"Out of memory" issues - where do I begin?

ember1205

Up until yesterday, my MM had been running reliably over a period of days without blanking the screen or locking up.

I have a few default modules (calendar, currentweather, and weatherforecast) enabled and only one third-party (MMM-ImagesPhotos). After correcting some CSS and config.js mistakes over the last couple of days so that it displays what I want, and how I want it, I decided to move on to tinkering with the images themselves.

I have a lot of photos that I am using in the background. Many were shot on my DSLR (24MP) and some came from other DSLR’s (Disney’s Photo Pass) and some smart phones. Using the default image size (3:2 ration) for landscape left gray letterbox bars on the sides of the screen. I began cropping the photos from 3:2 to 16:9 and replacing the original files.

While the photos display mostly “perfect” (for some reason, I am still seeing what looks like a 1-2 pixel “frame” around many images that should be full screen), the RPi (Model 3 v1.2) is back to blanking the screen after some period of run time and the logs show “Out of memory.” I don’t know if it’s related, but I was using a different module to display photos in the background (MMM-BackgroundSlideshow) and was using the exact same set of photos. With that module, I had not cropped ANY of the photos but was getting the same sort of “Out of memory” behavior.

I’ve seen some other possibly anecdotal ‘reports’ of certain image-related details causing this sort of error, so I’m wondering if it’s at all related. I also read through a thread from not-that-long-ago where various bugs in the notification module were being fixed (and the fixes being tested), but there wasn’t any follow-up in that thread as to the status.

I’ve also tried increasing swap space on the MM, but that has not made any difference.

Any tips on how to go about starting to troubleshoot this? The device itself remains 100% responsive and network access is fine. I can send signals to turn the display on and off and that works as well. The only way to “fix” the solid black screen is to restart MM or the entire device.

sdetweil

@ember1205 the only way I have found to prevent caching is the startup parm,

there is another way, but I haven’t been able to make it work in MagicMirror

const { webFrame } = require(‘electron’)

and then, periodically
webFrame.clearCache();
https://discuss.atom.io/t/single-page-application-show-image-use-too-much-ram/18722

BUT… this won’t work in a node helper, and I haven’t found a way to get access to electron in the module.js (as there are no requires() allowed)

I do have this working in a different mirror runtime, and it does what I wanted.

ember1205

I’m not sure that this is a caching error as it seems related very directly to the displaying of images but “rears up” differently based on the module. In the prior module I was using, displaying ANY graphical content lead to a quick demise of the system (within 15 minutes or so, typically). With the current module, everything was 100% fine until I cropped some of the images. The current module is one you are using with zero images, so there has to be something about the crop method I used that threw things out of whack.

Even if it IS a straight-up caching issue, I can’t disable caching because the image display process looks poor. I’ve tried with caching disabled and it is not at all smooth, and I also haven’t tested that anywhere near enough to actually validate that it corrects anything.

Is there a way for me to be able to at least understand the process in effect when this error gets thrown? The entry in the error log seems super generic with no real information along with it.

sdetweil

@ember1205 i have been chasing these gremlins for the last 3 years without any real success.

the addition of the call to webFrame.clearCache() has eliminated all my instability

with the updated MMM-ImagePhotos you shouldn’t have any slow painting… as my code waits til the image is ready to show. I use it EVERY day on multiple systems and OS’s) , and don’t see any slow paint.

ember1205

Turns out I’m wrong on two counts…

I -have- been testing the caching disable extensively as it’s currently disabled. :)
There is not a slow-paint issue for images with caching disabled.

Unfortunately, I can now say that I’m pretty comfortable with the assessment of this NOT being a caching issue. With caching disabled for electron, the issue has now occurred multiple times in just the last 4-5 hours.

sdetweil

@ember1205 s, I might try going thru each of the images u have adjusted with the electron browser on the pi and see what happens

from a commandline

cd ~/MagicMirror
export DISPLAY=:0
pi@raspberrypi4:~/MagicMirror $ node_modules/.bin/electron file://{url to pic}

mine looked like (from my mounted server volume)

 node_modules/.bin/electron file:///media/buildserver/media/Photos/DCIM/2018\ vacation/Page\ Arizona/DSC_0323.JPG

then close the browser, and use a different url

now… we are running electron 3.0.13, which is quite old…
you could try upgrading that to something newer… next version will use 6.0.12 (if my change is accepted)

ember1205

What would I be looking for?

sdetweil

@ember1205 the browser having trouble, crashing maybe

ember1205

I used an iterative command for this because there are so many photos:

cd ~MagicMirror
find < path to images >/* -exec node_modules/.bin/electron file:///{} \;

For each file in that directory, it would launch the electron browser and open the image. Because electron continues to run until it is terminated, this puts one image at a time on the screen. After each one loaded, I would simply click “X” to close the browser and it would move on to the next image.

While I -did- see one specific image file appear to load but not actually show any content (initially), I believe this may have been an issue with the fact that I walked away from the process for about ten minutes. Expanding the browser window caused it to re-render and it displayed fine. Additionally, this is an image that I have personally seen on the screen many times with no apparent link to any crashes.

It seems that any image-specific issues are possibly cumulative since loading each one in turn starts with a fresh memory allocation as the browser exits after each image.

ember1205

Thought I would update this…

Still broken.

I have the latest MM as a fresh install on a brand new RPi with a new power adapter, using a brand new SDCard, connected to a different monitor, using a different HDMI cable, and it’s all plugged in to an outlet at an entirely different house in a different area of the state (different sections of the grid). I have the same few, basic modules installed and have added the photos module discussed here. It crashes and hangs randomly with “out of memory” errors just like the other one.

The only thing the same between the two is the photo collection. I’ve not found any consistency in the past on it failing with any specific photos so troubleshooting this seems like it’s going to be impossible.

sdetweil

@ember1205 debugging is indeed very difficult w these problems.

I wonder if running w chromium browser would produce different results.

easiest way to test
download the run-start.sh script from here
https://github.com/sdetweil/MagicMirror_scripts
into the MagicMirror folder
edit the package.json and change the start command to

"start":"./run-start.sh",

edit config.js
and set

 serverOnly:"local",

this will start mm in server mode, and use chromium to display instead of electron

ember1205

Would that be any different than running it in server mode and accessing it from a browser on my computer across the network? That would be simpler to set up and wouldn’t add more running software onto the Pi…

sdetweil

@ember1205 same, no new software on the pi. mm install installs everything.

can run as
full ---- how it is now
server only
server only w different browser ( already installed)
or
client only

ember1205

Understood that there isn’t new software installed, but this would mean no additional software actively running on the Pi (Electron would be stopped, and all browser software would be running from a different machine).

sdetweil

@ember1205 right. that doesn’t mean u fixed the out of memory problem. it’s possible.

I was working on one change at a time.

u can install the run-start.sh and change config.js to serverOnly: true, then launch browser from your pc

ember1205

Your suggestion incorporates multiple changes, though… Stopping Electron and Starting Chromium while also changing the operating mode. No?

I’ve changed the config.js to both

serverOnly: true,

and

serverOnly:"true",

and neither one prevents Electron from starting.

I’ve now also tried

serverOnly:true,

sdetweil

@ember1205 u need the run-start.sh script for that parm to work

besides, u said u were going to run in server only mode anyway. and then u will have to change your startup process too

ember1205

By adding an external browser, it crashed in only a few minutes. I’m going to let it keep running with the additional external browser to see if it causes it crash much more quickly on a more continuous basis.

If I can cause it to crash quickly on a repeated basis, it gives me an opportunity to try testing specific image files. It’s possible that this is a cumulative problem that is related to a small memory loss that gets added to on the load of additional errant files.

UPDATE: So much for that theory… I am currently running Electron plus an external instance of Chrome on Windows and Safari on MacOS. No “quick crashes.”

sdetweil

@ember1205 i spent a year debugging an out of memory that only crashed one version of a module… another ran fine…

one line of code, my code, made the difference… but the problem ‘appeared’ to be a hang , suddenly the UI was dead…
randomnly displaying pics in the background.

not til I added a totally new logging approach did I find the actual problem, which was out of memory, which killed the UI, which made it LOOK hung…

ember1205

In my situation, the UI goes all black and the cursor appears (no mouse or keyboard connected). There’s a generic “out of memory” error entered into the log file. Restarting MM is the only way I’ve found to get it back up and running.

The photo module I’m running is your fork with the more recent changes to allow the blurring of the same image behind the full-view image. Any thoughts on anything that could/should be added to try and capture more in the logs? Is there an increased logging level I could be using in MM itself? Stack traces when it fails? Before it fails? Anything come to mind that might give me a new avenue?

The issue doesn’t seem to appear if I am not using the photo module. But, that could be because the other, more basic modules don’t use resources as intensively. We’ve compared notes on RPi, kernel, build, etc. and not found anything that seems to be materially different with the one exception being the specific image files in use.