src update without console access

[Bill Vermillion]

On Mon, Sep 13, 2004 at 12:25 , while denying his reply is 
spam, Per Engelbrecht prattled on endlessly saying:

> Hi Patrik

> Maybe I didn't make myself clear, if so, sorry. It's not the
> update/install procedure itself that's causing the problem at
> all.

> When running 'installworld',  'mergemaster -p' and 'mergemaster' the
> server needs to be in single-user mode.

Recommended but not needed.

I only manage a handful of servers and I've not had to go to the
colo for any work.  I do make sure someone is there >IF< things go
wrong.

The only time there was a problem was one older mobo that always
had the control keyboard connected and someone decided to stick
that on aother machine.  

> That is not possible for out customers! They often sit on the
> other side of the world on a ssh connection and in single-user
> mode you don't have ssh.That's the problem!

I run the buildworld and buildkernel under nohup so I have a
complete lot.  I start them up, and then log out, and check later
for any errors.

I then install the kernel and hope it comes up.  That part has
never failed for me.

> I use BSD only (Open, Free and a single Net) for our infracture/setup
> (bgp, dns, ntp, mail et al.) in our datacenter. No problem - I have
> consol access all the time for the things I can't do via ssh.

> At first we only offered Tux-boxes. Now we offer fbsd boxes as
> well. Now these customers need to update their servers from any
> location.

The problem is that you have no control over what your customers do
and they could easily modify things and wind up with a non-booting 
kernel.

Perhaps it would be good to set a policy so that customers notify
you when they are going to reinstall and have them do that only
when data center operators are there to copy the old kernel back
to a running one so the customer can sort things out.

When I do the remote updated [almost always after 1AM] I'm down for
about 2 minutes rebooting the new kernel. Then I perform
the installworld, then run mergemaster, then a second reboot.

So down time is about two periods of 2 minutes separated by
a 1/2 hour time frame.  I can get to the facility in about
20 minutes after 1AM so maybe that's why I never have a problem.
The machines must know that I can get there to fix them so they
don't bother breaking :-)

Bill


-- 
Bill Vermillion - bv @ wjv . com