[BUG]USB hard drive IO error occurs when using python3-picamera2 with rpi cam

[ss8688]

I encounter random usb hard disk IO errors whenever I use the mipi rpi camera and spyglass(https://github.com/mainsail-crew/spyglass) whitch dependents on python3-picamera2.Turning off webrtc(h264 encode) can alleviate some of the issues.When H264 encoding is not turned off, very few read and write operations cause error. For example, when running dmesg, this error only occurs when a large number of reads and writes occur after H264 encoding is turned off.The error completely disappear after unplugging the camera or turning off the spyglass.I have ruled out the power supply issue because my CM4 expansion board does not have a 1.2A limit and I have a 5v10A psu

[ss8688]

also see raspberrypi/linux#7274
mainsail-crew/spyglass#137

[davidplowman]

Hi, I'm afraid my first reaction is still to look at power supply issues, the whole situation feels so typical of that kind of problem. Just because your supply can deliver 10A (or whatever), doesn't mean that that it will correctly negotiate and reliably provide the power that is required. Do you have alternative power supplies (bench power supplies, or official Raspberry Pi ones) that you can test with? I also wonder if running your USB hard drive through a powered USB hub might help? Do you have any alternative USB drives to try? I expect some may draw more current than others.

[ss8688]

I have tested several different psu and also tested cables with separate power supply, but none of them have solved the problem. In my testing, the power supply has been consistently outputting a voltage of 5.02-5.03V without experiencing any voltage drop due to overload. The maximum recorded current is 1.31A, which is much lower than the maximum current.And this issue occurred during an update or after I switched from ayufan's camera-streamer to python3-picamera2 based spyglass, I did not make any hardware changes

[ss8688]

I haven't really tested any other USB drives, but I have tested replacing the ssd inside. I have replaced the Intel ngff SSD with the dramless Hynix BC711, which should consume less energy, but it hasn't solved the problem

[davidplowman]

This sort of thing is pretty tricky. I never known a problem that sounds like this not be power related, but I suppose there's always a first time. I'd definitely try a powered USB hub, and see if that makes any difference. Are you able to try without a USB drive at all? It feels like getting back to something stable would give you a place to start.

You could also try over-voltage-ing the Pi (over_voltage=x in config.txt, I think don't go above 6, maybe start at x=2), or down-clocking the board (e.g. arm_freq=1000).

[davidplowman]

One more thought, after "bad things" have happened, can you try running vcgencmd get_throttled. What does that report?

[ss8688]

vcgencmd get_throttled
throttled=0x0

And I have a software that can record low voltage and high temperature warnings that have occurred, but there is no record

[davidplowman]

So it seems like there's no evidence the Pi itself is having any problems, but that doesn't tell us whether the USB drive is getting enough power. Are you able to try running that off a powered hub?

[ss8688]

I have previously tested using a USB cable with a separate power supply to power the hard drive, but the issue has not been resolved

[davidplowman]

I'm not sure what else to suggest here. The problems do seem to me to be USB drive problems. If you can confirm that everything works fine when not writing to a USB drive, then I think you might have to try some completely different drives and adapters to discover if the problem is general or not. At that point, it might be worth asking in one of the more general support forums - I'm sorry that USB drives are not anything I really know about.

[ss8688]

The USB drive is definitely not a problem because it can work properly with spyglass python3-picamera2 off

[ss8688]

This is an uas issue that can be resolved by turning off uas, but I don't know why I didn't encounter this problem when not using picamera2 and mipi rpi-cameras.This uas issue only occurs when using picamera2 and mipi rpi-cameras

[davidplowman]

Thanks for the update. I'm afraid I'm just as baffled by this as you are, but I can try and ask around to see if there is anyone around who has any expertise here. Is it possible to create a relatively simple example, ideally not involving too many 3rd party packages, that reproduces the problem easily? As always, if things can be reproduced without too much difficulty, there's more chance of someone having time to take a look. Thanks.