Both nvme cache drives always end up corrupted! What can I do to prevent?

[xinil]

Hey all. First time poster, long time Unraid user. A few years ago I set up a streaming/automation server with 3 dockers (plex/jellyfin/telegraf)  and 1 vm (home assistant). This have worked ‘well’ for a while, but for at least the past year I’ve  been fighting a BTRFS cache pool corruption problem. The server works wonderfully….so long as I do not reboot. As soon as I reboot the VMs/Dockers will not start.

 

 My cache shows like this:

 [/dev/nvme1n1p1].write_io_errs    0

[/dev/nvme1n1p1].read_io_errs     0

[/dev/nvme1n1p1].flush_io_errs    0

[/dev/nvme1n1p1].corruption_errs  0

[/dev/nvme1n1p1].generation_errs  0

[/dev/nvme0n1p1].write_io_errs    1408272649

[/dev/nvme0n1p1].read_io_errs     820356362

[/dev/nvme0n1p1].flush_io_errs    40914172

[/dev/nvme0n1p1].corruption_errs  1630001

[/dev/nvme0n1p1].generation_errs  15037

 

So my ‘fix’ for this corruption (that I've done maybe 3 times in as many months) was to remove corrupted nvme0n1p1 from the pool, start the array, add it back to the pool (overwriting corrupted data) and then rebalance with the clean nvme1n1p1. The problem recently has been at some point BOTH nvme drives have started to become corrupt. Performing a scrub with ‘repair corruption’ checked does not fix the issue (even if I do it immediately after errors are detected. I installed a user script to monitor btrfs dev stats /mnt/cache so I’m aware of when the errors appear (usually many a day.)

 

What can I do? I ‘ve read on these forums that it could be memory issues. So I ran a memtest (free version 4 pass, came out 100% pass zero errors.) I’ve read it could be cable issues. But these are m.2 nvme drives, one connected directly to the motherboard and another connected via a PCIe 3.0 x4 adapter card. I’ve tried the latency change in the Linux config as well. Temps do regularly spike around 50-60C, but I figured that’s within normal range? One card is a Pony 1tb and another is Team 1tb.

 

Any advice would be greatly appreciated. I’m on Unraid 6.11.5.

Edited January 27 by xinil

[trurl]

Attach Diagnostics to your NEXT post in this thread 

[xinil]

Shoulda included that in the first post, apologies, here they are.

 

xtower-diagnostics-20230127-1101.zip

[JorgeB]

One of the NVMe devices dropped offline, is it always the same one?

[xinil]

49 minutes ago, JorgeB said:

One of the NVMe devices dropped offline, is it always the same one?

It's not always the same one. You can see in my first post it mentions nvme0n1p1 with the errors (from a past issue I saved), but in the logs it's now nvme1n1p1.

 

It seems to 'focus' on one for a good while (it was months last time before 'both' became corrupted.

Edited January 27 by xinil

[JorgeB]

Try this, it can sometimes help with dropping NVMe devices, on the main GUI page click on the flash drive, scroll down to "Syslinux Configuration", make sure it's set to "menu view" (top right) and add this to your default boot option, after "append initrd=/bzroot"

nvme_core.default_ps_max_latency_us=0 pcie_aspm=off

e.g.:

append initrd=/bzroot nvme_core.default_ps_max_latency_us=0 pcie_aspm=off

Reboot and see if it makes a difference.

 

 

[xinil]

44 minutes ago, JorgeB said:

Try this, it can sometimes help with dropping NVMe devices, on the main GUI page click on the flash drive, scroll down to "Syslinux Configuration", make sure it's set to "menu view" (top right) and add this to your default boot option, after "append initrd=/bzroot"

nvme_core.default_ps_max_latency_us=0 pcie_aspm=off

e.g.:

append initrd=/bzroot nvme_core.default_ps_max_latency_us=0 pcie_aspm=off

Reboot and see if it makes a difference.

 

 

 

 Ok, done, but before I reboot, is there any guidance on what to do with this situation? 

 

[email protected]:~# btrfs dev stats /mnt/cache
[/dev/nvme1n1p1].write_io_errs    47025602
[/dev/nvme1n1p1].read_io_errs     20207911
[/dev/nvme1n1p1].flush_io_errs    462569
[/dev/nvme1n1p1].corruption_errs  6
[/dev/nvme1n1p1].generation_errs  0
[/dev/nvme0n1p1].write_io_errs    0
[/dev/nvme0n1p1].read_io_errs     0
[/dev/nvme0n1p1].flush_io_errs    0
[/dev/nvme0n1p1].corruption_errs  0
[/dev/nvme0n1p1].generation_errs  0

 

Scrub with repair produces:

UUID:             03b3fe98-f7a1-4dc4-b69b-661202e041e1
Scrub started:    Fri Jan 27 13:06:14 2023
Status:           finished
Duration:         0:04:07
Total to scrub:   137.79GiB
Rate:             571.23MiB/s
Error summary:    read=17972852 super=3
  Corrected:      0
  Uncorrectable:  17972852
  Unverified:     0

 

Should I disconnect nvme1 and do what I've always done?: Rebuild with the non-corrupted drive.

Edited January 27 by xinil

[xinil]

I wiped the nvme1 drive, rebalanced with the nvme0 (uncorrupted). Rebooted and within about an hour seeing errors again on nvme1.

[/dev/nvme0n1p1].write_io_errs    0
[/dev/nvme0n1p1].read_io_errs     0
[/dev/nvme0n1p1].flush_io_errs    0
[/dev/nvme0n1p1].corruption_errs  0
[/dev/nvme0n1p1].generation_errs  0
[/dev/nvme1n1p1].write_io_errs    368768
[/dev/nvme1n1p1].read_io_errs     26856
[/dev/nvme1n1p1].flush_io_errs    3918
[/dev/nvme1n1p1].corruption_errs  0
[/dev/nvme1n1p1].generation_errs  0

 

This is what I have in the Linux config

default menu.c32
menu title Lime Technology, Inc.
prompt 0
timeout 50
label Unraid OS
  menu default
  kernel /bzimage
  append initrd=/bzroot nvme_core.default_ps_max_latency_us=0 pcie_aspm=off
label Unraid OS GUI Mode
  kernel /bzimage
  append initrd=/bzroot,/bzroot-gui nvme_core.default_ps_max_latency_us=0 pcie_aspm=off

 

[JorgeB]

7 hours ago, xinil said:

Rebooted and within about an hour seeing errors again on nvme1.

If the kernel boot options didn't help suggest using different NVMe devices.

[xinil]

Just wanted to chime back in on progress info. I've replaced one NVME SSD (the most recent one with errors) and switched to a different adapter card for the other NVME SSD. It's been 48 hours and not a single error has occurred yet. Crossing fingers it was just hardware that needed swapping! If I don't reply again, assume everything is fixed!