8017163: minor question regarding the calculation and logging of card set coarsening statistics

[Thomas Schatzl]

Hi,

   first, apologies for the late answer - I've been on vacation last 
week and then forgot a bit about this.

On 26.05.22 18:05, Tianqi Xia wrote:
> Hi,
> 
> One more observation on how the coarsening statistics are printed.
> 
> Right now coarsening stats are printed 3 times during a GC cycle: >
> The first one happens when G1HeapPrinterMark is constructed, this one 
> obtains its data from the _coarsenings field of G1RemSetSummary, and it 
> covers the coarsening stats from the end of last GC cycle to the 
> beginning of current GC cycle.
> 
> The second one happens when G1HeapPrinterMark is destructed, its source 
> is the _coarsen_stats and _last_coarsen_stats fields of G1CardSet, and 
> covers coarsening stats from the end of last GC cycle to the end of 
> current GC cycle, with both recent- and all- statistics.
> 
> The third printing happens immediately after the second one, its source 
> is again the _coarsenings field of G1RemSetSummary, and it covers the 
> coarsening stats from the beginning of current GC cycle to the end of 
> current GC cycle.
> 
> Based on testing results of BigRamTester and some other benchmarks, data 
> displayed by the first one and the second one are almost identical, 
> despite the fact that the second one is more verbose, and in most cases 
> the third one just displays a bunch of zeros.

The first one should always be a copy of the "Coarsening (recent)" 
statistics (at least with G1SummarizeRSetStatsPeriod=1).

The third one should always print zeros - gc does not (and should not) 
add any references directly to the remembered set.

> Is it possible to just keep the second one and remove others? This 
> change can keep the logging of coarsening stats more concise, and save 
> the effort of maintaining a copy of these data in G1RemSetSummary. >
> I understand there is nothing wrong with the current implementation, 
> just some thoughts to make the code a little bit cleaner.

I am fine with that although it looses a bit of information with 
G1SummarizeRSetStatsPeriod > 1.

I filed https://bugs.openjdk.java.net/browse/JDK-8287558.

Thanks,
   Thomas