Re: evbarm build using exorbitant disk space during build

[David Brownlee <abs%absd.org@localhost>]

On 30 April 2014 10:18, Martin Husemann <martin%duskware.de@localhost> wrote:
> Houston, we have a problem:
>
> the evbarm build uses up enormous amounts of temporary disk space:
>
> $DESTDIR is 602 MB
> $OBJDIR is 11.4GB
> $RELEASEDIR is 2.6GB
>
> Most of the space in $OBJDIR is used for kernel builds (> 6GB) and also for
> the cdimage.
>
> Now why is this a problem? Most nodes on the build cluster have a 16GB
> limited tmpfs space for this, so the evbarm builds only suceed, if they
> are run on one of the other nodes. Unfortunatley exactly those nodes
> are scheduled for replacement, and the new, faster ones will have the
> 16GB limit as well.
>
> There is an admin ticket and options are being discussed, but the easy
> solutions at the admin level will slow down all our other builds - or
> would be just stopgaps untill space runs out again.
>
> So, a solution at the build system level might be better (but I don't see
> any easy one).
>
> Or we drop all the INSTALL kernels and implement some hacks to load the
> install file system as a module.
>
> Personally I have (I think) never used an install kernel on some evbarm
> device, but instead used NFS root and manual install or similar schemes.
>
> Any smart ideas?

Possibly dumb idea - How about a tool to plug a ramdisk into a kernel
- maybe even with the ramdisk having a magic number start and end and
a magic number at the end and allow just concatenating the ramdisk
onto a kernel?