier method-attributed failure reporting and allowing remaining tests to proceed where possible.



##########
flink-connector-kafka/pom.xml:
##########
@@ -418,6 +418,24 @@ under the License.
 					</execution>
 				</executions>
 			</plugin>
+
+			<plugin>
+				<groupId>org.apache.maven.plugins</groupId>
+				<artifactId>maven-surefire-plugin</artifactId>
+				<configuration>
+					<properties>
+						<!-- Fail a hung test method after 15 minutes, with its stack trace, instead of
+							letting it consume the CI job's timeout. A test can override this with @Timeout.
+							The test body runs in a separate thread so that a method which swallows the
+							interrupt still fails on time and the rest of the suite runs. -->
+						<configurationParameters>
+							junit.jupiter.execution.timeout.testable.method.default = 15 m
+							junit.jupiter.execution.timeout.thread.mode.default = SEPARATE_THREAD
+							junit.jupiter.execution.timeout.mode = disabled_on_debug
+						</configurationParameters>

Review Comment:
   Should we also add
   ```
   junit.jupiter.execution.timeout.threaddump.enabled = true
   ```
   This captures the test JVM's thread stacks before interruption. 
   The existing CI dump step still runs on build failure, but the affected test JVM may already have exited or cleanup changed its state.
   Capturing a dump at the timeout would complement those CI diagnostics.
   [JUnit 5.13.3 documentation](https://docs.junit.org/5.13.3/user-guide/#writing-tests-declarative-timeouts-debugging-thread-dump)
   or am i missing something? 



##########
flink-connector-kafka/src/test/java/org/apache/flink/streaming/connectors/kafka/table/KafkaTableTestBase.java:
##########
@@ -53,10 +54,18 @@
 import java.util.Properties;
 import java.util.Timer;
 import java.util.TimerTask;
+import java.util.concurrent.TimeUnit;
 import java.util.stream.Collectors;
 
-/** Base class for Kafka Table IT Cases. */
+/**
+ * Base class for Kafka Table IT Cases.
+ *
+ * <p>Tests run on the thread that ran {@code @BeforeEach}, unlike the module's default of a
+ * separate thread: the table planner leaves Calcite's metadata handler provider in a thread-local
+ * there.
+ */
 @Testcontainers
+@Timeout(value = 15, unit = TimeUnit.MINUTES, threadMode = Timeout.ThreadMode.SAME_THREAD)

Review Comment:
   Is excluding the E2E modules intentional?



##########
flink-connector-kafka/src/test/java/org/apache/flink/streaming/connectors/kafka/table/KafkaTableTestBase.java:
##########
@@ -53,10 +54,18 @@
 import java.util.Properties;
 import java.util.Timer;
 import java.util.TimerTask;
+import java.util.concurrent.TimeUnit;
 import java.util.stream.Collectors;
 
-/** Base class for Kafka Table IT Cases. */
+/**
+ * Base class for Kafka Table IT Cases.
+ *
+ * <p>Tests run on the thread that ran {@code @BeforeEach}, unlike the module's default of a
+ * separate thread: the table planner leaves Calcite's metadata handler provider in a thread-local
+ * there.
+ */
 @Testcontainers
+@Timeout(value = 15, unit = TimeUnit.MINUTES, threadMode = Timeout.ThreadMode.SAME_THREAD)

Review Comment:
   Nit: could the Javadoc mention that this explicit 15-minute value should stay aligned with the Maven default? It will save up the next person from changing one and not the other.



-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: issues-unsubscribe@flink.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org