fix: use unique partition keys in QueryReturnTypesIT to avoid LWT contention

[nikagra]

Note

Corresponds to scylladb/scylla-java-driver-matrix#142.

Problem

QueryReturnTypesIT is annotated @Category(ParallelizableTests.class) and all test methods were using the same hardcoded partition key (id=1). On Scylla with tablets, initial LWT queries to the same partition key can be routed to random nodes, causing Paxos contention across parallel test threads and resulting in WriteTimeoutException.

Fix

Assign each test method instance a unique partition key via a static AtomicInteger counter incremented in @Before, so no two concurrently running tests contend on the same partition. The "not found" probe previously using id=2 now uses -(testId + 1), which is always negative and therefore guaranteed to never be assigned by the counter.

[Lorak-mmk]

Two questions:

• Why does this error even happen? In other word, why Scylla fails to achieve qorum in single-node scenario?
• Can't this be solved with a retry policy instead of separate function? Also, should the default retry policy maybe handle this case?

[dkropachev]

Two questions:

• Why does this error even happen? In other word, why Scylla fails to achieve qorum in single-node scenario?
• Can't this be solved with a retry policy instead of separate function? Also, should the default retry policy maybe handle this case?

This is acutally funny, it happened because query are run on the same PK and run in parallel, so it is regular LWT congestion .

[Lorak-mmk]

Contention and timeout from just 2 queries? Wow. I did not expect LWT to be THAT slow

[nikagra]

@Lorak-mmk, I've reworked fix following @dkropachev's feedback

[Copilot]

Pull request overview

This PR updates the QueryReturnTypesIT integration test to avoid LWT (Paxos) contention when tests are executed in parallel, by ensuring each test instance operates on a distinct partition key.

Changes:

• Introduced a per-test unique partition key generated from a static AtomicInteger.
• Updated all DAO calls/assertions to use the per-test testId instead of hardcoded IDs.
• Updated “not found” probes to use a guaranteed-unassigned negative ID derived from testId.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[dkropachev]

Contention and timeout from just 2 queries? Wow. I did not expect LWT to be THAT slow

Why only two ? integration-tests/pom.xml:41 sets test.parallel.threads=16, so the whole suit running at once, targeting same PK.

[dkropachev]

Contention and timeout from just 2 queries? Wow. I did not expect LWT to be THAT slow

Why only two ? integration-tests/pom.xml:41 sets test.parallel.threads=16, so the whole suit running at once, targeting same PK.

Actually, it is not what happening, tests in the test suit running one by one, but different test suits are running in parallel.
Anyways splitting them into different PKs will remove LWT extra sync anyway, so this PR still should work.