[NCLUG] sw raid, recovery after install

[Bob Proulx]

Stephen Warren wrote:
> This is well-known behaviour of Linux MD and was mentioned earlier in
> this thread IIRC. Basically, if the writes are scheduled, then the file
> deleted before the writes are sent to the disk due to write buffers,
> then the outstanding writes are cancelled. They may hit one disk but not
> the other.
> 
> Now, this is going to happen mostly on your /var or related system
> partition. It's much much less likely on a partition that's purely
> storing archive material, since that kind of access pattern doesn't
> usually occur there.

Agreed.

Here is a message from the upstream mdadm author concerning this
issue of mismatches.  And also a casual mention that RAID5 does not
display this symptom, as also mentioned earlier in the discussion.

  please explain mismatch_cnt so I can sleep better at night
  http://bugs.debian.org/cgi-bin/bugreport.cgi?bug=518834#38

Bob