Karanbir Singh wrote:
>>>> A killer on remote machines can be when the new kernel detects your
>>>> NIC cards in a different order and either skips initialization or
>>>> assigns the wrong IP's.
>>> isnt that why you use hwaddr in your network scripts ?
>> That's even worse. All of my remote machines have swappable disks and
>> almost all of them are cloned from a few masters, shipped, and swapped
>> into the destination machine with the IP address set on a temporary box.
>
> This is your site policy, and is not necessary how everyone runs their
> machines. I, for one, take it that each drive that is added to a machine
> is going to be empty - its trivial to remaster a machine on the fly with
> tools like cobbler+koan and use puppet to manage the machine, Capistrano
> to manage app rollout.
I'm not familiar with those tools, but I assume they require a certain
amount of nearby infrastructure which would be extra trouble to
duplicate and maintain in all our remote locations. And it doesn't cost
any more to ship a disk already loaded than empty.
> The concern you raised was about network interfaces not coming up in a
> predictable manner when people move from centos-4 to centos-5, the
> answer to which is, use hwaddr's in your network scripts.
The concern it more general - more about unpredictable differences.
Having seen a difference in what the hwaddr setting does even within a
version, I wouldn't count on it to work across versions without testing
it first.
--
Les Mikesell
lesmikesell at gmail.com