Question about trace-IDs

[Markus Gronlund]

Hi again,

-----Original Message-----
From: Roman Kennke <rkennke at redhat.com> 
Sent: den 21 september 2020 13:48
To: Markus Gronlund <markus.gronlund at oracle.com>; hotspot-jfr-dev at openjdk.java.net
Subject: Re: Question about trace-IDs

Hi Markus,



> Thanks for noticing some staleness here - remnants from having used 
> different atomic increment logic(s) in the past.
> 
> You can initialize all counters to 0, and also let LAST_TYPE_ID be 
> used as-is without the + 1.
> 

Thanks for confirming this!

While I'm at it, I could also replace the helper method atomic_inc() and its uses with Atomic::add() which has the advantage that it uses CPU instructions for atomic-add instead of a CAS-loop when the CPU has such an instruction. WDYT?

I have found using the templated versions of Atomic::inc() to be problematic on 32-bit platforms  (will compile but not link) as the traceid type is always 64-bit. Maybe Atomic::add() handles this better, but I don’t think it's worth modifying, and Atomic::cmpxchg() works on all platforms. 


Roman


> Thanks
> Markus
> 
> -----Original Message-----
> From: Roman Kennke <rkennke at redhat.com>
> Sent: den 21 september 2020 12:39
> To: hotspot-jfr-dev at openjdk.java.net
> Subject: Question about trace-IDs
> 
> Hello all,
> 
> While studying the code around JFR trace-IDs, I came about something 
> that I find a little odd: trace-IDs for modules, packages and class- 
> loaders start at 2, while trace-IDs for threads start at 1 (and 
> classes at LAST_TYPE_ID+1). The static counters (for modules, packages 
> and
> classloaders) are initialized at 1, and the first call to
> atomic_inc() increments this to 2 and this is the first value used. I 
> understand that 0 is the uninitialized value, but don't see that 1 has 
> any special meaning.
> 
> See for example here:
> https://urldefense.com/v3/__https://github.com/openjdk/jdk/blob/master
> /src/hotspot/share/jfr/recorder/checkpoint/types/traceid/jfrTraceId.cp
> p*L65__;Iw!!GqivPVa7Brio!JGBhdXPwm7skiWvt6c0M2CXIXa_4CdjKsp6UKdp-b-19g
> _WFUOsJlWJl32N6avBT21Ai$
> 
> vs
> 
> https://urldefense.com/v3/__https://github.com/openjdk/jdk/blob/master
> /src/hotspot/share/jfr/recorder/checkpoint/types/traceid/jfrTraceId.cp
> p*L60__;Iw!!GqivPVa7Brio!JGBhdXPwm7skiWvt6c0M2CXIXa_4CdjKsp6UKdp-b-19g
> _WFUOsJlWJl32N6ajNzFPfr$
> 
> Would it hurt to start counting trace-IDs for packages, modules and 
> classloaders at 1 too? (If not, I'd like to submit a patch - I'm happy 
> about something simple as my first contributions to JFR ;-) ).
> 
> Cheers,
> Roman
>