Gmirror question - what am I doing wrong?

[Karl Denninger]

Ok, so I have Gmirror working for two exactly-equal sized disks, either on
the raw device or on the slice.

Here's the issue.

I have two vendors of a particuar size disk.  They have slightly different
actual sizes, and detect as slightly different in sysinstall and FDISK.
The displayed geometries differ only in cylinder counts, BUT the partition
function in either sysinstall or FDISK (whether I use the command line or
the "tool") will not permit an EXACT size match for the slices between the
two vendors.

Now here's the problem...

Let's posit D1 and D2, where D1 is from vendor 1 and D2 is from vendor 2.

There is also a "D1'" and "D2'", which are "spares" from each vendor.

I wish to be able to do the following:

1. Run on the mirrored system normally.

2. Attach any of the remaining two disks (D1' or D2') as a third mirror
   component, allow it to synchronize, detach it, and put it in a safe
   offsite.  This provides me with a simple offsite backup, although I 
   understand that I risk data inconsistency unless I quiesce the system
   prior to detaching the backup volume.

3. If either D1 or D2 fail, I need to be able to replace the failed disk
   with EITHER D1' or D2', and have the rebuild succeed.

Now here's what I see as a problem.

I have built the system on the SMALLER capacity disk (D1), using the
maximum available size for the first (and only) slice on that disk.  

When I attempt to set up the larger capacity disk with the exact same 
slice size in blocks as the smaller disk has, the system refuses, and 
rounds either up or down (presumably due to an internal difference in how 
it sees geometry - although the FDISK display only differs in the cylinder 
count!)

Gmirror refuses to attach a slice which has FEWER blocks than the online
volume with a cryptic ("Bad address") error message.  

This makes sense; it can't actually mirror it.

It will, however, attach a disk slice which has MORE blocks quite happily.

Now here's the problem.

Let's say that I have the system sync'd with the larger and smaller block
count disks online.  The SMALLER one fails.  I replace it, but the other 
larger disk is not available, because its in the vault.  I thus replace the
disk with the SMALLER one, but I can't attach the slice on the replacement 
disk, because that disk has a smaller capacity than the only remaining
active mirror component, and I can't make a slice bigger than the number 
of actual blocks that exist on the drive!

I'm screwed, and have no way out of this other than to ditch the entire
scheme, manually recreate the smaller volume, copy all the data over, and
restart gmirror from scratch.  

There goes my online mirroring - the very intent of the mirroring has been 
lost!

It appears that you have to somehow force FDISK to create a slice with 
the EXACT same block count to resolve this.  

This, however, appears to be flatly impossible to coerce FDISK to do, as 
whatever is preventing this from being done is not "exposed" and visible.

There has to be a way around this problem - its not credible to insist that
someone buy the exact same model and make of disks all the time, is it?

So... how do you work around this - or can you?

--
-- 
Karl Denninger (karl at denninger.net) Internet Consultant & Kids Rights Activist
http://www.denninger.net	My home on the net - links to everything I do!
http://scubaforum.org		Your UNCENSORED place to talk about DIVING!
http://www.spamcuda.net		SPAM FREE mailboxes - FREE FOR A LIMITED TIME!
http://genesis3.blogspot.com	Musings Of A Sentient Mind