[SPARK-2208] Fix for local metrics tests can fail on fast machines

[joan38]

What changes were proposed in this pull request?

Second trial to fix local metrics tests that can fail on fast machines.
This is probably what is suggested here #3380 by @aarondav?

How was this patch tested?

This need to run multiple times CI to be sure that the tests doesn't accidentally pass like in the previous trial #11747.

Cheers

[JoshRosen]

Is this worth fixing? It's not flaky in Jenkins and the last few attempts at fixing it have made flakiness worse in our CI environment.

[SparkQA]

Test build #54061 has finished for PR 11937 at commit 2f03ab3.

• This patch fails Spark unit tests.
• This patch merges cleanly.
• This patch adds no public classes.

[srowen]

Yeah it's not flaky now because it's disabled. Although this was a good guess at the fix and passed locally and in the PR builder, it looks like there's a substantial difference in how it runs in the Jenkins jobs. It has been disabled a long time and for better or worse may be just tough to fix, at least by going down this particular road. I think we might want to record this attempt for posterity but just give up on it.

[joan38]

I've done some benchmark and the Thread.sleep I've added on SortShuffleManager.getBlockData actually doesn't affect the fetchWaitTime. So this change is actually useless and I don't really know how to fix this.