Performance issue with psusensor / dbus-sensors

[Vijay Khemka]

On 2/14/20, 2:30 PM, "openbmc on behalf of James Feist" <openbmc-bounces+vijaykhemka=fb.com at lists.ozlabs.org on behalf of james.feist at linux.intel.com> wrote:

    On 2/14/20 2:14 PM, Peter Lundgren wrote:
    > We're running into some occasional and hard to reproduce performance 
    > issues with sensors on the entity-manager/dbus-sensors/intel-ipmi-oem 
    > software stack. I don't have much concrete to say on the subject, but I 
    > want to put a feeler out to see if anyone else has seen similar issues. 
    > Here's what we think so far:
    > 
    > Complaints range from IPMI sensor reads being slower than normal to 
    > sensors "never" updating.
    > 
    > Josh got access to one machine in a bad state and observed this:
    > 
    >  1. All the I2C buses were working normally. i2cdetect ran successfully
    >     on each bus.
    >  2. hwmon was working fine. He wrote a shell script to read all of the
    >     *_input sysfs files and could read every sensor in the system in 3
    >     seconds.
    >  3. psusensor was running.
    >  4. busctl --no-pager monitor | grep -i PropertiesChanged shows no
    >     traffic. On a healthy system, it shows many updates per second. No
    >     obvious error messages in journalctl --no-pager -f.
    >  5. Restarting psusensor alleviates the problem.
    
    Not sure if it's 100% related but Jae is looking into an issue with the 
    CPU sensor not reporting data after DC cycles. We haven't root caused it 
    completely yet but we'll let you know what we find.

I have also seen many time cpu sensors not reporting data after power off and on,
I didn't test enough or follow it through. But I see some issues on i2c not sending ack
For few messages after power on.
    
    -James
    
    >