RFR(XS): 8138607: gc/g1/mixedgc/TestLogging.java - G1 Evacuation Pause missing from output

Thomas Schatzl thomas.schatzl at oracle.com
Mon Apr 18 15:10:04 UTC 2016


Hi Kirill,

On Fri, 2016-04-15 at 16:58 +0300, Kirill Zhaldybin wrote:
> Jon,
> 
> Thank you for reviewing the fix!
> 
> The test is designed to  create a situation when Mixed GC should 
> happen for sure. The fact of Mixed GC is checked by gc log analysis. 
> The test failed previously because under some circumstances an 
> unexpected Full GC occurred.
> 
> The test starts CMC by itself (WB.g1StartConcMarkCycle();)
> So, to be more deterministic IHOP is set to 100 to avoid CMC 
> initiated by GC, not by the test.

  I had the same question :)

I kind of recommend indicating that in the comment (like "set initial
CMC threshold to avoid additional concurrent cycles caused by
ergonomics").

Maybe also disable adaptive IHOP (-XX:-G1UseAdaptiveIHOP), although I
think since the test starts only a single CMC, it should be good.

I do not need to re-review any potential change of the comment.

Thanks,
  Thomas


More information about the hotspot-gc-dev mailing list