Raid5 degraded mode : what happens for real, and when resync will occur ?

Raid5 degraded mode : what happens for real, and when resync will occur ?

[Francois Barre]

Hi all,

Please forgive me for my early-morning wonderings, but I am planning
to make a non-raid -> raid5 migration in a way that could be summed up
like this :
Let sda5 be the originial partition. Let sd[bcd]5 be the additionnal
ones to build up the raid5.
After security backups and sd[bcd]5 partition creation, I would try and do :
Step 0 :
mdadm -l 5 -n 4 /dev/sd[bcd]5 missing
Step 1 :
(start the array as degraded)
Step 2 :
Format /dev/md0, copy sda5 contents to md0, re-create partitions on sda
Step 3 :
mdadm -a /dev/md0 /dev/sda5

At the end of Step 3, we have a working raid5 4-disk array, with the
contents of the previous sda5.
My questions are :
* When will the (real/final) resync occur ? I guess it's in step3,
because during all steps before, there is nothing to resync... Am I
right or wrong ?
* What happens for real during Step1/2 ? Id est for a given stripe,
will the parity block be computed and written as well ? Does a
degraded array work exactly as a normal one for blocks layout ? (I
can't help wondering, because IHMO, this is a terrific algorithmic
issue).

Well, having written down my questions I guess I have the answer, so
if you just can say 'yes little boy, you're right, md is good for
you', I would be pleased.

Re: Raid5 degraded mode : what happens for real, and when resync will occur ?

[Neil Brown]

On Tuesday January 24, francois.barre@gmail.com wrote:
> Hi all,
> 
> Please forgive me for my early-morning wonderings, but I am planning
> to make a non-raid -> raid5 migration in a way that could be summed up
> like this :
> Let sda5 be the originial partition. Let sd[bcd]5 be the additionnal
> ones to build up the raid5.
> After security backups and sd[bcd]5 partition creation, I would try and do :
> Step 0 :
> mdadm -l 5 -n 4 /dev/sd[bcd]5 missing

You'll need "--create" or "-C", and /dev/md0
  mdadm -C /dev/md0 -l 5 -n 4 /dev/sd[bcd]5 missing

> Step 1 :
> (start the array as degraded)

It's already started.

> Step 2 :
> Format /dev/md0, copy sda5 contents to md0, re-create partitions on sda
> Step 3 :
> mdadm -a /dev/md0 /dev/sda5
> 
> At the end of Step 3, we have a working raid5 4-disk array, with the
> contents of the previous sda5.
> My questions are :
> * When will the (real/final) resync occur ? I guess it's in step3,
> because during all steps before, there is nothing to resync... Am I
> right or wrong ?

As the array is created degraded, it will be assumed to be
in-sync. i.e. it will be assumed that the parity is already correct,
and there is no 4th device to prove that wrong.
When you add /dev/sda5, it will 'recover' the missing device onto
sda5.  "recovery" is slightly different to "resync", but for you it
probably isn't an important difference.
So yes, the resync (really recovery) will happen at step 3.


> * What happens for real during Step1/2 ? Id est for a given stripe,
> will the parity block be computed and written as well ? Does a
> degraded array work exactly as a normal one for blocks layout ? (I
> can't help wondering, because IHMO, this is a terrific algorithmic
> issue).

Yes, the parity block will be computed just as normal.  It will be
written if it lives on one of the first 3 devices.

> 
> Well, having written down my questions I guess I have the answer, so
> if you just can say 'yes little boy, you're right, md is good for
> you', I would be pleased.

Yes, you are right, md is good for you.

NeilBrown