Please don't confuse zfsonlinux with ZFS. zfsonlinux is not feature complete with ZFS and in my book not ready for enterprise usages!
mdadm
- Thread starter Alessandro 123
- Start date
Sure, but PVE is based on zfs on linux......
So, saying that ZFS is much more secure than mdadm just because it has bitrot protection, is false.
Between two scrubs, you can loose data.
This doesn't happen with hw raid or mdadm because both trigger a rewrite in case of unreadable sector
So, saying that ZFS is much more secure than mdadm just because it has bitrot protection, is false.
Between two scrubs, you can loose data.
This doesn't happen with hw raid or mdadm because both trigger a rewrite in case of unreadable sector
We don't use anymore hardware raid because of a lot of firmware issues.
On other installation (no proxmox) we are using mdadm
Because of this discussion we are trying to use zfs with proxmox.
After a hour load test on proxmox V5.4.3 we got this error:
May 12 11:17:24 serv1020 kernel: [419044.195091] INFO: task zvol:2398 blocked for more than 120 seconds.
May 12 11:17:24 serv1020 kernel: [419044.195125] Tainted: P O 4.15.18-12-pve #1
May 12 11:17:24 serv1020 kernel: [419044.195142] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
May 12 11:17:24 serv1020 kernel: [419044.195164] zvol D 0 2398 2 0x80000000
......
And the zfs is dead.
Sorry to say that, but zfs is not ready for production use....
On other installation (no proxmox) we are using mdadm
Because of this discussion we are trying to use zfs with proxmox.
After a hour load test on proxmox V5.4.3 we got this error:
May 12 11:17:24 serv1020 kernel: [419044.195091] INFO: task zvol:2398 blocked for more than 120 seconds.
May 12 11:17:24 serv1020 kernel: [419044.195125] Tainted: P O 4.15.18-12-pve #1
May 12 11:17:24 serv1020 kernel: [419044.195142] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
May 12 11:17:24 serv1020 kernel: [419044.195164] zvol D 0 2398 2 0x80000000
......
And the zfs is dead.
Sorry to say that, but zfs is not ready for production use....
It would be instructive to look at the 5 minutes prior to those messages. its possible (likely) that there is an underlying memory or storage fault.
yes please name some? - if you say cache issues with power failure then your an idiot for setting your drives up that way with mdadm. You should always DIRECTLY pass through (non raid, no cache, simple HBA like a direct connected SAS/SATA drives) your drives for mdam. If you say cpu/memory usage then you can pound sand if your counter recommendation is zfs!
Last edited:
Quick example:
Once a month ( First Sunday of the month) my proxmox server stops responding for a few hours because of a mdadm cron job installed by default with mdadm package starts to check the raid.
I'm running a totally unsupported configuration: 3 proxmox nodes;
* one hosting 3 HDD in raid 5 exposed to the cluster thanks to NFS ( one node is thus nfs client of itself)
* Asynmetric quorum to be able to use my cluster with only my main server powered on
I'm thankful to proxmox to let me go "off-road" with their product on my homelab, but it's also normal for me to be on my own and I do require nor request any support from them.
How to solve this problem need to build a new initrd without a dmraid-driver the following command does this on a system with the "2.6.18-8.1.6.el5"-kernel: