Gmirror on a partition of a slice

[n j]

Pawel wrote:
> > 6) fdisk -BI /dev/mirror/gm0 to create a slice in gm0
> Don't do this.
> > 7) bsdlabel -w -B /dev/mirror/gm0s1 to create root partition
> And this.
> newfs /dev/mirror/gm0. Copy the data from da0s1a to /dev/mirror/gm0.
> Modify /etc/fstab on da0s1a to mount root as /dev/mirror/gm0.

Just wanted to say thanks. I don't know if the solution should have
been obvious to me; perhaps it is to someone with more geom
experience. For me, the two "skip this" instructions were an
enlightenment. Anyway, here is what I learned, maybe the archives will
help someone in the future. Put simply:

a) if you are using whole disk as consumers, you can partition (fdisk)
and label (bsdlabel) the resulting gmirror provider,

b) if you are using slices as consumers, you should skip the
partitioning and only label the partitions of the resulting gmirror
provider,

c) finally, if you are using partitions as consumers, you must skip
both the partition and label steps and directly do a newfs on the
resulting gmirror provider.

The logic behind this is probably that if you already partition or
label the underlying consumers, you must not do the same with the
provided mirror, i.e. you can't really treat the provided mirror as an
entirely fresh new disk. As I said, this is maybe obvious to someone,
but for me it was new.

For the record, here's my current setup using two 17G disks. I hope
that mounting swap on a gstripe partition is not a Bad Thing.

# df -h
Filesystem            Size    Used   Avail Capacity  Mounted on
/dev/mirror/gm0       989M     36M    874M     4%    /
devfs                 1.0K    1.0K      0B   100%    /dev
/dev/stripe/st0s1a     29G    718M     26G     3%    /big

# swapinfo -h
Device          1K-blocks     Used    Avail Capacity
/dev/stripe/st0s1b   2097152       0B     2.0G     0%

# gmirror status
      Name    Status  Components
mirror/gm0  COMPLETE  da0s1a
                      da1s1a

# gstripe status
      Name  Status  Components
stripe/st0      UP  da0s1d
                    da1s1d

Once again, Pawel, thanks for help (and for the geom system, of course).

Regards,
-- 
Nino