Electron CPU usage

martinkooij

@mattn thanks! I love my modules, so difficult to lose one ;).

It however seems you have a slow leak too, growing memory usage of ca 200Mbyte in 6 days. Much less than me, but still more than it should be, I guess. Just let us know how you fare! Thanks for sharing.

martinkooij

@mattn and others.
I have a stable situation with the following:

electron 1.4.15 (withELECTRON_FORCE_WINDOW_MENU_BAR=1).
higher elektron version gives black screen, the environment variable lowers the cpu-usage by elektron considerably).
Magic Mirror 2.3.1 (actually I was quick to install MM2.3.0 but manually downgraded electron, but I am quite sure that amounts to the same as installing MM 2.3.1).
nodejs version v9.11.1. (I used to have a 1.6.x nodejs, but that one caused memory leaks. )

Is now running in this configuration almost 7 days without interrupt. And no signs of stopping soon. Thanks you all in this forum for the help.

(Rasberry pi 3, Jessie, CPU Hardware: BCM2835, Revision: a02082)

cdelaorden

Hi guys

Im new to the forum and this project but I have some small experience with the Pis and having the same issue

I think that its related to the video driver, using ‘sudo raspi-config’ and changing the video driver to one of the GL drivers resulted in a much lower cpu load in my case.

Doesn’t matter if you choose full KMS or fake KMS as long as you dont use the legacy non-GL driver.

It took my 4 CPU from full 100% to idle bellow 35%

Im sure someone else with this problem can confirm this

E3V3A

@cdelaorden

How can we check the video driver without using rapsi-conf?
in dmesg? in systemd journal?
How/where do you set the: ELECTRON_FORCE_WINDOW_MENU_BAR=1.
What would be the easiest way for users to check electron version from command line?

cdelaorden

@e3v3a
You can check it in the config.txt

or just do a “tail /boot/config.txt | grep dtoverlay” to check what version is enabled

all the rest of my configuration is the default one after a fresh install from the current version of MM and Raspbian, I havent touched anything from electron

Juju

Omg, yes finally! :D
First I followed this installation guide: https://forum.magicmirror.builders/topic/236/complete-setup-tutorial/3
I had the electron CPU issue. So I downgraded to electron version 1.4.15. The CPU usage went down from 90% to 70%. It was still laggy as hell.
I then formatted the microSD card and followed that youtube guide: https://www.youtube.com/watch?v=0Q04SKvibik
Downgraded to 1.4.15 again and now it’s running great with a CPU usage of like 5% with spikes to 50%, when there is an animation on the screen.

But I have one question.

After I finished the youtube tutorial I have the “MagicMirror” and the “mm”. Both autostart when I boot the raspberry pi.
Which use has the “mm”?

Mykle1

@juju

pm2 stop mm
pm2 delete mm
pm2 save
rm ~/mm.sh

MichMich

@cdelaorden I think you’ve found the issue! Awesome work!

For others please note that when you start raspi-config and go to “Advanced Options” > “GL Driver”, the top one will always be selected, no matter which one is active. After selecting one of the two top drivers (GL), it might solve the hi CPU load issue …

It solved it on a clean installed Pi 3 for me.

MichMich

I added some info about the Open GL driver to the Wiki:
https://github.com/MichMich/MagicMirror/wiki/Configuring-the-Raspberry-Pi#enable-the-open-gl-driver-to-decrease-electrons-cpu-usage

Hein-Jan

@michmich

I seem to be running into some kind if trouble. My SD card failed, so I had to do a clean install. I decided to immediately go for the full KMS driver.
“$ tail /boot/config.txt | grep dtoverlay” results in “dtoverlay=vc4-kms-v3d”.

The CPU usage is alright, but “/home/pi/.pm2/logs/mm-error-o.log” states:

libGL error: MESA-LOADER: failed to retrieve device information
MESA-LOADER: failed to retrieve device information
MESA-LOADER: failed to retrieve device information

I am having trouble with the monitor. I am using MMM-Remote-Control.
“MONITOROFF” results in “/home/pi/.pm2/logs/mm-error-o.log”:

Powering off HDMI

“MONITORON” results in “/home/pi/.pm2/logs/mm-error-o.log”:

Powering on HDMI with preferred settings

[E] Failed to power on HDMI with preferred settings

{ Error: Command failed: tvservice --preferred && sudo chvt 6 && sudo chvt 7
[E] Failed to power on HDMI with preferred settings

    at ChildProcess.exithandler (child_process.js:217:12)
    at emitTwo (events.js:106:13)
    at ChildProcess.emit (events.js:194:7)
    at maybeClose (internal/child_process.js:899:16)
    at Process.ChildProcess._handle.onexit (internal/child_process.js:226:5)
  killed: true,
  code: null,
  signal: 'SIGTERM',
  cmd: 'tvservice --preferred && sudo chvt 6 && sudo chvt 7' }
Powering off HDMI

I may be wrong but I feel this is related to the Full KMS driver.

Can anyone confirm? Does anyone have a solution?
I am way out of my depth here…

Dom1n1c

@Hein-Jan I got the same issue.
A few minutes, the monitor could be controlled, but then “tvservice -p” fails to power on the monitor.

cdelaorden

@hein-jan @Dom1n1c

Please use the fake kms driver instead “dtoverlay=vc4-fkms-v3d”

it its known that the full kms driver isnt compatible with tvservice plus other restrictions

EDIT:
Also with the fake KMS you can rotate the display from /boot/config.txt so I suggest @michmich using it as the preferred method in the wiki

schmo90

@cdelaorden said in Electron CPU usage:

the full kms driver isnt compatib

hy now i use the fake kms driver, everything is working fine and realy fast.
but i have the problem that the mirror will go to a black screen after some hours.

only a pm2 restart mm helps in this case -.-

cdelaorden

@schmo90 mmm havent had that problem, everything its working fine

Have you deactivated the screensaver as its shown in the wiki?

Also, are you using a module like MMM-Remote-Control to turn off the screen?

If you see a black screen and the mouse but nothing else you are probably missing “sudo chvt 6 && sudo chvt 7” in your statement after turning hdmi back on. You can test it in ssh after forcing the issue.

Also if you are concatenating commands like Hein-Jan did with “tvservice --preferred && sudo chvt 6 && sudo chvt 7” have in mind that commands after && are only executed if no errors are found. So maybe “tvservice -p” its working correctly but its giving some feedback and things after && aren’t executed.

You could force it by using ; as a separator instead of && so it doesn’t look for an answear

schmo90

hy i didnot turn off the hdmi,

i will turn off the lcd with the gpio’s

screen is turning on normal but electron browser is showing a black screen (mous appears at moving)

binderth

adding another flavour:
As my Foscam FI9900P got an ONVIF-Update lately and is now capable of RTSP-streaming I installed MMM-RTSPStream.

activated fake (or full) OpenGL Desktop driver:
omxplayer crashes
activated non-GL desktop driver:
omxplayer runs

So, as I see it, the omxplayer problem doesn’t concern MagicMirror - is there another way to reduce Electron CPU usage - except using an Open-GL dekstop driver?

PS: activating fake KMS reduced CPU usage as recommended in this thread - but I’d like to have the video-stream of my cam on the MM (unfortunately it doesn’t provide MJPEG)

cdelaorden

hi @binderth
the fake-kms was created for keeping omxplayer compatibility but you need to add the “–no-osd” flag to the omxplayer command line

you might want to try it

dont bet on it, but changing line 101 of node_helper.js to

var args = ["--no-osd", "--live", "--video_queue", "4", "--fps", "30",

might do the trick

vinp

MM 2.4.1 - if anyone is trying to look for it while searching for forums. Apologies if it has already been stated

So i’ve done it both ways:
dtoverlay=vc4-kms-v3d
I noticed that my CPU temp is well into the 60-70 degree Celsius with CPU loads mostly in the 1.x, 2.x, and sometimes 3.x ranges

When i try the other way :
dtoverlay=vc4-fkms-v3d
CPU is much lower in temp under 50 degrees Celsius and CPU load rarely goes above 1.x
and i am able to use the tvservice -o and tvservice -p to schedule my monitor to go on and off at certain times

Hopefully this helps anyone searching for this issue on new installs like myself

jaffons

So i ran into this issue as well. I started MM stuff 2-3 weeks ago.

When everything was working my logs show:

I had the previous mm version ( probably 2.3.1)
Enable the Open GL driver, fake KMS (via raspi-config)
Disabling the screensaver https://www.raspberrypi.org/forums/viewtopic.php?t=57552
Autohiding the Mouse Pointer
rotate screen 270deg(CW), display_rotate=3 at /boot/config.txt
I also had the dtoverlay=vc4-fkms-v3d line there
I was getting this error though (thrunpm start ):

libGL error: MESA-LOADER: failed to retrieve device information
MESA-LOADER: failed to retrieve device information
MESA-LOADER: failed to retrieve device information

Anyway I messed something on my modules and had to reinstall the Raspbian. Same as before, Stretch 7/-18

Rotate screen at /boot/config.txt
installed the MM 2.4.1 and tried it straight away default config and modules. It was working alright
Disabling the screensaver https://www.raspberrypi.org/forums/viewtopic.php?t=57552
Autohiding the Mouse Pointer
I think I added the dtoverlay=vc4-fkms-v3d line…
- and boom animations started lagging (+high CPU)
- also npm start dev didnt work. It launched but no mirror was shown and if i had mm running it would stop refreshing.

Any way I figured out that taking that commenting out that # dtoverlay=vc4-fkms-v3d everything works flawlessy. I dont get the “libGL”-error
and I also updated npm 5.6.0 → 6.2.0 sudo npm i -g npm and npm start dev started working (dev or no dev).

My Rpi temperature is under 50C when mm is running (no heatsink).

Hope this helps out someone!

adamlove86

@vinp and @jaffons WOW! Thank you for your comments… Made a huge difference.
After some months I needed to reinstall from scratch (now also hosting octoprint and a remote desktop simultaneously). I thought MM animations were just generally slow, but this made a huge difference compared to the more orthodox setup.
I don’t see why this configuration shouldn’t be preferred over the normal Pi configuration, and the OpenGL driver was also much trickier to get working.