[SPARK-48139][CONNECT][TESTS] Try stabilising multi-thread tests in CI

[xupefei]

What changes were proposed in this pull request?

This PR tries to stabilise flaky tests which involve thread pools. Some tests are failing due to the thread pool having 2 threads instead of 3 or 4.

Why are the changes needed?

To let CI pass.

Does this PR introduce any user-facing change?

No.

How was this patch tested?

CI will tell.

Was this patch authored or co-authored using generative AI tooling?

No.

[dongjoon-hyun]

Thank you for working on this, @xupefei .

[dongjoon-hyun]

To @xupefei , could you add multiple empty commits on this PR in order to verify your claim?

CI will tell.

[xupefei]

An inrelavent test is failing:

TypeError: Channel.unary_stream() got an unexpected keyword argument '_registered_method'
[info] - python listener process: process terminates after listener is removed *** FAILED *** (2 seconds, 445 milliseconds)

[xupefei]

Failing on unrelated tests.

[dongjoon-hyun]

+1, LGTM.

Thank you, @xupefei and @HyukjinKwon .

Yes, the irrelevant failures seem to block the testing. Let's merge this PR.

[dongjoon-hyun]

For the record, unfortunately, it turns out this doesn't help.

• https://github.com/apache/spark/actions/runs/11614520002/job/32342932206
• https://github.com/apache/spark/actions/runs/11605545799/job/32316140558

[info] - interrupt tag *** FAILED *** (1 minute)
[info]   Expected exception org.apache.spark.SparkException to be thrown, but java.util.concurrent.TimeoutException was thrown (SparkSessionE2ESuite.scala:250)

[dongjoon-hyun]

Sorry but let me revert this because this failed twice a day already in master commit builder, @xupefei .

I guess there might be more failures on PR Builders after this PR.

[dongjoon-hyun]

This is reverted via the following.

• 125c1da

[xupefei]

Sounds good!
We should refactor these tests to use only two threads. That would definitely resolve this issue.

[xupefei]

@dongjoon-hyun Could you disable Cancellation APIs in SparkSession are isolated and link it to SPARK-48139?

[dongjoon-hyun]

To @xupefei , I believe we should handle them separately. You can apply most of your patch if you exclude these lines.

-  // TODO(SPARK-48139): Re-enable `SparkSessionE2ESuite.interrupt tag`
-  ignore("interrupt tag") {
+  test("interrupt tag") {

In other words, you should file and use a different JIRA ID for the rest of your contribution because SPARK-48139 is dedicated to this ignore("interrupt tag") exactly.

[dongjoon-hyun]

If you need the rest of patch, please open a new PR (by excluding ignore part) with new ID. Then, we can merge it back without brining back this flakiness.

[xupefei]

It's weird that only the interrupt tag test is failing, as I believe they are suffering from the same issue (thread pool only got 2 threads instead of requested 4). If my change didn't fix interrupt tag then I think Cancellation APIs in SparkSession are isolated is still flaky. I'd disable it for now.

[xupefei]

PR created at #48736.