Spam and extra MX records + cool dual-db setup
Lloyd Kvam
python at venix.com
Mon Apr 21 12:31:22 EDT 2008
On Mon, 2008-04-21 at 10:24 -0400, Neil Joseph Schelly wrote:
> On Saturday 19 April 2008 07:44, Lloyd Kvam wrote:
> > Was there any reason for not using the MySQL replication feature to copy
> > from the writeable database to the read-only database?
> >
> > I rely on replication mostly for off-site backup, but I've read of
> > people splitting databases as you have done for performance reasons.
>
> I imagine a workable solution could be made, but I've only setup replication
> between multiple boxes, not multiple databases on the same box. I'm not even
> sure that's a supported configuration for replication - not sure how you'd do
> it.
I've done it by specifying host and port for the master.
> And any real-time replication would probably face the same locking
> problems. If you just restricted replication to play catchup at regular
> intervals and then you'd still want to disable the Bayesian filtering at
> those times before doing the replication, because it would hang up matters.
The replication stream is handled at a low priority, but I am not going
to argue against your decision.
>
> One problem I did run into was that the binlogs filled up fast. After only a
> few hours of a full load of traffic, the binary logs had filled up several
> gigabytes of space. The MySQL traffic is an overwhelmming percentage of
> INSERT statements:
> http://jenandneil.com/sites/jenandneil.com/files/sf00.dc0.oasis-open.net-mysql_queries-week.png
>
> In my original partitioning, this was especially a problem since those, the
> databases themselves, the logs, the Exim queueing and spooling and tmp space,
> etc were all on one partition for /var. I've broken it up some to get some
> isolation, but it's still just one physical disk. The IO of keeping up a
> binlog with everything else happening would result it more IO overhead than I
> want to spend on replication.
OK. Thanks very much for taking the time to explain. That is very
convincing and enlightening. The IO is even worse than you outlined:
binlog, relaylog, replicated table.
>
> I would venture that the computing effort required to replicate all the
> queries that happen in a few hours time would be far more costly than the
> couple-of-minutes spent re-duplicating the database.
> -N
--
Lloyd Kvam
Venix Corp
DLSLUG/GNHLUG library
http://www.librarything.com/catalog/dlslug
http://www.librarything.com/profile/dlslug
http://www.librarything.com/rsshtml/recent/dlslug
More information about the gnhlug-discuss
mailing list