RFR: 8265436: G1: Improve gc+phases log output during full gc

Thomas Schatzl tschatzl at openjdk.java.net
Tue Apr 20 09:39:07 UTC 2021


On Tue, 20 Apr 2021 02:14:39 GMT, Hamlin Li <mli at openjdk.org> wrote:

>> Hi all,
>> 
>>   can I have reviews for this fix to remove recently introduced log messages from `debug` to `trace` level?
>> 
>> If you select `gc+phases=debug` logging with [JDK-8262068](https://bugs.openjdk.java.net/browse/JDK-8262068), you get:
>> 
>>     [0.695s][debug][gc,phases] GC(51) Phase 1: Weak Processing 0.309ms
>>     [0.695s][debug][gc,phases] GC(51) ClassLoaderData 0.003ms
>>     [0.695s][debug][gc,phases] GC(51) Trigger cleanups 0.002ms
>>     [0.699s][debug][gc,phases] GC(51) Phase 1: Class Unloading and Cleanup 4.056ms
>>     [0.699s][info ][gc,phases] GC(51) Phase 1: Mark live objects 7.722ms
>>     [0.699s][debug][gc,phases] GC(51) Phase 2: skip compaction region index: 0, live words: 125610
>>     [repeated for every region skipped] 
>> 
>> the last per-region message is quite annyoing and just fills up logs as the code prints one line per affected region (which may be a lot), so I suggest to move it to `trace` level. I did suggest to add some nice phase 2 region summary in [JDK-8265437](https://bugs.openjdk.java.net/browse/JDK-8265437), but that is out of scope for this change.
>> 
>> Testing: local testing
>> 
>> Thanks,
>>   Thomas
>
> just realized, the test code needs to be adjusted accordingly.
> '''
> public class TestG1SkipCompaction {
>     public static void runTest() throws Exception {
>         final String[] arguments = {
>             "-XX:+UseG1GC",
>             "-XX:MarkSweepDeadRatio=3",
>             "-Xmx8m",
>             "-Xms8M",
>             "-Xlog:gc+phases=debug",            // here, from debug to trace
>             "-XX:G1HeapRegionSize=1m",
>             GCTest.class.getName()
>             };
> '''

I will integrate after tier1-2 testing that runs that test passed. Thanks @Hamlin-Li @albertnetymk for your reviews.

-------------

PR: https://git.openjdk.java.net/jdk/pull/3569



More information about the hotspot-gc-dev mailing list