RFR: Do not abandon RP discovery on conc cancel

Aleksey Shipilev shade at redhat.com
Wed Jun 7 18:09:30 UTC 2017


On 06/07/2017 07:51 PM, Aleksey Shipilev wrote:
> Hi,
> 
> I think our code that handles concurrent GC cancellation is still under belief
> that cancellation means Full GC. Which makes it correct to abandon reference
> discovery on cancel. But this is not true anymore for degenerate CM/UR -- we
> *want* to continue with RP, and we don't want to lose references. It smells like
> a bug.
> 
> The fix is to move abandonment to Full GC:
>   http://cr.openjdk.java.net/~shade/shenandoah/fullgc-abandon-rp/webrev.01/

Maybe even stronger, abandon unconditionally before Full GC:
  http://cr.openjdk.java.net/~shade/shenandoah/fullgc-abandon-rp/webrev.02/

-Aleksey



More information about the shenandoah-dev mailing list