EXT4 Filesystem Corrupted on v4.4, Is ZFS a Safer Option for v5.1?

[CashewCaliphate]

For the past 4 months I had PVE 4.4 running exclusively on a 250GB m.2 SSD via PCIe adapter (EXT4 filesystem). The drive had less than 4 total TBs written to it, I had 12 Ubuntu containers, and <50% of overall total space utilization as well as <50% of consumed space within each container. Then, I started getting logical I/O errors, PVE locked to read-only, and I had to back-up what I could before throwing in the towel.

I ran diagnostics on the SSD, and gparted showed 0 physical bad blocks, and the firmware is the latest available firmware from Samsung (960 EVO). I searched through Google trying to find out what could have gone wrong, and I found some other folks that mentioned containers and VMs outgrowing their allotted storage, despite PVE reporting otherwise.

Regardless, I'm hoping that fresh installing with ZFS on v5.1 will eliminate this problem, but even then I was reading about some people with arc_prune issues on Proxmox. Is that one of the issues being fixed with the addition of ZFS v0.7.2 on PVE v5.1 ?

This is my approach for re-installing with v5.1; let me know if you think I'm missing something to avoid data corruption in the future:

• I bought a 2nd identical m.2 Samsung 250GB SSD, and am going to install PVE v5.1 using ZFS RAID-1 instead of EXT4. I already have 64GB of ECC RAM, which should be more than plenty for now and the future.
• Additionally, I mounted a spare 2TB HDD for writes, logging, and plex transcoding.
• Finally, I'm going to schedule automatic container/VM backups to a separate FreeNAS machine on my LAN via an NFS share.

 

[LnxBil]

I would not run ZFS on such a cheap SSD right now. ZFS does not support TRIM, yet ext4 does. Normally, it's not ext4's fault if something is corrupt, it's always the underlying system. There are billions even trillions or maybe quadrillons of ext4 installations worldwide without any problems.

Have you checked the ssd with smartctl? If not, please do so.

 

[CashewCaliphate]

I would not run ZFS on such a cheap SSD right now. ZFS does not support TRIM, yet ext4 does. Normally, it's not ext4's fault if something is corrupt, it's always the underlying system. There are billions even trillions or maybe quadrillons of ext4 installations worldwide without any problems.

Have you checked the ssd with smartctl? If not, please do so.

I would also think to blame the OS before the filesystem, but I have no way of seeing what went wrong. There was no evidence from Proxmox that something went awry. smartctl and all other disk diagnostics showed no physical deficiency in the drive itself.

I already ordered and have the new SATA SSDs. I agree that it's not ideal, but without knowing what went wrong, I'd be doing the exact same thing I did last time while expecting different results.

 

[guletz]

Hi,

First I would try to move the SSDs in a different system, and then see if the errors are happening again. If no errors are happening then you will know that your hw server is the problem(memory, cables, psu, ...)

It is true that zfs has no support for trim. But I think that now the trim is not a real problem. I have use for 2 years a SSD for zfs cache (Samsung evo) and it was no problem . Now in your case (zfs mirror ), if you realy need trim, you can do like this(one time at 3 months for example):

- remove one ssd from zfs pool
- format the SSD with ext4 ant then run fstrim
- then re-add your fresh trimmed SSD to the zfs pool

And I can say that zfs with proxmox(5.1) is safe. I have see only a minor problem with grub(easy to solve).