[rescue] So, I'm bringing up my E3000....

Skeezics Boondoggle skeezics at q7.com
Wed Apr 7 21:47:39 CDT 2004


On Wed, 7 Apr 2004, Brian Dunbar wrote:

> Yah, I've thought some and the situation applied to a specific 
> instance, specific machine, a few jobs ago.  My production boxes boot 
> to a mirrored disk, so I dunno why that folklore was jammed in there.  
> I've excised it.

Sorry to chime in late, but I've always used mirrored root disks (with
DiskSuite, not Veritas) for a couple of reasons:

1.  If one half fails, pull it and boot the other half - DiskSuite doesn't
muck with the partition labels or require "encapsulation" and either
surviving half of the mirror is bootable.

2.  Software RAID 5 can be pretty slow... and where do you intend to put 
your swap slices?  Interleaved across each drive in the RAID set, or in 
the filesystem instead of on the raw device?  Seems messy.

3.  Sun now actually recommends doing it the way I've been doing it for
years before they published their Blueprints(tm) series of "best
practices"...

4.  Makes upgrades totally painless.

Before the "Live Upgrade" and "WebStart Flash" stuff was introduced, we
figured out that a mirrored root volume allowed for quick and painless
Solaris upgrades with minimal downtime:

1.  Detach the secondary submirror.

2.  Jumpstart the 2nd drive with the machine up, and configure it, patch 
it, load apps, etc.  Don't configure mirroring on the 2nd drive yet.

3.  Reboot the box off the newly-installed submirror.

4.  If everything works and you're happy, redefine the mirrors and <some
handwaving here over specifics> resync with the original drive.  Ta-da!  
An OS upgrade with less than 2 minutes of downtime (depending on how fast
your machine can POST and reboot. :-)

5.  If something breaks or isn't right, simply reboot off the primary boot
disk and be back in production while you figure out what went wrong and
try again.

(We even used a similar method under SunOS 4, upgrading a bunch of
machines from 4.1.3 to 4.1.4 in no time.)

I may be old fashioned, but I hate that you can't find appropriately-
sized boot drives anymore.  A pair of 9gig drives is plenty in most
situations - I still set up a mirrored / & /var, interleave swap - and all
user data on a separate set of drives, or a NetApp. :-)

Anyway, that's my $0.02,

-- Chris



More information about the rescue mailing list