Using the *real* sector/block size of a mass storage device for ZFS

[Mark Felder]

On Fri, Feb 7, 2014, at 16:19, Warren Block wrote:
> On Fri, 7 Feb 2014, Mark Felder wrote:
> 
> > On Fri, Feb 7, 2014, at 14:44, Dustin Wenz wrote:
> >> We have been upgrading systems from FreeBSD 9.2 to 10.0-RELEASE, and I'm
> >> noticing that all of my zpools now show this status: "One or more devices
> >> are configured to use a non-native block size. Expect reduced
> >> performance." Specifically, each disk reports: "block size: 512B
> >> configured, 4096B native".
> >>
> >> I've checked these disks with diskinfo and smartctl, and they report a
> >> sector size of 512B. I understand that modern disks often use larger
> >> sectors due to addressing limits, but I'm unsure how ZFS can disagree
> >> with these other tools.
> >>
> >> In any case, it looks like I will need to rebuild every zpool. There are
> >> many thousands of disks involved and the process will take months (if not
> >> years). How can I be I sure that this is done correctly this time? Will
> >> ZFS automatically choose the correct block size, assuming that it's
> >> really capable of this?
> >>
> >> In the meantime, how can I turn off that warning message on all of my
> >> disks? "zpool status -x" is almost worthless due to the extreme number of
> >> errors reported.
> >>
> >
> > ZFS is doing the right thing by telling you that you should expect
> > degraded performance. The best way to fix this is to use the gnop method
> > when you build your zpools:
> >
> > gnop create -S 4096 /dev/da0
> > gnop create -S 4096 /dev/da1
> > zpool create data mirror /dev/da0.nop /dev/da1.nop
> >
> > Next reboot or import of the zpool will use the regular device names
> > with the correct ashift for 4K drives.
> 
> But remember that this does not fix alignment, and if the partitions are 
> not aligned with 4K blocks, at least write performance will suffer.
> 

I often use raw devices ever since ZFS has had the ability to tolerate
slight differences in disk sizes.