Switch from SATA-RAID to gmirror?

Ashley Moran work at ashleymoran.me.uk
Thu Apr 6 20:29:10 UTC 2006


On Apr 06, 2006, at 5:35 pm, Duane Whitty wrote:

> Hi Ashley,
>
> I'm glad things worked well for you.  Faith got you this far
> but how long do you want to depend upon it?
>
> A long time ago I was tasked with the administration of some
> HP-UX boxes running on K-series hardware.  I didn't setup the
> hardware and I didn't do the system install but I was expected,
> as the systems consultant, to give reasonable assurances that in
> the case of system failure the recovery procedures would work.
> As it turns out I had to also write those procedures.  After I did
> so I insisted that a failure be simulated and that it be determined
> whether or not we could recover our operation starting from
> scratch with just our backups and system tapes.  After all, there is
> no one easier to fire than a consultant and it's always the  
> consultant's
> fault :)
>
> So my recommendation is that you simulate a disk going bad now
> before it happens for real.  For instance, what happens if you unplug
> the disk from the controller, or remove its power connection, etc?
>
> Just my $0.02


Duane,

Your $0.02 is probably worth a lot more than that...  I'm not in a  
hurry to put things to the test but I will eventually.  Fortunately,  
we've just bought redundant servers for everything (apart from a  
Win2k3 server running SQL Server, which cost us more in licensing  
than hardware, and which we are unfortunately stuck with for the  
foreseeable future).  This server is one of them - so even if the  
whole array fails, we will have another machine to fall back on.  But  
when it's settled down, I'll pull the plug on the primary drive and  
see if it will reboot.  We have two more servers on the way destined  
to run Postgres.  We've bought them with Areca RAID 6 cards, and I  
will definitely enjoy pulling two of the drives just to see what it  
does.

Our new policy is redundant EVERYTHING in the live environment.   
Mainly this is not for the reduced protection from failure, but for  
the freedom to take servers offline for upgrades or testing.   
Currently we're in a situation where a guy's whole business depends  
on a single-disk webserver running Postgres (because it was the only  
BSD machine we had at the time), which desperately needs upgrading  
for performance tuning, but which we just can't do.

Ashley


More information about the freebsd-questions mailing list