I am fighting a bit with my installation of HA. It uses almost all memory it can get, like almost 8GB of RAM but only 27% of 4 CPUs. It is running in a Proxmox VM and was running till one of the latest updates.
e-Welink now fails and UI integration cannot open the console. Tasmo Add-On cannot get nginx server running.
It feels like everything is falling apart.
#home assistant slow and using loads of memory
1 messages Β· Page 1 of 1 (latest)
proxmox>HA VM>Hardware>advanced>enter value max RAM>DISable ballooning. Helped for me at least.
2cpu is enough also.
It uses almost all memory it can get
That's normal for pretty much every OS you virtualize: https://www.google.com/search?q=wrong+memory+usage+site%3Aforum.proxmox.com
27% of 4 CPUs
That seems high. Install theglancesaddon, open it and presszandcand see why.
I can't see how ballooning should help with this unless your host has > 80% memory usage. Right but giving a VM more cores usually does not hurt.
got that tip after memory increase after installing music assistant. So you think this is unrelated? Then i could set it back.. Will try
must be dns then π
Due to caching (see link) a VM will usually continually get more and more memory assigned and will use it as cache or whatever it wants to.
This memory is only taken back if the host reaches 80% memory usage (by default) and ballooning kicks in.
I'm not saying don't use ballooning. I usually use it. I'm just saying it shouldn't change the cache behavior under normal circumstances.
thanks
That HA just eats what it gets. And I have other virtual systems going fast on 512MB of RAM and a single core
And that Ballooning option was set already.
Unrelated but you should configure the guest agent.
As for CPU usage. Install the glances addon, open it and press z and c and share what you see.
There are reports about that high usage of RAM and initially that was blamed to Proxmox showing the "out view" of all RAM used, but inside HA the system just built up larger caches but not real HA RAM usage. However that request for cache must come from somewhere.
But then there where also some add-ons found that blew up memory.
Then also press m and share that as well.
@drifting solar Wilco. Just let me check how to get in there to add glances
I need to figure out how to select the right ssh key.
Also relevant: https://www.linuxatemyram.com/
For what? Install the addon via the GUI.
I am taking my first steps on HA and now I got two add-ons that require ssl keys. But I use my proxy manager on Proxmox to provide the HA ssl encryption
Ah. Because you wrote SSH.
Ups
glances doesn't start. Just falls back to START
Other then with the failing e-welink there is no log coming up
Check logs. You likely have to disable protection mode.
Ingress error: cannot connecto to host 172.30.xx.y:64858
Oh, even I left ssl disabled, it argues about ssl...
Is there an option to get "supervisor logs" and filter it for ERROR only. The Proxmox console into the VM is not very cooperative with textual cut and paste. Only screen snippets are possible
ha supervisor logs | grep -i "error".
About the copy pasting: https://gist.github.com/Impact123/3dbd7e0ddaf47c5539708a9cbcaab9e3#enabling-a-vms-serial-console
That command gives an error about | is not a valid command line thing
I expected you to use a SSH addon, not the VM CLI.
I am sure I was on the linux command line yesterday
Use login first in the VM CLI.
It looks like I am using a PVE
Can't really judge without seeing a picture or something.
I meant the CLI where you wrote the command π
Its the virtual console kvm
Okay, that is nuts. HA doesn't provide any serial-getty systemd
found it but cannot copy it as the /etc/ filesystem is read-only
HAOS is very opinionated appliance. What are you trying to do?
Actually it worked absolutely fine. Then a few updates came in and it went crasy in memory usage, e-welink add-on stopped working and I am sitting hree with a load of new sensors that I can't use
The user panel is working very sloppy and slow, but it was fast a few days ago
I had glances on "start at boot" what failed after adding a serial console and repbooting
Now HAOS can't get of the error, that the add-on wasn't willing to start. I disabled it in the bubble, now HAOS hangs
Let's try to keep this on topic with the memory. Log in on the OS CLI and run these
top -co%MEM
docker stats
Looks okay at the moment.
Yes
For the other issues I'd recommend
ha core logs -vf
ha supervisor logs -vf
currently memory is also in Proxmox reported as <4GB of 8GB
That one repeats
Nothing special here
Btw! Thank you for supporting me! Really appreciate it
A reboot of the HAOS got rid of the persistent error about glances
You have a serial terminal now, you can copy & paste π
I think the version warning is to be fixed upstream.
Serial: Sweeet! 
Works, so now what do you need next?
Interestingly, even only half of RAM is used and CPU load is low, I have that effect that there is the message like disconnected, reconnecting on the bottom of the screen
I need nothing. I see no memory problem to be fixed and that's why I opened the thread π
I'd probably try and see if it happens in safe mode too. Also check the browser's developer console and follow the core logs as mentioned earlier. See if a log message happens when a disconnect happens.
Unfortunately that is not the case.
Memory stays below 4GB, but HA still is slow on reactions to clicks and the disconnect doesn't drop a line in any of the two logs (supervisor or core)
Not currently. Got glances to work
Glances is not showing any value that makes me wondering.
With the console and glances at hand I do have some more tools to track the issue when it re-appears
Okay, next thing to address are these e-welink things... Would love to make them independent of Chinese repos and clouds. But that is another topic. And it is >10:35pm and the alarm goes off at 06:00am.
Maybe you can replace them with a zigbee device.
Thank you for your help!I learned something and know now what to look for before re-openign this thread. I really don't like if things work after just rebooting. It is like slapping a thing and then it works.
I don't want to replace them, just free them.
Unfortunately that happens more than I like with the supervisor.