implications of partitioning and raid

implications of partitioning and raid

[Peter W. Morreale]

I'm wondering what the implications are for having multiple raid sets on
a partitioned disk wrt to disk failures. 

For example, suppose I create two partitions on a set of disks and
create raid sets on those partitions.  Further, not all raid sets
reference the same disks.  IOW, md0 references disks 1 and 2, and md1
references disk 1 and 3.  (overly simplistic for discussion purposes)

Assume a portion of disk 1 goes 'bad' (localized within one of those
partitions), is noticed by md and a rebuild is warranted.  

What is the behavior?  

Will both raid sets start a rebuild?  Or only the affected raid set?

IOW, would there be two rebuild tasks, one for each raid set?  Or a
single rebuild that encompasses all raid sets (within the same raid
level, of course) on the disk in question? 

What I am getting at is whether there would be any advantage to
partitioning disks for failure purposes.   

Thanks,
-PWM

Re: implications of partitioning and raid

[Robin Hill]

[-- Attachment #1: Type: text/plain, Size: 2275 bytes --]

On Thu Jan 05, 2012 at 08:40:34AM -0700, Peter W. Morreale wrote:

> 
> I'm wondering what the implications are for having multiple raid sets on
> a partitioned disk wrt to disk failures. 
> 
> For example, suppose I create two partitions on a set of disks and
> create raid sets on those partitions.  Further, not all raid sets
> reference the same disks.  IOW, md0 references disks 1 and 2, and md1
> references disk 1 and 3.  (overly simplistic for discussion purposes)
> 
> Assume a portion of disk 1 goes 'bad' (localized within one of those
> partitions), is noticed by md and a rebuild is warranted.  
> 
> What is the behavior?  
> 
Any failure will only affect the single array member where the error
occurred. In most cases this is due to a catastrophic disk failure
though, so will also occur on other partitions of the disk (and the
arrays using them) as well.

> Will both raid sets start a rebuild?  Or only the affected raid set?
> 
> IOW, would there be two rebuild tasks, one for each raid set?  Or a
> single rebuild that encompasses all raid sets (within the same raid
> level, of course) on the disk in question? 
> 
Rebuild processes are per raid set, but only one rebuild will be done at
a time using the same disk (so if disk 1 fails and is replaced, then
there's separate rebuilds pending for md0 and md1, but only one will run
at a time).

> What I am getting at is whether there would be any advantage to
> partitioning disks for failure purposes.   
> 
There's an advantage in being able to prioritise the rebuilds (so I can
rebuild the array containing / before that containing /opt for example)
and there's some cases where there's a transient failure (a write error
causes the array member to be failed, but testing shows up no errors) in
which case it reduces the rebuild time as only a single array needs to
be rebuilt. The latter is pretty rare though, unless you have other
underlying problems (such as power issues or poor quality I/O
chipsets/drivers).

HTH,
    Robin
-- 
     ___        
    ( ' }     |       Robin Hill        <robin@robinhill.me.uk> |
   / / )      | Little Jim says ....                            |
  // !!       |      "He fallen in de water !!"                 |

[-- Attachment #2: Type: application/pgp-signature, Size: 198 bytes --]

Re: implications of partitioning and raid

[Stan Hoeppner]

On 1/5/2012 9:40 AM, Peter W. Morreale wrote:

> Assume a portion of disk 1 goes 'bad' (localized within one of those
> partitions), is noticed by md and a rebuild is warranted.

How often does only a "portion" of a disk go bad these days?  Typically
this would mean unrecoverable read/write errors for a set of LBA
sectors.  Physically, with the majority of today's disks, these
perceived sector defects are actually the result of head actuator and/or
spindle bearing wear beyond tolerances, not magnetic defects in the
platters.

Thus, any such defective "portions" are typically doing to "grow" fairly
rapidly.  Which means, in your scenario, both arrays are going to need
new spare partitions and a rebuild in short order, as an entire disk is
failing, not just a portion thereof.

RAID - Redundant Array of Inexpensive Disks
RAIP - Redundant Array of Inexpensive Partitions

Aside from sounding like a sexual crime, there are many other obvious
reasons why the latter has never been coined nor considered a storage
standard by anyone.

There are valid reasons for partition based arrays, such as booting from
disks with wonky offset requirements (e.g. advanced format drives).
IMHO the scenario you've presented here is not a valid case for
partition based arrays.

-- 
Stan