On Mon, Jul 17, 2017 at 04:02:14PM +0200, Ludovic Courtès wrote:
> The log shows that building the disk-image derivation starts with:
> 
> --8<---------------cut here---------------start------------->8---
> creating raw image of 102400.00 MiB...
> Formatting '/gnu/store/yv5r65584aaml86hc0xrgyffnp70ri36-disk-image', fmt=raw size=107374182400
> --8<---------------cut here---------------end--------------->8---
> 
> That’s a lot, no?

Yes, I specified it manually while debugging.

> Regardless, memory consumption in the VM is not supposed to be
> proportional to the size of the image being created.

Indeed, the problem exists also when I don't pick a size manually.

> The allocation failure happens while copying files:

[...]

> Does that work on previous master with Linux-libre 4.12.0 (current
> master is at 4.12.2)?  (This would allow us to determine if this is an
> ext4 bug, who knows…)

Yes, I tested with a variety of kernels and on the master branch as
well.

> If it does, then the only other issue I can think of is if Guile itself,
> while running ‘copy-recursively’ from (guix build utils), eats memory
> proportional to the number of files, leading to an OOM condition.
> However the kernel message don’t report it as an OOM, AFAICS.

I'm wondering if it's related to the Guile 2.0 / 2.2 mismatch in the
initrd, discussed previously:

https://lists.gnu.org/archive/html/guix-devel/2017-07/msg00220.html

Guile 2.0 would need to rebuild all the 2.2 modules it finds.