top | item 36589408

(no title)

Running in production for about 3 years with Ubuntu 20.04 / zfs 0.8.3. ZFS is being used as the datastore for a cluster of LXD/LXC instances over multiple physical hosts. I have the OS setup on its own dedicated drive and ZFS striped/cloned over 4 NVMe drives.

No gotchas / issues, works well, easy to setup.

I am looking forward to the Direct IO speed improvements for NVMe drives with https://github.com/openzfs/zfs/pull/10018

edit: one thing I forgot to mention is, when creating your pool make sure to import your drives by ID (zpool import -d /dev/disk/by-id/ <poolname>) instead of name in case name assignments change somehow [1]

[1] https://superuser.com/questions/1732532/zfs-disk-drive-lette...

discuss

throw0101c|2 years ago

> I am looking forward to the Direct IO speed improvements for NVMe drives with https://github.com/openzfs/zfs/pull/10018

See also "Scaling ZFS for NVMe" by Allan Jude at EuroBSDcon 2022:

* https://www.youtube.com/watch?v=v8sl8gj9UnA

noaheverett|2 years ago

Sweet, I appreciate the link!

doctor_eval|2 years ago

Sorry if this is a newbie question - but what are you using to share the ZFS file system between physical hosts?

I’ve been out of this particular game for a long time.

noaheverett|2 years ago

In our case each node's ZFS store operates independently, it's not shared. We spin up "VMs" (LXC based) on each node and they talk to each via our private network. Hope that helps!