Hotspot: GC "logs" from hell

[Jesper Wilhelmsson]

Bernd Eckenfels skrev 23/5/13 11:48 PM:
> Am 23.05.2013, 20:56 Uhr, schrieb Jon Masamitsu <jon.masamitsu at oracle.com>:
>> This has been on our to-do list for
>> quite a while, but  it seems like every time we try to push it up,
>> something new gets dropped on top and pushes it back
>> down.
>
> Sure, what I think might be a good strategy is to leverage the event work. If I
> understand it correctly there is already a event ringbuffer for the
> flightrecorder stuff. And if all relevant GC events are pushed (with most
> relevant metrics) to that ring already (for the commercial products) the dumping
> of that buffer to a logfile is a more isolated extension to the JVM (as opposed
> to collecting all gc events which is all over the Hotspot gc subsystem).

There is a flag -XX:+EnableTracing to print all the tracing events.

> This should also allow to reduce the diagnostic code (printf formats) all over
> the place.

Hopefully the tracing events can replace some of the printf logging in there. It 
would be nice to get rid of some code for a change :-)
/Jesper

>
>> If in the end all we can do is provide guidance, is there
>> any sentiment in the community to try and get this done
>> separate from Oracle?
>
> I would certainly contribute, however my C is very rusty (I do not even have a
> working build environment locally).
>
> Gruss
> Bernd