CentOS - May 2016 - Inconsistancy in RAID documentation

On the wiki page for creating a mirrored root drive, at
<https://wiki.centos.org/HowTos/Install_On_Partitionable_RAID1>, it first
tells you, in the install, to chose manual partitioning, and to leave at
least 1M at the end of the drive.... and then it tells you to create the
RAID 1 using /dev/sda and /dev/sdb - the *whole* raw drives.

         mark, currently creating a RAID 1 using just the raw drives, no
partitions

On Tue, May 3, 2016 at 12:49 PM,  <m.roth at 5-cent.us>
wrote:
> On the wiki page for creating a mirrored root drive, at
> <https://wiki.centos.org/HowTos/Install_On_Partitionable_RAID1>, it
first
Way back in 2012, I had some problems with partition-able RAID1, when
one of the disks is missing.
<https://lists.centos.org/pipermail/centos/2012-June/126927.html>

A client's ERP system would not boot because one of the disks had gone bad.

A search "partitionable raid 1 site:lists.centos.org" also throws up
other discussions on partition-able RAID1"

Indeed the problem may have been fixed but I have not done any
partition-able RAID1 since 2012.

-- Arun Khan

On 2016-05-03 15:49, m.roth at 5-cent.us wrote:
> On the wiki page for creating a mirrored root drive, at
> <https://wiki.centos.org/HowTos/Install_On_Partitionable_RAID1>, it
first
> tells you, in the install, to chose manual partitioning, and to leave at
> least 1M at the end of the drive.... and then it tells you to create the
> RAID 1 using /dev/sda and /dev/sdb - the *whole* raw drives.
> 
>          mark, currently creating a RAID 1 using just the raw drives, no
> partitions
Hi Mark,

The advice to subtract a certain amount from the drive size
is still sound. I subtract ~ 1GB (~ US$0.20), here's why.

A typical vendor disclaimer for drive capacity: "some of the
listed capacity is used for formatting and other functions, and
thus will not be available for data storage". "Other functions"
includes dynamic things like an inventory of spare sectors to
remap sectors that develop unrecoverable errors, replacing them
from a pre-existing inventory of spares. After a failure, a
new, replacement drive may be short of the capacity needed to
mirror all the sectors of the surviving RAID pair. AFAIK the
Linux RAID driver won't try to reason about the need to mirror
the final few blocks in the element, it just refuses to create a
mirror from a mismatched pair. Trimming some from the "actual"
size when creating a brand new mirror gives you some room to
finagle a replacement.

Probably should mention that in the last few years (?) the
position of the md superblock was repositioned from 128K from
the logical last block of the drive to 8K to 12K from the end of
the drive, see man 4 md.

HTH, HAND,
-- 
Charles Polisher