-
Notifications
You must be signed in to change notification settings - Fork 28.9k
[SPARK-53673][CONNECT][TESTS] Fix a flaky test failure in SparkSessionE2ESuite - interrupt tag
caused by the usage of ForkJoinPool
#52417
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Conversation
Intentionally keep the test as ignore because there is another cause of the flakiness. |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
It sounds reasonable to me as a partial improvement. Thank you, @sarutak .
Feel free to merge since the CI result is irrelevant to this PR because the test case is still ignored. |
Thank yoou @dongjoon-hyun @zhengruifeng @yaooqinn for the review! BTW, as far as I know, there is the last one issue which blocks SPARK-48139 and a PR tries to resolve it. |
…SessionE2ESuite - interrupt tag` caused by the usage of `ForkJoinPool` ### What changes were proposed in this pull request? This PR backports #52417 to `branch-4.0`. This PR aims to fix one of the issues which block SPARK-48139. In the problematic test `interrupt tag` in `SparkSessionE2ESuite`, four futures run on threads in `ForkJoinPool` and try to interrupt through tags. A thread in a `ForkJoinPool` can create a spare thread and make it available in the pool so any of threads can be parent and child. It can happen when a thread performs a blocking operation. One example is `ArrayBlockingQueue.take` and it is called in a method provided by [gRPC](https://github.com/grpc/grpc-java/blob/24085103b926559659ecd3941a3308479876f084/stub/src/main/java/io/grpc/stub/ClientCalls.java#L607). On the other hand, tags are implemented as [InheritableThreadLocal](https://github.com/apache/spark/blob/13e70100426233e62fd9edf13e229f91f4941ff8/sql/connect/common/src/main/scala/org/apache/spark/sql/connect/client/SparkConnectClient.scala#L285). So, if the futures q1 and q4, or q2 and q3 are parent and child, tags should be inherited, which causes the flaky test failure. You can easily reproduce the issue by inserting a sleep into the problematic test like as follows (don't forget to replace `ignore` with `test`). ``` // TODO(SPARK-48139): Re-enable `SparkSessionE2ESuite.interrupt tag` - ignore("interrupt tag") { + test("interrupt tag") { val session = spark import session.implicits._ -204,6 +204,7 class SparkSessionE2ESuite extends ConnectFunSuite with RemoteSparkSession { spark.clearTags() // clear for the case of thread reuse by another Future } }(executionContext) + Thread.sleep(1000) val q4 = Future { assert(spark.getTags() == Set()) spark.addTag("one") ``` And then, run the test. ``` $ build/sbt 'connect-client-jvm/testOnly org.apache.spark.sql.connect.SparkSessionE2ESuite -- -z "interrupt tag"' ``` ### Why are the changes needed? For test stability. ### Does this PR introduce _any_ user-facing change? No. ### How was this patch tested? Ran the problematic test with inserting sleep like mentioned above and it passed. ### Was this patch authored or co-authored using generative AI tooling? No. Closes #52476 from sarutak/fix-thread-pool-issue-4.0. Authored-by: Kousuke Saruta <sarutak@amazon.co.jp> Signed-off-by: Dongjoon Hyun <dongjoon@apache.org>
…SessionE2ESuite - interrupt tag` caused by the usage of `ForkJoinPool` ### What changes were proposed in this pull request? This PR backports #52417 to `branch-3.5`. Different from `master` and `branch-4.0`, the SPARK-53673 doesn't seem to affect the `branch-3.5` at this time because the implementation of `ClientCalls#waitForNext` in `gRPC 1.56.0` which `branch-3.5` depends on is different from the one in `gRPC 1.67.1` which `master` and `branch-4.0` depend on. More specifically, the test doesn't go through the pass which calls `ArrayBlockingQueue#take` but go through [this else block](https://github.com/grpc/grpc-java/blob/v1.56.0/stub/src/main/java/io/grpc/stub/ClientCalls.java#L634). But I think it's better to backport #52417 to `branch-3.5` to prevent future changes from causing that issue. ### Why are the changes needed? Just in case to prevent future changes from causing that issue. ### Does this PR introduce _any_ user-facing change? No. ### How was this patch tested? 1. Temporarily enable the test and insert a sleep into the test like as follows ``` // TODO(SPARK-48139): Re-enable `SparkSessionE2ESuite.interrupt tag` - ignore("interrupt tag") { + test("interrupt tag") { val session = spark import session.implicits._ -204,6 +204,7 class SparkSessionE2ESuite extends ConnectFunSuite with RemoteSparkSession { spark.clearTags() // clear for the case of thread reuse by another Future } }(executionContext) + Thread.sleep(1000) val q4 = Future { assert(spark.getTags() == Set()) spark.addTag("one") ``` 2. Run the test and confirm it passes ``` $ build/sbt 'connect-client-jvm/testOnly org.apache.spark.sql.SparkSessionE2ESuite -- -z "interrupt tag"' ``` ### Was this patch authored or co-authored using generative AI tooling? No. Closes #52477 from sarutak/fix-thread-pool-issue-3.5. Authored-by: Kousuke Saruta <sarutak@amazon.co.jp> Signed-off-by: Dongjoon Hyun <dongjoon@apache.org>
What changes were proposed in this pull request?
This PR aims to fix one of the issues which block SPARK-48139.
In the problematic test
interrupt tag
inSparkSessionE2ESuite
, four futures run on threads inForkJoinPool
and try to interrupt through tags.A thread in a
ForkJoinPool
can create a spare thread and make it available in the pool so any of threads can be parent and child. It can happen when a thread performs a blocking operation. One example isArrayBlockingQueue.take
and it is called in a method provided by gRPC.On the other hand, tags are inplemented as InheritableThreadLocal.
So, if the futures q1 and q4, or q2 and q3 are parent and child, tags should be inheritd, which causes the flaky test faulure.
You can easily reprodue the issue by inserting a sleep into the problematic test like as follows (don't forget to replace
ignore
withtest
).And then, run the test.
Why are the changes needed?
For test stability.
Does this PR introduce any user-facing change?
No.
How was this patch tested?
Ran the problematic test with inserting sleep like mentioned above and it passed.
Was this patch authored or co-authored using generative AI tooling?
No.