PXE Boot and Disk for persistent Storage

[Marc Haber]

Hi,

even when I am booting CoreOS via PXE, I might want to have some
persistent storage, at least, for example, for /var/lib/etcd. The docs
give me the impression that this is a rather normal mode of operation,
and so I am a bit astonished that the rest of the docs don't mention
a "best-practice" way to actually configure this.

To me, it would be the nicest method to have a disk-backed filesystem
overlayed over the tmpfs that CoreOS boots into, so that write access
goes to the disk instead of the tmpfs. /run and the usual suspects
would have to be exempt to that. Does CoreOS support this operation
mode?

The second nicest method would be to mount a disk-backed filesytem,
for example, to /mnt/disk, with a mount unit, with appropriate
symlinks to /var/lib/etcd (how do I make those with cloud-config?).

The least nice method that comes to my mind is mounting one
disk-backed filesystem each per directory that should be disk-backed.

What is the recommended system of achieving persistence with
PXE-booted CoreOS?

Greetings
Marc

--
-----------------------------------------------------------------------------
Marc Haber | "I don't trust Computers. They | Mailadresse im Header
Leimen, Germany | lose things." Winona Ryder | Fon: *49 6224 1600402
Nordisch by Nature | How to make an American Quilt | Fax: *49 6224 1600421

[paul...@coreos.com]

Hey Marc,

Good question. Out of curiosity, does this doc achieve what you're looking for?

https://coreos.com/os/docs/latest/booting-with-pxe.html

Cheers,

[Marc Haber]

Hi Paul,

thanks for your answer.

On Tue, Jan 31, 2017 at 08:58:05AM -0800, paul...@coreos.com wrote:
> Good question. Out of curiosity, does this doc achieve what you're looking
> for?
> https://coreos.com/os/docs/latest/booting-with-pxe.html

Only partly. This is actually the documentation that started my
confusion, as it gives so many options without really explaining them,
and does not give advice like "this ist the best practice for this
environment". See, for example, the etc web page that explains three
or four ways of building differently sized etcd clusters (which has
the disadvantage that I fail to find it each and every time I want to
look at it).

The "booting with PXE" web page tells me that I can use rootfstype= to
choose which file sysetm to use in RAM, which does not give
persistency, that I can use root= to use a disk-backed local
filesysem which will be initialied on boot (thus not giving
persistency as well).

The only sentence referring to my issue is "Once booted it is possible
to install Container Linux on a local disk or to just use local
storage for the writable root filesystem while continuing to boot
Container Linux itself via PXE" under installation which gives two
options that might address my issue but doesn't give advice _at_ _all_
about _how_ to do things.

That being said, I feel like being basically on my own which will
probably mean that I'm going to use an operating system that I am more
familiar with instead of exploring the surely interesting concept of a
minimal base installation like CoreOS.