[Systems] DBMS on sunjammer

[Bernie Innocenti]

El mié, 09-12-2009 a las 11:41 +0100, Sascha Silbe escribió:
> I still wonder why they all have to link directly to the native database 
> library instead of using the database abstraction layer almost every 
> language has, but no point ranting about that here...

Yeah, this is just stupid. Even PHP offers a good DBI abstraction, but
everyone uses the mysql_* API directly.

Rails actually does it right, abstracting away (most of) SQL syntax
differences.

</rant>


> > What I really dislike is having to run both on the same machine. At 
> > this time PostgreSQL is not even being backed up!
> The latter is especially bad, we should fix that ASAP.
>   From scanning the documentation, there are three options that would 
> make sense for us:
> - SQL dump (smallest, (mostly) standard format, only full backup)
> - LVM snapshot (only full backup)
> - file-level full backup + write ahead logs (more complex to set up, 
> large, continuous backup)

I'd go for option one. Only very large datasets require the advanced
snapshotting schemes.

Our current MySQL backup script locks one table at a time, which may
lead to inconsistencies in some unlucky situations, but... what the
hell, we're not running a bank :-)


> Depending on how bad it is to loose recent data I'd suggest either the 
> SQL dump (easy to set up, potentially useful for migrations) or the 
> continuous backup (which can be "upgraded" to hot standby servers, i.e. 
> high availability).

I was hoping to setup a slave DB server on beamrider. MySQL supports a
very simple replication scheme using binary logs that are transfered to
the slave and replayed.

The slaves are of course read-only, but can relay write queries to the
master and do automatic takeover.  This has the advantage of being
simple and transparent for client applications.

The disadvantage is that it is quite naive compared to real clustering,
but I don't think we want/need to introduce so much complexity at this
time.


> > We still have a ridiculously small dataset :-)
> In that case running two DBMS shouldn't matter too much anyway. ;)

You're right. I guess it's more of a maintenance issue than a real
resource usage issue.

I swear I'm not being anti-PostgreSQL. In fact, I find the religious DB
wars quite pointless. All RDBMSes suck, and that's it :-)

-- 
   // Bernie Innocenti - http://codewiz.org/
 \X/  Sugar Labs       - http://sugarlabs.org/