[baylisa] Re: wtf: hostid gives '0' as a hostid

Sun Feb 17 16:05:10 PST 2008

I must say, this is the first time I can recall an applicable LISA topic
discussion in quite a while. :)

What's funny is how nothing's really changed, of course.  Same sort of
problem we were handling when we started being SAs.  Different tools, same
limiting issues.  Anyway, back to the topic:

On Sun, Feb 17, 2008 at 03:10:03PM -0800, Jeff Younker wrote:
> Depending on your hardware vendor it may be possible to get a real
> unique ID.  Dells have a software accessible serial number.  The
> magic incantation is something like:
> 
>     dmidecode | grep Serial\ Number | head -n1.
> 
> We used this at the last company I worked for.  Other vendors may have
> similar systems.

Everyone does, the issue is that no one's consistent.  If you're lucky enough
to manage an environment with a sole exclusive h/w provider, you're in good
shape.  Otherwise, you've got different length IDs, and some vendors don't
bother to set a serial, or set them all to the same (number of zeroes or
whatever).  Dell's great for that, other h/w suppliers that may already
exist in many environments, not so much.

Does, however, work great for new systems, and provide a useful reminder about
planning ahead and establishing sane requirements.  If you're fortunate enough
to be arriving early enough in a company's life to choose the first vendor and
establish those practices.  In my experience, most SAs fall to pressure to get
things done immediately and the trap of "we'll automate it later," "we'll get
the specs and install system up once we ship this first batch of machines," etc.

While I do like to try to get it "right the first time" when I set things up
for new companies/new environments, my bread and butter has always been
to come into companies that have realized they've built themselves into a
corner of integration limitations and pick the pieces up.  Most don't have
budget to refresh all the heterogenous systems to bring to consistency, so an
artificial consistency is more likely.  I suspect most SAs see the same thing
as there are more SAs than there are new companies, afaik.  Or maybe there
are a lot of new companies that keep hiring the same fools to mess things up
to start ;) 

> There are advantages to using an infrastructure specific identifier.    
> The
> switch port that a machine lives on would be a good one.  This way
> location determines function, and you can replace a system simply by
> unracking the old box, racking the new box, and turning it on.

easier said than done IME.  Figuring out which port a machine is on relies
on a lot of chained trusts between network gear, often of different vintages
and different authentication options.

Knowing what VLAN is a lot easier, and often works well enough, but again
only with a planned environment (essentially, a dedicated VLAN per type of
host/installation), though it does lend itself well to transition:  as hosts
are made compliant, bring them onto their VLAN.  legacy installations get
left in the swap VLAN, with a clear agreement in place that these machines are
"best effort" support and will never be held to any SLA standards, and
more importantly, will *always* be lower priority than the structured
VLAN system, both new setup and maintenance.  (yes, you still fix them when
production goes wack.  You just get to make a big deal out of how your team
made up for the lack of support and negotiate some immediate form of reward
for them, and announce how this shortsighted need has delayed ALL your teams
deliverables by however long it took to fix the problem and follow-through)