Alright, so I've ddrescued 2 disks now:
sda - clean
sdb - SMART errors, kicked out of array
sdc - SMART errors, kicked out of array
sde - ddrescued copy of sdb
sdf - original failed disk, replaced.  is now a ddrescued copy of sdc

So I run 

eduardo ~ # mdadm --assemble --force /dev/md127 /dev/sd{f,g,e,a}1
mdadm: cannot open device /dev/sdf1: Device or resource busy
mdadm: /dev/sdf1 has no superblock - assembly aborted
eduardo ~ # 

Any ideas?

On Apr 30, 2013, at 7:48 AM, Gimpbully [off-list ref] wrote:

I have no intentions of recreating at all.  I fully understand the implications (I've had stripe orders go bad on truly massive scales before, I don't ever wanna experience that again), I just don't know mdadm as well as other vendor storage.

I was able to ddrescue a copy of sdb with no errors.  I assembled the array and got a file system metadata dump (this is HUGE) and it's currently rebuilding before I even look at data.  Thank you all for your help, patience was absolutely key here.


On Apr 30, 2013, at 7:45 AM, Phil Turmel [off-list ref] wrote:

quoted

On 04/30/2013 02:20 AM, Sam Bingner wrote:

quoted

On Apr 29, 2013, at 4:33 PM, "Gimpbully" [off-list ref]
wrote:

quoted

On Apr 13, 2013, at 7:20 AM, Sam Bingner [off-list ref] wrote:

quoted

After that you can try to recreate the array with the proper
order (sdc1, sdb1, sde1, missing, sda1) and copy data off or add
the spare in again depending on if you were able to recover all
the data wih GNU ddrescue.

What do you mean recreate?  what's the specific command?  something
like:
mdadm --create --assume-clean --level=5 --raid-devices=5 /dev/md127
/dev/sdc1 /dev/sdb1 /sdv/sde1 missing  /dev/sda1

Don't recreate it - I said the wrong thing... You want to do an
assemble on them with force if possible... Recreate is last ditch and
make sure you have another copy if you do the previous command in
case it doesn't work right due to offsets etc...

try:
mdadm --stop /dev/md127
mdadm --assemble --force /dev/md127 /dev/sd{c,b,e,a}1

Yes.

quoted

If you DO need to recreate it, what you showed looks correct.

NO!

The OP has *not* shared sufficient information on the array members to
say that.  Since it has "worked for years", the odds of an offset error
is *very* high.  Chunk size defaults are also likely to be different.

*Complete* output of "mdadm -E" for the array members is needed before
any "--create" operation is attempted.  Plus the distro info, kernel
version, and mdadm version .

Phil