On Tue, 15 May 2012, Dan Williams wrote:
> On Mon, May 14, 2012 at 5:04 PM, Mark Hills <[email protected]> wrote:
> > On Fri, 11 May 2012, Dan Williams wrote:
> >
> >> This is a hybrid proposal for supporting bcache as a md device.
> >> Somewhat similar to the v1.x metadata format, where array assembly is
> >> handled in userspace, but managed in the kernel. In the bcache case it
> >> is an "external" metadata format, but then the expectation is that the
> >> kernel "bcache" personality takes over runtime maintenance of the
> >> metadata.
> >
> > I am having some trouble with this, can you clarify (perhaps by example)
> > how to create a pairing of a cache and backing device?
[...]
> > # make-bcache -C /dev/sdb
> > # make-bcache -B /dev/sdc
> > # mdadm -A /dev/md0 /dev/sdb /dev/sdc
> > mdadm: Cannot assemble mbr metadata on /dev/sdb
>
> I should have been more explicit in the changelog. This current patch
> was only tested to assemble an existing bcache configuration. I.e. it
> assumes the backing device has been attached to the cache set at least
> once. By default make-bcache always creates a new cache-set id per
> invocation. So in the above example it won't find sdb and sdc belong
> to the same md device because the cache-set id's differ. You can
> verify this with mdadm -E.
Thanks for explaining. I suppose this is just expecially awkward at the
moment because attaching the devices can't be done without a different
kernel.
So it will be good to see the "create" support, when it is ready.
> [..]
> >
> > Maybe I missed something basic here, but I'm afraid can't see what? If
> > not, hopefully this information is useful.
> >
> > I use bcache v13 patches and the MD conversion, and this patch to mdadm.
>
> For reference:
>
> [root@fedora-virt ~]# mdadm -E /dev/vd[bc]
> /dev/vdb:
> Magic : <bcache>
> Version : 3
> Role : cache
> Set UUID : e8670bc3:974b489c:aeb68ea1:d1adf6d1
> Cache Devs : 1
> Device UUID : a1b25e26:5a4c4db6:2716c1aa:eba44ec7
> Flags : sync
> Policy : lru
> Label :
> Device Size : 16776192 (8.00 GiB 8.59 GB)
> Bucket Size : 1024
> Num Buckets : 16384
> this dev : 0
> First Bucket : 1
> Checksum : 45fa97cda53b0b6 correct
> /dev/vdc:
> Magic : <bcache>
> Version : 1
> Role : backing-device
> Set UUID : e8670bc3:974b489c:aeb68ea1:d1adf6d1
> Cache Devs : 1
> Device UUID : 7045ff87:aa4b13c4:670d8bb2:0b4ad628
> Flags : sync
> State : clean
> Mode : writeback
> Label :
> Device Size : 16776192 (8.00 GiB 8.59 GB)
> Bucket Size : 1024
> Num Buckets : 16384
> this dev : 0
> First Bucket : 1
> Checksum : f36910e51451d91c correct
Presumably there's a little more too this, as I can't even investigate the
UIDs in use:
# make-bcache -C /dev/sdb
[...]
# make-bcache -B /dev/sdc
[...]
# mdadm -E /dev/sd[bc]
/dev/sdb:
MBR Magic : aa55
/dev/sdc:
MBR Magic : aa55
I confirmed that I'm using the patched mdadm, but it still doesn't seem to
recognise the magic.
But, of course, I understand this is all work in progress. So this isn't a
complaint at all, I am just looking forward to code in this area and happy
to test where I can.
Thanks
--
Mark
