RFR: EvilSyncBug test is too slow

[Aleksey Shipilev]

On 09/08/2018 03:16 PM, Roman Kennke wrote:
> Not sure. Back then I needed to crank up the number of runs to make the test fail at least 
> somewhat reliably. The risk is that something gets changed in thread termination and this thread 
> coming back.

> This one haunted us for some weeks, and it was an actual customer/early adopters bug. The test 
> prevents that from coming back.

IIRC, it was a jcstress bug found by us.

I think we have enough safety net now: we have gc-state that is verified/asserted specially, we have
oom-evac thing that has its own test configurations, we even have VerifyJCStress test that is
derived from jcstress test.

This is why I propose to trim the test down, not to remove it. The bugs, if any not captured by the
safety net above, would show up in continuous testing, without prolonging the usual tier3 too much.

-Aleksey