number of cpus detection still buggy

[Aleksandar Ivanisevic]

As pointed out to me in this thread

https://discuss.linuxcontainers.org/t/online-cpu-numbers-in-sys-devices-system-cpu-online-and-sched-getaffinity-masks/13991

this is perfectly possible and reproducible also in physical machines and full VMs, not only in containers. It is enough to offline a (couple of) CPUs and the bug will be triggered.

Is there any interest by anyone in this being fixed?

> On 4. May 2022, at 17:51, Aleksandar Ivanisevic <aleksandar.ivanisevic at 2e-systems.com> wrote:
> 
> Hi,
> 
> looks like number of CPUs detection is still buggy in u332, if there is a core online with a logical number bigger than the total number of cores allocated to the (LXD, maybe also some docker implemantations) container in /sys/devices/system/cpu/online, it does not get counted, JRE counts only the ones with a “possible” number, i.e. if there are 8 cores allocated, only cores 0-7 are counted
> 
> # cat Cpus.java
> class Cpus {
>    public static void main(String[] args) {
>        System.out.println("Number of processors available to this JVM: " + Runtime.getRuntime().availableProcessors());
>    }
> }
> 
> # java -version
> openjdk version "1.8.0_332"
> OpenJDK Runtime Environment (Zulu 8.62.0.19-CA-linux64) (build 1.8.0_332-b09)
> OpenJDK 64-Bit Server VM (Zulu 8.62.0.19-CA-linux64) (build 25.332-b09, mixed mode)
> 
> # cat /sys/devices/system/cpu/online
> 3,8-10,13,22,25,28
> 
> # java Cpus
> Number of processors available to this JVM: 1
> 
> so only core number 3 is counted. This is especially funny when no cores are within range, them jre thinks it has 0 CPUs and all sorts of hilarity ensues ;)
> 
> if I get rid of /sys/devices/system/cpu/online, or override number of cores with -XX:ActiveProcessorCount=  all is well
> 
> # umount /sys/devices/system/cpu
> # java Cpus
> Number of processors available to this JVM: 8
> 
> but this is really hard to diagnose, as the errors are for some reason mostly 
> 
> random java.lang.NoClassDefFoundError exceptions
> JRE crashing on startup with a segmentation fault in "# V  [libjvm.so+0x95eb25]  ParNewGeneration::collect(bool, bool, unsigned long, bool)+0x95”. 
> maven dying with "Error assembling JAR: / by zero”
> 
> (just putting this down in hope people might find it googling ;)
> 
> java11 does not have that problem. How does one go about getting this fixed so that people do not waste more time with this than its necessary?  I’m just a lowly sysadmin that lost two days of his life on this, not really enlightened how things are going with JVM development.
> 
> Thank you.
> 
> -- 
> Aleksandar Ivanisevic
> Head of Operations and Support
> 2e Systems
> 
> Tel: +49 - 6196 - 950 58 14
> Fax: +49 - 6196 - 950 58 94
> E-mail: Aleksandar.Ivanisevic at 2e-systems.com
> 
> Address: 2e Systems GmbH,  Koenigsteiner Str. 107, 65812 Bad Soden am Taunus
> Company registration: Amtsgericht Koenigstein (Germany), HRB 7303
> Director: Philip Douglas
> 
> http://www.2e-systems.com - making your business fly!
> 
> 

-- 
Aleksandar Ivanisevic
Head of Operations and Support
2e Systems

Tel: +49 - 6196 - 950 58 14
Fax: +49 - 6196 - 950 58 94
E-mail: Aleksandar.Ivanisevic at 2e-systems.com

Address: 2e Systems GmbH,  Koenigsteiner Str. 107, 65812 Bad Soden am Taunus
Company registration: Amtsgericht Koenigstein (Germany), HRB 7303
Director: Philip Douglas

http://www.2e-systems.com - making your business fly!