Low-Overhead Heap Profiling

[JC Beyler]

And it has been exactly two months since I posted an update here :)

Thanksgiving, Christmas, and handling
https://bugs.openjdk.java.net/browse/JDK-8190862 will do that to you
:)

I have gotten back to this item now that JDK-8190862 is done and I
have the following webrev ready:
http://cr.openjdk.java.net/~jcbeyler/8171119/webrev.02/

With the incremental here:
http://cr.openjdk.java.net/~jcbeyler/8171119/webrev.01_02/

The updates are:

a) GetCachedTraces

- I added a GC collection if the GetLiveTraces is called, this is
because you really want to know what is currently live when you call
that method
- If you are worried about performance and only care about what was
live at the last GC, you can now call GetCachedTraces, which does not
provoke a GC

Let me know if there are any questions here or concerns. I'm happy to
explain and defend the choices :).
Note: I added a new test for this and updated other tests to double
check the live call. (see for example
http://cr.openjdk.java.net/~jcbeyler/8171119/webrev.01_02/test/hotspot/jtreg/serviceability/jvmti/HeapMonitor/MyPackage/HeapMonitorCachedTest.java.html)

b) Disabling samples wipes out the data

I had originally implemented for OpenJdk a version that keeps the data
in memory after disabling the sampler, allowing a user to get traces
post-sampling. Because of this, we would always do the weak_oops_do
method, whether enabled or disabled. This led to a slight regression
in performance for GC reference processing time. I had initially fixed
this with a small "was this ever enabled" flag. This would have
allowed a program that never uses this to not have a regression but a
program that enables the disabled the code for the rest of the
execution would still pay the price after disabling the sampler.

Because of this, I have moved things back to where they probably
should be: if you disable the sampler, you lose the data. But this
allows a simpler code path: if the sampler is disabled, skip over the
GC weak_oops_do method.

Let me know what you think and happy 2018 to all!
Jc

On Thu, Nov 23, 2017 at 7:20 AM, Thomas Schatzl
<thomas.schatzl at oracle.com> wrote:
> On Tue, 2017-11-21 at 13:50 -0800, JC Beyler wrote:
>> Hi all,
>>
>> I have a new webrev here:
>> http://cr.openjdk.java.net/~rasbold/8171119/webrev.15/
>>
>> With the incremental one here:
>> http://cr.openjdk.java.net/~rasbold/8171119/webrev.14a_15/
>>
>> I think I did most items from Thomans and Robbin. I've especially
>> added a new test:
>
> Thanks. Looks good.
>
> Thanks,
>   Thomas
>