HAOS 12.2 (with 13.2 there is a still unresolved problem with “brcmfmac: brcmf_set_channel: set chanspec 0xd0…fail, reason -52 ")
HA Core 2024.10.4
Supervisor 2024.10.3
Frontend 20241002.4
Addons list
Advanced SSH & Web terminal ver 19.0
with “start at boot” flag to create swap and then normally I manually stop it
DuckDNS ver 1.18
Home Asisstant Google Drive Backup ver 0.112.1
Mosquitto Broker ver 6.4.1
Network UPS Tool ver 0.13.2
Ring-MQTT with Video Streaming ver 5.7.2
Description of various situation after almost 2 years of HA running on my PI3A+
In every situation (described below) where the system is not responding (no web interface) I always noticed that the Observer (except very rare cases) says:
Home Assistant observer
Supervisor: Connected
Supported: Supported
Healthy: Healthy
The “standard running time” (before spontaneous restart due to watchdog) is normally few hours… so normally I see 3 to 5 HA core restart every day… I had to accept this from beginning…
After a reboot for any reason (not HA Core restart but full Device/OS manual reboot) it can happen that the used memory after HA is totally started is about 70/72% and in this situation the system is totally unstable (which means it can restart every hour or even more frequently) … Need to do several reboots or Core restart to get a stable situation. I see a stable situation when memory usage is about 84/85% (this is my indicator to understand if HA will run in a stable way or not from now on)
In some circumstances where the system is not responding I see that the SD activity is almost always running (green led always ON) for hours and hours … and in this situation it’s useless to force reboot (from local console) because even after a reboot it continue to do this for hours and hours (in some case up to 2 days)… While in this situation, I noticed that normally HA automatically disable MQTT broker at next reboot… then HQ starts normally and I have to manually re-enable MQTT broker and then it works fine (it’s like HA takes up to 2 days of full SD activity - curious to understand what is it doing? – to realize that disabling MQTT broker the system returns to be stable)… The it can run for several months without this problem…
Given the previous point MQTT appears to be the most problematic component that affects system stability (is it the “broker” or the “integration”?)… several times I also see various errors related to MQTT even when apparently it is working fine…
Netatmo official integration (Netatmo - Home Assistant ): for over one year there was no problem… recently it happens quite often that this integration doesn’t start and need to restart it manually (I created a button in the home web page to restart it without going to device/integration)
Tuya official integration (Tuya - Home Assistant ): for over one year there was no problem… recently it can happen that need to re-authenticate it 3 times in a week after subsequential reboot … or never… it’s totally random.
Duckdns certificate expiration service most of time doesn’t work… sometimes if I restart it manually it works but not always…
Most of the time if I don’t access the web interface (I only access some sensor I have as widget on my mobile phone/WearOS device) it doesn’t restart for days… so it appears to be more stable if I DON’T use the web interface (or the mobile app) versus when don’t using those…
Most of the time, with HA running from quite long time, as soon as I access the web interface (or mobile app) it starts to show problems (restart/web GUI not accessible, ecc)
Few days ago (Core 2024.10.4) I had the HA running for over 2 days (a kind of miracle) but during these days I wasn’t accessing at all the Web interface… and then I noticed that SSH and web terminal add on was stopped (I incidentally did stop it without noticing it)… This appears to mean that also this component affects system stability/durability but in the reality even if I keep it always disabled the system can have stability problems anyway…
Quite often the cause (well, the visible one) seems to be the API Call error (Timeout → watchdog) Typical log is:
2024-11-04 10:03:34.411 ERROR (MainThread) [supervisor.homeassistant.api] Timeout on call https://172.30.32.1:8123/api/core/state.
2024-11-04 10:03:36.557 ERROR (MainThread) [supervisor.misc.tasks] Watchdog missed 2 Home Assistant Core API responses in a row. Restarting Home Assistant Core! Home Assistant automatic restart for API call error? - #54 by remimikalsen
My hope is that some of HA developers would really be able to understand why all of the above happens and fix the issues… so that system can run in stable way…
If anyone wants to get some specific logs to better understand the situation feel free to ask me… there are tons of different logs so I need to be exactly guided on which one to provide.
Well… In my case it is running from about 2 years now… and apparently it is used by over 1300 users (source https://analytics.home-assistant.io/)
Ram never goes over 85% and there is also a 2GB Swap
CPU is always running at 2-4%
Not to dogpile, but I used to run HA on a Pi3B+ (so twice as much RAM as you) and while it worked, I had some similar issues as you’ve experienced (random restarts in particular). Moving off an SD card and increasing the swap size helped some, but only a little. Spent a lot of time chasing my tail before upgrading to better hardware (specifically more RAM and an SSD) which fixed every issue I had.
From what I am seeing/experiencing I personally believe thar if there was good intentions to optimize code it could run also within SD and 512Mb…
But of course it’s easier to add features without optimization … relying on the fact that there is powerful enough HW to run the code…
One example is the “fixed” timeout for API call… which often cause the Core restart.
I was suggesting to make it adjustable based on owned HW but noboby will modify that…
Please follow the others’ suggestions above and get better hardware. The Pi3 (of any variant) is too low on RAM to be reliable nowadays. In my opinion, the docs should remove any mention of the Pi 3 from the minimum requirements.
The issues you’re seeing with different addons aren’t tied to a particular addon. The cold truth is that a Pi3A does not have the resources necessary to run all addons on startup at once without potentially running into issues.
The reason you were seeing small periods of stability is probably because your addons didn’t restart at the same time, or because watchdog gracefully handled restarting in the background when they failed.
I’ve read your “Big Thanks” post, and noticed that your joy only lasted two weeks. Your following posts in that thread (and the creation of this thread) seem to confirm that you are not running in a stable environment.
Yes, you could increase the swap to 2GB like you did, but that will massively increase wear on your SD card. Yes, you did a good job of getting an Endurance version, but then you went and got the lowest capacity card required to run HA. Essentially, you’re shaving off 2GB out of the minimum storage requirement and continuously hammering away at those 2GB.
Software bloat is a factor too. As time goes by, database history (both short & long-term) will keep accumulating, entities might increase, backups will take more and more space if you forget to delete them, and recent OS updates have not been very kind to the lowest-specced machines.
Please consider buying (at least) a Pi4, 4GB version if power consumption is your main concern. While you’re there, buy a new 64GB Endurance SD card, because your current one will probably have at least 2GB worth of sectors who are nearing their end of life.
Ask yourself - do you want to babysit a 6+ year old machine which isn’t even supported and makes you wonder whether you can even access the web interface? Or would you rather bite the bullet and get a more robust solution which will avoid these headaches?
Thanks for your honest opinion…
I am now evaluating to migrate to PI5 which apparently is even more powerfull (and it cost 55€ now)…
I hope that backup/restore process will maintain my entire configuration done in about 2 years (I couldn’t remember all the modifications done)
Thanks
Do a full backup of current. Pull usb sticks, etc.
Shut it down.
Build new hardware. Give it the OLD HA box IP address, install HA according to. Your hardware instructions. As long as you select a build type that uses supervisor you’re fine.
Restore your backup.
Shut it down.
Install USB sticks etc
Start the box.
Check everything.
Get used mini pc and you never have perfomance issues again. I bought one for 50€ i5 quad core processor, 16gb ram and 256gb ssd. Cheaper and more realiable choice.
I couldn’t believe…
Going from PI3A+ to PI5 (with SSD) totally changed the game…
It’s not 2 or 3 or 5 times faster… It’s 200 times faster!!
The backup/restore process was quite smooth and really fast: I saved the Backup locally on SD using CLI command with Core stopped, then read it on my Windows PC wit Linux reader and uploaded to new system at 1st run/initial wizard on RPI5…
I just had a doubt about the fact that the restore wizard apparently remained stalled (the step after the backup upload)… but really it was completed…
This RPI5 (55€ + 16€ for PS) was the best purchase I’ve ever made
After few days or RPI5 usage…
Is it “normal” this slow but constant memory increase??
I think I have to schedule (with automation) HA restart before it reach a too high memory usage
sure… and thanks for the reply…
For now I am monitoring the situation…
Will see how it behaves after many days of running (which it was never possible with RPI3A+)
Thanks again
