High CPU usage

[lazer 38]

Hi,

 

I am running the new 5.071 beta firmware for some days now, and I noticed higher CPU usage.

 

On this cores usage chart, you can easily guess when I made the update :

Please login or register to see this image.

/monthly_2021_06/image.png.80177815f1158c1b1e250496d5d9bf5c.png" />

 

It didn't took me too long to find the origin : every device reporting energy property.

Each time the energy property is updated (new value in kWh), one core is used at 100% for few seconds.

 

See the diagnostic graph, in which case I have a QuickApp child device whose property value is updated every minute (whole house electrical consumption) :

 

Please login or register to see this attachment.

 

Other users on French forum have also reported the same issue.

 

It's getting worse when you have many energy devices (Wall Plugs, Dimmers, Switches, etc...), because each time a new energy measurement is reported, one CPU core is 100% used, up to a point where the HC3 gets very slow, or even unusable with dozens of devices.

 

Obviously, the new energy panel is be the source of the problem.

I guess some internal computation are made each time a new energy value is recorded.

 

I didn't see any user who have reported this issue on the forum right now, I hope Fibaro is aware and problem will be solved in next stable firmware.

@A.Socha are you aware ?

 

For the record, the above graph come from my test HC3 box, on which I have only 4 energy metering devices.

I can't imagine applying the new firmware on my production HC3 box, managing 91 devices with energy interface.

 

 

 

[Sjakie 135]

Hello,

On my HC3 I dont see this problem or I am looking wrong?

I updated 19-6  8.31

my CPU usage is equal as in the previous FW.

I have around 90 devices and I use some power values in QA. The energy panel I dont use but the data should be in.

Just for information.

//Sjakie

Please login or register to see this attachment.

[jyd 0]

Have the same problem with this beta version. I have a few devices (more than 20) reporting energy values, and at least one CPU is nearly at 100% and web interface is the so slow that it becomes completely unable to use.

Had to come back to the previous stable version.

[joseph91 0]

Hello,

I have the same problem with this beta version. I have some device reporting energy values and one CPU Core is at 100% (see figure). Another problem is that the web interface is so slow and it is difficult to use it

Please login or register to see this attachment.

[couillerot 89]

Hi,

since the last update, I have also seen an unexplained increase in CPU usage (nearly 100% !)... 

I have about 70 devices.

Stef

Edited June 23, 2021 by couillerot

[lazer 38]

18 hours ago, Sjakie said:

I have around 90 devices and I use some power values in QA

It's not the "power" value which is involved, but the "energy" value.

 

Usually, the power value is updated right after any power consumption change (examples : you have just turned on/off the lights or the TV)

 

The energy value is updated more rarely. It may be once every hour, or even day, it depends if the connected device has very low power consumption. Or maybe never if there is not electric consumption.

In that case, you might not see the higher CPU usage if you don't stay in front of the charts for hours...

 

As I said, I have a QuickApp device which report my main house energy, and it's updated every 60 seconds, so the issue is clearly visible.

 

[Sjakie 135]

Because of high memory use I disabled it a month ago

If I enable my energy O1 monitor the CPU looks like:

Please login or register to see this attachment.

 

//Sjakie

[m.roszak 470]

We would like to take a look, I am testing this on several devices with short interval of energy reports and my CPU usage looks normal

Please login or register to see this attachment.

 

[lazer 38]

Sure.

What to you need ? Remote Web & SSH access ?

I can provide you those information in private message.

[m.roszak 470]

Sure, this will be enough. 
Remote access for support will be sufficient.

Please pm me with details (Serial number of your HC3) - I will add a developer to the conversation and tomorrow we will check it out

[Brunanchon 5]

Hello, 

 

Have you found a way to solve this issue?

I use 2 x 3-phase meter from qubino, and my CPU load is probematic....

 

Please login or register to see this attachment.

[lazer 38]

Problem have been solved right after by a firmware update.... about 1 year ago !

 

So either you are running a very old firmware, or you have a new problem.

[Brunanchon 5]

thx, so I have to figure out why my CPUs become overloaded 

[petrkl12 203]

I had also problem with overloaded CPU and thanks to Fibaro investigation I know reason:

CPU was overheated and CPU frequency was cut from 1200 to 200MHz. Solution: put the gateway in cooler/more ventilated place

 

It was new information for me that CPU in HC3 can reduce the frequency in case of overheating ...

 

btw I have the newest beta 5.111.48

 

Please login or register to see this image.

/monthly_2022_06/image.png.4cbb486bf06e76442419e13ea46a0ec1.png" />

Edited June 30, 2022 by petrkl12

[petergebruers 1,261]

12 hours ago, petrkl12 said:

CPU was overheated and CPU frequency was cut from 1200 to 200MHz

 

I am old enough to have vivid memories of AMD CPUs burning

 

That was around 2005

 

I think most reasonable designs in this application field now have thermal throttling which may not only turn down the speed but also shut down indiviual cores temporarily...

 

 

[Brunanchon 5]

I think I might have figured it out in my case...

 

I have a LUA scene to monitor water consumption. I use a pulse water meter with a smart implant.

With each liter of water consumed, the scene is triggered.

This can generate a lot of occurrences when watering the garden in particular.

 

I use a consumption history table (day/month/year) stored in a global variable, which I implement each time the scene is triggered.

 

I feel like the repeated getglobalvariable, json.decode json.encode and setglobalvariable each time the scene is triggered is the cause of the CPU overload.
I transformed the scene to use only a simple daily variable (without table) and the problem seems solved so far.

 

Does that seem likely to you?

Thx.

[SmartHomeEddy 811]

13 hours ago, petrkl12 said:

It was new information for me that CPU in HC3 can reduce the frequency in case of overheating ...

Sounds logical, but your CPU load doesn’t look ok (or you just fully used the HC3]. 

[Martin_N 143]

Now that I have a very high CPU load (but not constant), I am thinking about needing access to some logging processes.
Only CPU load can be found via Swager and that still has to be calculated by periodic reading, so no glory + RAM either.
Now that we know that heat slows down the CPU and overclocks, it would be nice to have access to that information.
Thus:
1) CPU temperature
2) CPU speed
3) motherboard temperature
4) others..

 

eM.