Io_uring + ext4 + data=journal causes filesystem errors

[adamb]

Running into a interesting issue with the following pve-manager/7.1-6/4e61e21c (running kernel: 5.13.19-1-pve).

This is a standard 3 node cluster with HP DL 380 Gen10's and Nimble Hybrid iSCSI Storage.

We use data=journal within our guests as server crashes can be harsh on our database enviroment.

However, I found that when using io_uring on the disk, and mounting it in the guest with data=journal results in filesystem errors and the mount then goes read only.

I can reproduce this consistently, if I mount with data=ordered all is well. Or if I go back to native with data=journal all is well. I have 30 or so clusters and I can reproduce it on all of them.

I know the use case for data=journal is very small so I don't expect many others to have the issue.

Any ideas what it could be, or is it just the nature of the two combined?

 

[mmenaz]

Have you tired the "non subscription" new kernel pve-kernel-5.13.19-2-pve and kvm package pve-qemu-kvm 6.1.0-3? I think they are addressing lot of these issues, judging from other related posts in this forum.

 

[adamb]

Have you tired the "non subscription" new kernel pve-kernel-5.13.19-2-pve and kvm package pve-qemu-kvm 6.1.0-3? I think they are addressing lot of these issues, judging from other related posts in this forum.

Packages from the non subscription repo's seem to have resolved this issue. Appreciate the input.