On 7/31/19 12:01 PM, Kaleb Keithley wrote:
>
> https://bugzilla.redhat.com/show_bug.cgi?id=1733602
>
> One of the suggestions there is to "drop the arch." I.e. i686.
>
> If that ends up being the solution that pretty much would force me to drop
> the arch too for glusterfs. (GlusterFS has a bit of plumbing around opening
> ports in the firewall. It might just fail — silently or not so silently.
> It's hard to know, nobody has tested it.
>
> I suspect dropping the arch might cause some amount of heartache in some
> circles.
>
> OTOH, I haven't paid close enough attention to really understand what it
> means to stop building i686 kernels. Does that mean no Fedora distribution
> for i686 hardware? Does it even make sense to keep building glusterfs for
> i686?
I would drop the kernel dependency. It doesn't make sense already in
some contexts (containers) and this is a Fedora package for Fedora
users, so I think anyone who would install it would have a kernel, and
if it's a supported Fedora release it would be larger than 4.18.0.
glusterfs is not the package with the dependency on kernel. Its dependency is on firewalld, which is where the kernel dependency comes from.
The firewalld packager doesn't seem to know how to add an ExcludeArch: to the .spec or how to remove the 'Requires: kernel ...' line.
Maybe someone with some authority on the subject can give some guidance?
Personally I'd be perfectly happy to stop building glusterfs for 32-bit platforms. I honestly don't think anyone runs on 32-bit and the developers aren't really thinking about 32-bit as they work on it. (I had to add a 32-bit build in the CI just to catch all the 32-bit sprintf format mistakes they were making.)
And FWIW, Ceph has abandoned 32-bit platforms starting with Nautilus/14.x.x. I have suggested that GlusterFS officially abandon 32-bit starting with 7.0. Not sure whether that hasn't fallen on deaf ears.
--
Kaleb