MDADM, Raid 5 and shutting down (getting mdadm errors before power off)

#1

Hi,

I have a newbie question. I have been getting errors on reboot with my software raid 5 (i use 4x2 TB drives). I created it with MDADM, it takes 6 to 8 hours to set up the 4th drive with parity.

When I shutdown, i see errors that drives have failed (just before shutdown) from mdadm. Then after reboot, my raid 5 is in auto/read-only and says it needs to rebuild again.

I’ve been googling around but I have not been successful in finding a solution so far. Is there a safe way to shutdown the raid 5 I guess, before the machines shuts down or reboots?

Any tips, links or info would be very helpful. Thanks everyone!

#2

What’s cool is that i did
sudo mdadm --readwrite /dev/md0

and it went right back into active. However, how do I prevent the problem from occuring in the first place? I have not yet mounted the raid.

#3

The problem that you are seeing is probably due to the gpios that enable the two usb/sata chips
Your problem is that the sata hat service drops them as it is shut down.

This is done in /usr/bin/rockpi-sata/misc.py in the routine “disk_turn_off”. The two GPIOs, 26 and25 are set low, which disables the chips and drops them out of the system before a sync could be completed on the filesystem.

If you comment out those two set_mode (xx, GPIO.LOW) statements the driver chips will remain enabled until the system drops power. It should have completed all file operations well before that.

#4

Please confirm which sata hat you are using?

Pcie or usb?

#5

I am using USB I believe (this is the quad sata hat for Raspberry PI which has the USB adapter that needs to be plugged in).

Thank you @oket I will definitely try that!

#6

I think the issue is solved now. Still sometimes at boot, only 2 drives light up. I have to reboot at least one more time or shutdown and restart for it to detect all 4 again. Once detected, the RAID automatically goes back to active. Not sure why that is. I have also followed all the steps to get the hdd serial numbers, not from the hat.

#7

Rather than stop the power-down. I think a better way would be to unmount the software raid first and then power it down. I have it automounting on boot (following a novaspirittech youtube video). I just need to figure out how to unmount it before it executes that shutdown script. Any ideas?

#8

Just add “mdadm --stop /dev/md0” in script

#9

Thanks! The shutoff script seems to be a python script if I’m not mistaken. I’m not that familiar with python, but I see we can execute normal commands. I will try something like this:

from subprocess import call
call(['echo', 'mdadm --stop /dev/md0'])
#10

Sure, go for it