[RFR]: Per thread IO statistics in JFR

[Alan Bateman]

On 01/02/2019 10:52, Volker Simonis wrote:
> :
> As I wrote in my previous mail - this information has proved very
> valuable for both, our support engineers as well as our developers.
> What do you mean by "start with a summary"? The feature is trivial -
> you get exact, per thread IO information in the same way you currently
> already get allocation information per thread. Why should the one be
> useful and the other not?
>
Allocation rates are useful for tuning and troubleshooting and are 
better understood.  I think I/O events require a lot more discussion. 
I've no doubt that your app server or support folks have found it useful 
but all the discussion and info on how they use it is missing here. We 
are also missing the bigger discussion on extending serviceability 
features to the libraries, and this at least partly related to a 
discussion on how serviceability features will work when more of the 
runtime and compilers are written in Java. There was some ground work 
done on this in JDK 12 to allow code in java.base generate JFR events 
that may be part of the discussion and exploration. The other thing that 
is missing is a write up of the different approaches that you have 
tried. From a maintenance point of view it is important to see the 
solution that does the instrumentation in the java code, rather the 
native methods as we want less native code. Finally there is a issue on 
where the approaches fit with the bigger projects, hence the 
concerns/questions on where I/O stats per thread might go in the future.

-Alan.