direction criterium for synchronisation raid1

direction criterium for synchronisation raid1

[Christina Braun]

Hello,
which is in raid1 the criterion for the direction of synchronisation? In 
case of  disk-replacing
or removing one mirror-leg (working actual on the other leg ) and add 
the same removed leg later, which has the old data now, again to the mirror?

How can I tell the system which mirrored partition is  now the 
data-source without
making the raid1 new or zero the superblock?  Is the  destination in 
every case the device
in mdadm manage after the add ? Can I see the source or destination by a 
info like mdstat or superblock?
 
I cannot find the  answers of this simple questions. It works by 
testing, but I feel better about my production-system and a later 
replace  if I know it. The found examples describes only the replace of  
ascending order second drives.

Thank You,
Christina

Re: direction criterium for synchronisation raid1

[Mario 'BitKoenig' Holbe]

Christina Braun <braun@mail.informatik.uni-essen.de> wrote:
> which is in raid1 the criterion for the direction of synchronisation? In 

The event counter :)

> How can I tell the system which mirrored partition is  now the 
> data-source without
> making the raid1 new or zero the superblock?  Is the  destination in 

Usually you don't need to do this. md manages a per-mirror event counter
which always gets increased when relevant events occur like assembling
or stopping an array, adding or removing mirrors etc.
Due to this, whenever you remove a mirror off an raid1, the event
counter of the remaining mirrors gets increased. When you shut down your
machine, plug a disk off and turn the machine on again, once the raid
gets assembled, the event counter of the remaining mirrors gets
increased.

Thus, as long as you access those devices through md only (and don't
mount the device of one of the mirrors as plain filesystem, for
example), the remaining mirrors will always be newer than removed ones
and thus md knows the sync-direction (when they are equally "old", they
are in sync per definition).

There are some exceptions to this:
1. When you like to use the older mirror as source of synchronization,
you have to take care and better zero the superblock of the newer one
before (make sure your raid device did not get assembled, probably based
on the wrong mirror).
2. When you plug a foreign mirror into the system which - however -
refers to the same raid-device (especially having the same UUID etc.)
as your own mirrors but has a bigger event counter, you have to take
care a lot :) This should usually not happen accidentially, as long as
you avoid to assign UUIDs to new raids manually.

> every case the device
> in mdadm manage after the add ? Can I see the source or destination by a 
> info like mdstat or superblock?

Have a look at mdadm -E. This shows you the superblocks of single
mirrors and within them their respective event counters.


regards
   Mario
-- 
A promise to a woman is just a lie that hasn't happened yet.
                                               -- Mr. Floppy

Re: direction criterium for synchronisation raid1

[Christina Braun]

Thank you very much for your quick and detailed elaboration! That is my 
searched answer and more.
I have made a short test for quick looking at the event-counters with my 
testpartitions. I  failed and removed the local mirror-leg  and keep the 
remaining active iscsi-leg on the other server (I use raid1 with iscsi 
on 2 servers),  copied  some  files to the md, added then the removed 
local partition again.  I watched the counters in all phases  and can 
see the differences and after syncing the same counters. The criterion 
event-counter  removes also my little doubt about using iscsi-raid1.  
Also I am very glad for your advice in case of mounting md-devices  as 
plain-filesystems and remaining older data in relationship of this 
counter. I will now pay great attention to this event-counter :-)

Best regards,
Christina


Mario 'BitKoenig' Holbe schrieb:
> Christina Braun <braun@mail.informatik.uni-essen.de> wrote:
>   
>> which is in raid1 the criterion for the direction of synchronisation? In 
>>     
>
> The event counter :)
>
>   
>> How can I tell the system which mirrored partition is  now the 
>> data-source without
>> making the raid1 new or zero the superblock?  Is the  destination in 
>>     
>
> Usually you don't need to do this. md manages a per-mirror event counter
> which always gets increased when relevant events occur like assembling
> or stopping an array, adding or removing mirrors etc.
> Due to this, whenever you remove a mirror off an raid1, the event
> counter of the remaining mirrors gets increased. When you shut down your
> machine, plug a disk off and turn the machine on again, once the raid
> gets assembled, the event counter of the remaining mirrors gets
> increased.
>
> Thus, as long as you access those devices through md only (and don't
> mount the device of one of the mirrors as plain filesystem, for
> example), the remaining mirrors will always be newer than removed ones
> and thus md knows the sync-direction (when they are equally "old", they
> are in sync per definition).
>
> There are some exceptions to this:
> 1. When you like to use the older mirror as source of synchronization,
> you have to take care and better zero the superblock of the newer one
> before (make sure your raid device did not get assembled, probably based
> on the wrong mirror).
> 2. When you plug a foreign mirror into the system which - however -
> refers to the same raid-device (especially having the same UUID etc.)
> as your own mirrors but has a bigger event counter, you have to take
> care a lot :) This should usually not happen accidentially, as long as
> you avoid to assign UUIDs to new raids manually.
>
>   
>> every case the device
>> in mdadm manage after the add ? Can I see the source or destination by a 
>> info like mdstat or superblock?
>>     
>
> Have a look at mdadm -E. This shows you the superblocks of single
> mirrors and within them their respective event counters.
>
>
> regards
>    Mario
>   


-- 
**********************************************   /\_/\
* email: braun@dc.uni-due.de                    / 0 0 \
* Christina Braun                              ====v====
* FB. 5, ICB / Informatik                       \  W  /	
* University DE                                 |    |         
* Schuetzenbahn 70  tel.: + 49    201 183-3929  / ___ \    / 
* D-45127 Essen     fax.: + 49    201 183-2419 / /   \ \  |  
**********************************************(((-----)))-'