Forum Discussion

bijan_a_183509's avatar
bijan_a_183509
Icon for Nimbostratus rankNimbostratus
Aug 04, 2018

increasing cpu process ubnormally

Hi,

 

i recently run into CPU spike process during mentioned time(9am-11am) at morning on both chassis(Active/Standby). when i digged it out, found out on CPUs which handle load of Control plan happened on odd cores.

 

The setting which has been applied on both includes LTM, ASM, AVR and HA configuration. Could you please let me know any idea about this?

 

application delivery
ASM Advanced WAF
LTM
  • bijan_a_183509's avatar
    bijan_a_183509
    Icon for Nimbostratus rankNimbostratus
    Aug 06, 2018

    To add some more information, i'd say that the most Memory usage is related to Java process. Is there any idea about that?

     

  • MvdG's avatar
    MvdG
    Icon for Cirrus rankCirrus
    Aug 06, 2018

    Hi,

     

    I had a customer with a simular issue. Once a day the CPU spiked for several hours. LTM, APM, ASM and AFW are provisioned. Also a HA pair. I opened a case with F5 support and this is what they told me.

     

    "To confirm, this is definitely datasyncd and the ASM obfuscation process, from the logs, we can get a clear indication of when the obfuscation process starts and ends for each day.

     

    On v11.6 and later, datasynced periodically invokes the generator (a Java based application) in order to generate data for obfuscation and cryptographic processes (see K14020). When ASM initially gets provisioned, datasyncd and the generator run in 'urgent mode', taking maximum CPU cycles and so the unit goes to offline for a couple of minutes. While the unit is in day-to-day running, the task is scheduled every 24 hours and when the time comes (based on /var/log/datasync/datasyncd.log it is around 13:19hrs for this unit), it runs in 'normal mode' that limits CPU usage for the job. As CPU usage is limited, the job takes much longer time, like a few hours in general. This explains control plane usage on odd cores occurring every day staring around 13:19hours.

     

    Datasyncd will not commence the key generation tasks if it the device does not have enough CPU cycles available on it. This will explain why we see it change state regularly from QUEUED to RUNNING and vice versa. The time the overall tasks will take will depend on how busy the box is.

     

    In short this is normal behavior and F5 do not recommend changing the time of this process unless there is an impact to traffic or a degraded user experience, there is a feature request to add an easy way to reschedule this job from the GUI in a future version but this has not been completed yet."

     

    Hoped this helps.

     

    Regards, Martijn