management

[Jon Simola]

On 1/12/06, Matthew D. Fuller <fullermd at over-yonder.net> wrote:

> > What I'm doing now is all my machines have a common NFS mounted /usr
> > and /var/db/pkg so installing a port/package on any one of them
> > means they all have the package installed.
>
> I would tend toward instead using rsync/rdist to manage /usr from a
> central location, and leave it on local disks.  It saves having your
> whole network die when your NFS server goes down, and is also a lot
> faster.

Yes, I would agree completely. This setup was made with some specific
choices in mind, based on the OS (OpenBSD), the running services
(qmail and friends), and the requirement that adding new servers to
the cluster require minimal knowledge (at the expense of other
things).

> Plus, it lets you more easily maintain individual machine
> configs in /usr/local/etc, and handle some things (PostgreSQL comes to
> mind) which write their running data under /usr/local.

Yep. On these machines, I have local filesystems for a backup root
(OpenBSD config is all contained in /etc), /var (DB and logs) and
/queue (mail queue).

> Which is pretty much what it all boils down to; EVERY situation is
> unique in some way, and every person finds a slightly different layout
> works for them.

Yes, and a good admin can tell you why certain choices were made and
is able to debate the issue from all sides, much as choosing between
Linux and Free/Open/NetBSD, or ipfw and PF, or ...

In the few months I've been working on this cluster, I've still
learned things that make me want to go back and redo the entire setup
(the near-zero config I mentioned above in particular). The new design
(running on basically the same hardware) increased our mail processing
ability by at least 2 orders of magnitude, so that I have a single
frontend mail server that's 95% idle instead of a pair of heavily
loaded servers. Makes the zero-config not as high a concern.

Anyways, now that I've rambled on too much, the point is agreed. You
never stop learning, and there's too much knowledge based on
experience that cannot be distilled into a book. And even so, many
times what is printed in books is known to not be best practice (eg,
Cisco's subnet zero).

--
Jon Simola
Systems Administrator
ABC Communications