Nashorn heap profiling

[Marcus Lagergren]

Oh, by the way - we never seem to get any free cycles for this, but in the Java 9 timeframe I hope (forward looking statement - so it can be a complete lie - disclaimer), that we will also have some basic JFR events in Nashorn itself, which will provide JavaScript specific data, such as “this callsite is megamorphic” and similar things.

/M

> On 04 May 2015, at 11:31, Marcus Lagergren <marcus.lagergren at oracle.com> wrote:
> 
> Personally, I use Java Mission control to record and analyze the heap operating set. This is free for non production use. Mission control also gives you source lines in your original JavaScript program for views.
> 
> Attila and I have previously also used the http://eclipse.org/mat/ <http://eclipse.org/mat/> Eclipse plugin.
> 
> Regards
> Marcus
> 
>> On 30 Apr 2015, at 15:32, Josh Fleming <jfleming at linkedin.com <mailto:jfleming at linkedin.com>> wrote:
>> 
>> Hi folks,
>> 
>> What's the best way to do heap profiling of JavaScript code in Nashorn? We're running a large JavaScript framework on Nashorn, and recently I had to track down a couple memory leaks. This turned out to be a painful process. When using a heap profiler like YourKit, it's very tedious to find the correspondence between what you see in a JVM retention path and the exact JavaScript code it originates from: there are multiple layers of JVM representation, and the naming is a bit cryptic.
>> 
>> Eventually I fell back on a manual process of drilling down to the "JO* -> map -> properties -> list" linked list to find a unique property name within each JavaScript object, and used what I found there to grep the JavaScript code. The situation is better if you happen to come upon a closure, because the "data -> functionName" property in ScriptFunctionImpl includes a line number. But it seems like there must be a better way.
>> 
>> Any guidance would be much appreciated.
>> 
>> 
>> Thanks,
>> 
>> jf
>> 
>