RFR: 8288474: Move EventContinuationFreezeOld from try_freeze_fast to freeze_slow

Coleen Phillimore coleenp at openjdk.org
Wed Sep 14 12:21:28 UTC 2022


On Wed, 14 Sep 2022 11:45:42 GMT, Ron Pressler <rpressler at openjdk.org> wrote:

>> This moves the event ContinuationFreezeOld to freeze_slow so it's reported on the different paths to freeze_slow, and renames the events Continuation*Old/Continuation*Slow and Continuation*Young/Continuation*Fast to reflect that they're on the fast/slow paths.  I turned on CONT_JFR and fixed some bugs and turned it back off again.  Some output:
>> 
>> $ jfr summary hotspot-pid-31671-id-1-2022_09_13_19_42_01.jfr | grep Continuation
>>  jdk.ContinuationThaw                    12193        211929
>>  jdk.ContinuationThawFast                10139        165420
>>  jdk.ContinuationFreeze                   3196         57135
>>  jdk.ContinuationFreezeFast               2788         46397
>>  jdk.ContinuationThawSlow                 2054         28543
>>  jdk.ContinuationFreezeSlow                408          5688
>> 
>> Tested locally and with tier7 which seems to run JFR tests.
>
> src/hotspot/share/runtime/continuationFreezeThaw.cpp line 723:
> 
>> 721: #if CONT_JFR
>> 722:   EventContinuationFreezeSlow e;
>> 723:   if (e.should_commit()) {
> 
> Committing the event should happen at the end of the method to measure the time.

That makes sense but the time recorded for these is startTime:
jdk.ContinuationThawSlow {
  startTime = 11:54:23.301 (2022-09-14)
  id = 4260565248
  eventThread = "MainThread" (javaThreadId = 36)
}

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

PR: https://git.openjdk.org/jdk/pull/10254


More information about the hotspot-dev mailing list