On 30/06/2006 5:31 p.m., Stewart Smith wrote:
> On Mon, 2006-06-26 at 12:55 -0700, Yong Lee wrote:
>> We currently have a 2 node cluster database (2 hosts with a ndbd and sql
>> node on each) setup and I want to do some realtime backups of the database
>> by using the replication feature. I was wondering if there was a way to
>> capture the cluster updates on a single slave, eg:
> Why not use the consistent online backup already in Cluster?
> ndb_mgm -e 'start backup'
> is the thing that does it.
The problem with this is there's no way to use it in conjunction with
binary logs for "always current" or point in time backup and recovery.
Recovering from an ndb backup only gets you back to whatever point in
time the backup was taken.
It would be great if MySQL Cluster had some functionality for
centralised binary logging (e.g. on the management node). Then one could
do a cluster backup (which would record the current binary log position)
and then recover from there by re-applying queries using the binlogs. I
don't know if that's technically feasible...
At the moment we have to force all update queries to go through one
"master" server (we need this anyway because we can't use NDB for all
our tables), which has binary logging and replicated slaves which do
daily backups with mysqldump. I don't bother with backups from the
ndb_mgm any more, because I can't relate them to any position in the
binlog so I can't actually use them for recovery.