Data corruption detected, but files are not corrupted

[rs225]

There is a possibility that only reads are corrupting, and they are only corrupting under certain conditions. When you write a file, it writes fine. When you read a file, it reads fine. When you scrub, read errors are occurring in large amounts and ZFS claims everything is broken. Then when you go check again, individual files look fine.

 

[Stux]

I'm thinking the hypervisor is buggy or the onboard controllers.

Could you test it by booting bare metal when you can?

 

[MoltenMoose]

or the onboard controllers.

Potentially, although because I have two completely different controllers (onboard SATA and a discrete SATA controller card) and both exhibit the same issue, I find it unlikely.

Since I last updated this thread I've changed all the SATA cables and reinstalled Freenas from scratch in a fresh virtual machine; hasn't helped.

Could you test it by booting bare metal when you can?

What I'm trying now is removing Freenas and virtualisation from the equation entirely. I've imported the zfs pool to the hypervisor directly; it's now mounted natively and I'm not running Freenas at all.

We'll see how I get on with this setup over the next few days.

 

[bleomycin]

Potentially, although because I have two completely different controllers (onboard SATA and a discrete SATA controller card) and both exhibit the same issue, I find it unlikely.

Since I last updated this thread I've changed all the SATA cables and reinstalled Freenas from scratch in a fresh virtual machine; hasn't helped.



What I'm trying now is removing Freenas and virtualisation from the equation entirely. I've imported the zfs pool to the hypervisor directly; it's now mounted natively and I'm not running Freenas at all.

We'll see how I get on with this setup over the next few days.

Did you ever find the problem? I'm having the exact same issue as you, albeit running ZoL on debian jessie. I've had the issue both virtualized under proxmox and running bare metal. I DO have ECC memory, which also memtests fine and there are no ECC errors. I'm running on 2x m1015's flashed to IT mode, I also replaced all the cables AND the supermicro sata backplane. Other than those differences my problem exactly mirrors yours. The system is 100% stable other than this crazy zfs problem. Very strange..

 

[DrKK]

You might consider talking to some of the guys in the #zfs channel on freenode. They are pretty hard core.