RFR: 8257774: G1: Trigger collect when free region count drops below threshold to prevent evacuation failures [v6]
Kirk Pepperdine
kirk at kodewerk.com
Sun May 23 02:42:33 UTC 2021
His Stefan,
What we have here are essentially speculative collections. Historically, speculative collections have been problematic and and it would have been a real issue to not be able to turn them off. Also, int has been traditional to have “arbitrary thresholds” backed by the ability to configure/adjust them. DGC triggered collections is a prime example of this. Another example is OOME which is controlled by two different flag settings, throughput and heap recovered. I have had occasions, albeit rare ones, where the immediate need to was reset the heap recovered from 2% to 5%. There are other cases where “assumed” thresholds simply didn’t work well with the workload that the JVM was presented with. In each of these cases, and many, having the ability to adjust thresholds and configure behavior was a life line. I would like to think that we would continue to allow those of us in the field the ability to make these adjustments so that we might technically tune collectors to cope with workloads that you’re unlikely to see in a benchmark.
Kind regards,
Kirk
> On May 21, 2021, at 12:52 AM, Stefan Johansson <sjohanss at openjdk.java.net> wrote:
>
> On Fri, 21 May 2021 01:55:02 GMT, Aditya Mandaleeka <adityam at openjdk.org> wrote:
>
>>> What are your thoughts on further steps involved before getting this change merged? For instance, is there value/interest in putting the proactive GCs behind an -XX: option? Every such flag adds complexity and surface area for testing, so I'm somewhat hesitant but I'm interested to hear others' opinions on this...
>
> I agree, we should avoid adding a new flag unless we see a real potential problem with having this always on. At this point I don't think we do. I hope to find time later today to do the final review, but have some other things to finish first so this might get pushed to beginning of next week.
>
> -------------
>
> PR: https://git.openjdk.java.net/jdk/pull/3143
More information about the hotspot-gc-dev
mailing list