[CentOS-devel] Disk size on Vagrant image

[Laurentiu Pancescu]

On 25/08/17 17:28, Jeff Sheltren wrote:
> A few questions as I look at
> https://github.com/lpancescu/sig-cloud-instance-build
> 
> 1) I don't see where the 'dd' is happening, could you point to that?

Not implemented yet in CentOS, but reducing the image size is something 
I intend to do as soon as 7.4.1708 is released.  You can take a look at 
Fedora's kickstart (no idea why they don't use zerofree, Fedora images 
only use ext4):

https://pagure.io/fedora-kickstarts/blob/master/f/fedora-cloud-base.ks#_242-246

> 2) Is there a strong reason to use xfs instead of ext4 on the Vagrant VMs?

"Strong" is somewhat subjective, but XFS is the default filesystem in 
CentOS and some people ran out of inodes with our ext4 Vagrant images, 
without running out of space.  I wanted to go back to ext4 (with a small 
bytes-per-inode parameter) and get rid of the swap partition, like 
Fedora Cloud, but there was quite enthusiastic opposition in 
#centos-devel to moving away from XFS.

> 3) It seems you could shrink down the install size by adjusting the
> kickstart. Doing i.e. '%packages --nobase', and also removing '@core' and
> instead adding just the individual desired packages to the list should help
> reduce the size (and the need to remove stuff after install).

I know, but...  Our official images are quite popular and users seem to 
prefer an experience close to what you'd get by installing CentOS 
yourself, so we tend to err on the side of not changing things and not 
straying too far away from users' expectations.  We're still using an 
IDE controller for the VirtualBox images, even though the VirtualBox 
docs recommend using SATA or SCSI for performance reasons, again due to 
vocal opposition to changing the (bad) defaults - mostly from Marcin.  I 
didn't even remove the man pages from the images yet (like Fedora does), 
and I think users would be confused if @core wouldn't be part of the 
images.  We could also reduce the image size considerably if the CentOS 
kernels were also split in packages for VMs and for real hardware, like 
it's been in Fedora for several releases - I'm afraid that's up to Red 
Hat, though.

I started to go through the XFS documentation, to see if it's possible 
to implement something similar to zerofree.  I'm quite tired during the 
week after work and commuting (there are constant traffic jams around 
Stuttgart - think 35km one-way in 60-90 minutes), so progress is really 
slow.

Best regards,
Laurențiu