[SPARK-36900][BUILD][TESTS][FOLLOWUP] Try 5g, not 6g, for test mem limit

[srowen]

What changes were proposed in this pull request?

Use 5g of memory for tests, not 6g.

Why are the changes needed?

6g might be flaky, per #34214

Does this PR introduce any user-facing change?

No

How was this patch tested?

Existing tests on Java 11

[HyukjinKwon]

im gonna just push this in. the tests already passed with both 4g and 6g. To actually test this change, we should monitor other builds.

[HyukjinKwon]

Merged to master.

[dongjoon-hyun]

Thank you for keeping trying this optimization, @srowen and @HyukjinKwon .

[HyukjinKwon]

Ah .. flakiness still persists. I could hack with calling GC for TPC-DS but seems like other tests are flaky too (e.g., https://github.com/apache/spark/runs/3871942012).

Probably we should find other ways around to 1. reduce memory usage with JDK 17 in tests or 2. add another environment variable (or respecting JAVA_OPTS) in these places we have ..

[HyukjinKwon]

Since we don't support JDK 17 yet (and don't run any tests in CI), let me revert this one first since this blocks other PRs for now 😢 .

[srowen]

That's fine, but I'm guessing it's not actually related to this change, but, who knows

[HyukjinKwon]

I will follow up and bring this in back if this wasn't the reason. Thanks @srowen.