[CentOS] Clustering MySQL

Ryan Ordway rordway at oregonstate.edu
Tue Dec 11 20:31:25 UTC 2007


On Dec 11, 2007, at 11:29 AM, Matt Shields wrote:

> On Dec 11, 2007 1:39 PM, J. Potter <jpotter-centos at codepuppy.com>  
> wrote:
>>
>>> ...  But I saw a presentation at the Boston MySQL Meetup.com
>>> group about how to do master-master in mysql 5.  We're about to
>>> implement this in the next few weeks.  ...
>>
>> I've run into issues with crash recovery in master-master mode:
>>
>>  - master A is at position X
>>  - master B, replicating from A, gets to position X
>>  - master A syncs to its filesystem that it's at position X
>>
>>  - master A receives some inserts, and is now at position Y
>>  - master B, replicating from A, gets to position Y
>>  - master A crashes before the position gets synced to filesystem
>>  - master A gets rebooted, recovers from innodb log, but has itself
>> only marked at position X
>>  - master B requests position Y from master A, but that position
>> doesn't exist yet, so replication breaks.
>>
>> Perhaps someone here knows the proper recovery procedure at this  
>> point?
>
> If this were master-slave, I'd probably do an LVM Snapshot and get a
> fresh copy of the master db.  The same could be done for
> master-master.


The problem is you'll have some inconsistency between your master A's  
view of the database and the master B's view. You lose any changes to  
the data on master B. It would be nice to be able to merge any changes  
from B that hadn't made their way to master A yet. At that point  
you're examining binlogs.

Ryan

--
Ryan Ordway                           E-mail: rordway at oregonstate.edu
Unix Systems Administrator               rordway at library.oregonstate.edu
OSU Libraries, Corvallis, OR 97331    Office: Valley Library #4657












More information about the CentOS mailing list