Device network configuration resets during power cuts

[santievil]

Unfortunately, power outages are common in the area where I live.
Since I’ve had HyperK installed, I've noticed that when the power goes out, all settings are saved except for the network configuration.
The AP (Access Point) reappears, and I have to connect to it again to reassign the Wi-Fi.
I'm not sure if this is a bug or if it's supposed to be this way... but I’d swear this didn't happen with WLED.

[awawa-dev]

Hello, could you specify the board model and the exact settings you are referring to so I can attempt to reproduce the issue (if it's an actual bug it should be easily reproducible).

[santievil]

Its a GLEDOPTO GL-C-015WL-D (ESP32 dont know what version).

/json/info return this:
{"state":{"on":true ,"bri":255 ,"mainseg":0,"lor":0,"seg":[{"id":0,"fx":0,"pal":0,"sel":true,"cct":0,"o1":false,"o2":false,"o3":false,"si":0,"m12":0,"bri":255,"start":0,"stop":167 ,"on":true ,"col":[[0 ,0 ,0 ],[0,0,0],[0,0,0]]}],"nl":{"on":false,"dur":60,"mode":1,"tbri":0},"udpn":{"send":false,"recv":false}},"info":{"ver":"0.15.3","vid":2508020,"cn":"Hyperk_0.0.4" ,"name":"Hyperk" ,"mac":"3c8a1f05add4","arch":"ESP32" ,"uptime":26445 ,"live":true ,"freeheap":181396 ,"leds":{"count":167 ,"maxseg":1,"lc":1,"seglc":[1],"cct":0,"wv":0,"maxpwr":0,"rgbw":true },"wifi":{"rssi":-54 ,"signal":92 ,"channel":11 },"fs":{"u":16,"t":61,"pmt":0}},"effects":["Solid"],"palettes":["Default"]}

Settings saved: Led Hardware and Light Preset (all settings).
Settings losed: Wifi configuration (no SSID and pasword saved).

Its like Network->Switch to AP option is activated.

[awawa-dev]

Is this a consistent behavior during a power loss or disconnection from power grid?

[santievil]

Yes, it has happened every time the power went out since I've had HyperK Installed (2 times at least , with the help of my microwave, which keeps resetting the time).

I've tried manually disconnecting the power, but I haven't achieved the issue (settings were kept). It seems to behace differently during a real power grid failure than when simply unplug it.

I've run out of things to try.

[awawa-dev]

So it’s not going to be that easy to reproduce after all ;-) I can only take an educated guess: in version 0.0.4, if Hyperk detects a configuration file but it’s not a valid JSON (i.e., it's corrupted), it formats the partition and creates a new one. Unstable power supply can cause a "dirty" start where the LittleFS partition fails to mount properly or mounts only partially, resulting in the config read returning garbage data. Alternatively, the flash might not be detected at all, in which case Hyperk starts with default settings - though in that scenario, a subsequent hard reset when power supply is stable should fix the issue.

There are no perfect solutions here:

• Stick with the current behavior.
• Hard-block the Hyperk boot process until LittleFS mounts and a config file is either read or recreated - but in case of persistent LittleFS issues or corrupted JSON, this would lead to a bootloop, effectively bricking the device.
• In case of a "dirty" start, boot with default settings without formatting the partition or saving a new config. However, this is somewhat pointless, as we wouldn't be able to read the SSID/password required to connect to the network anyway.

[awawa-dev]

Hard-block the Hyperk boot process until LittleFS mounts and a config file is either read or recreated - but in case of persistent LittleFS issues or corrupted JSON, this would lead to a bootloop, effectively bricking the device.

There is one possibility to avoid bootlloop but probably only for esp32: a boot counter stored in RTC. I'll test it.

[santievil]

You know, I never make it easy.
I would swear that the first time it happened (and I didn't think much of it) was with version 0.0.2.
The other times the power went out, I had WLED installed and it never lost any settings.

[awawa-dev]

I would swear that the first time it happened (and I didn't think much of it) was with version 0.0.2.

That’s important information. It means there’s only one place in the code that formats the LittleFS partition when it fails to mount after reboot. I’ll add a retry/timeout to Hyperk to make it more graceful before it reformats the partition.

[awawa-dev]

But are you sure that restarting the device doesn't help when this happens? The behavior is identical to the fail-safe mechanism that switches Hyperk to AP mode when it can't connect to Wi-Fi (without formatting anything, so the config and LED settings remain valid). Maybe the router also stopped working after the power outage and takes longer to reboot than the ESP32, even though Hyperk waits for about 12 seconds. Next time it happens, make sure to check if the LED settings were restored. Then, perform a full power cycle of the ESP32 once the Wi-Fi network is available and check if it reconnects.

[awawa-dev]

Could you power cycle the ESP32 and your router simultaneously to see if we can trigger that Wi-Fi race condition?

[santievil]

You nailed it.
I cut the power to the ESP32 and rebooted the router at the same time.
I powered it back on while the router was rebooting, and the 'HyperK-Setup' AP appeared for configuration.
Once the router finishes booting up and I have Wi-Fi again, if I reboot the ESP32, it recovers the network configuration and works again (ole).

[awawa-dev]

Could you check how long your router takes to broadcast the Wi-Fi network after being powered on? (You might need a stopwatch and to observe your phone to see when the SSID appears). Then I'll increase current ~12 seconds to something longer and safer.

[santievil]

I don't know if mine is a good example.
I have an ASUS ADSL-AC68U with Entware, Diversion, and other services... it takes about 120 to 150 seconds until it is usable (I've been looking at the logs).

[awawa-dev]

I think wled has also around 12 sec. in default AP init mode: https://github.com/wled/WLED/blob/0f30c54ca6d621feed1cb2ccedd06d527d36f5d4/wled00/wled.cpp#L943

[santievil]

Yes, but if no one connects to the AP within 18 seconds, it calls initconnection again; if someone connects but does nothing, every 5 minutes.

https://github.com/wled/WLED/blob/0f30c54ca6d621feed1cb2ccedd06d527d36f5d4/wled00/wled.cpp#L937

AP only is accesible WLED_AP_TIMEOUT (take a look in comment)
https://github.com/wled/WLED/blob/0f30c54ca6d621feed1cb2ccedd06d527d36f5d4/wled00/wled.cpp#L949

And shut it down when connect to a network
https://github.com/wled/WLED/blob/0f30c54ca6d621feed1cb2ccedd06d527d36f5d4/wled00/wled.cpp#L976

So by those 5 minutes, my router was already running fine and ESP32 connected to network.