RFR: JDK-8152952: Allow G1 phase logging to use individual number of threads

Bengt Rutisson bengt.rutisson at oracle.com
Tue Mar 29 12:03:18 UTC 2016


Hi everyone,

Could I have a couple of reviews for this change?

http://cr.openjdk.java.net/~brutisso/8152952/webrev.00/
https://bugs.openjdk.java.net/browse/JDK-8152952

Currently if you run with UseDynamicNumberOfGCThreads you can 
potentially get a different number of worker threads each GC. There are 
improvements coming where we want to select a different number of worker 
threads for individual phases. The G1GCPhaseTimes and WorkerDataArray 
structures need to support this.

The proposed patch sets all slots in the WorkerDataArray to an 
uninitialized value and then only print any values that have actually 
been set for that phase.

The patch also extends the log message about the number for worker 
threads to also say how many it could potentially have used. And it also 
fixes a missing space in the level 3 and level 4 indentation.

After applying this patch and running with -Xlog:gc*,phases*=trace you 
get output like:

[0,581s][info][gc,task ] GC(0) GC Workers: using 2 out of 23
[0,588s][info][gc,phases] GC(0)   Evacuate Collection Set: 5,0ms
[0,588s][trace][gc,phases] GC(0)       GC Worker Start (ms):     Min: 
580,9, Avg: 580,9, Max: 580,9, Diff:  0,0
[0,588s][trace][gc,phases,task] GC(0)                                 
580,9 580,9   -   -   - -   -   -   -   -   -   -   -   -   -   -   -   
-   -   -   - -   -
[0,588s][debug][gc,phases     ] GC(0)     Ext Root Scanning (ms):   
Min:  1,7, Avg:  1,7, Max:  1,8, Diff:  0,0, Sum:  3,5
[0,589s][trace][gc,phases,task] GC(0)                                
1,8  1,7   -   -   -   - -   -   -   -   -   -   -   -   -   -   -   -   
-   -   -   -   -
[0,589s][trace][gc,phases     ] GC(0)       Thread Roots (ms):        
Min:  0,0, Avg:  0,7, Max:  1,5, Diff:  1,5, Sum: 1,5
[0,589s][trace][gc,phases,task] GC(0)                                  
0,0  1,5   -   -   -   - -   -   -   -   -   -   -   -   -   -   -   -   
-   -   -   -   -

Thanks,
Bengt




More information about the hotspot-gc-dev mailing list