On Wed, Jul 4, 2018 at 12:33 PM, Fabrice Bacchella <fabrice.bacchella@orange.fr> wrote:

There is a lot of other dependencies to declare other than postfix, MTA
would have been better.

I agree, and suggest to open an RFE on ovirt-host (and elsewhere?
didn't check) to change the Requires:.

ovirt-host have a lot of dependencies:

Indeed

rpm -q --requires ovirt-host
NetworkManager-config-server
cockpit
cockpit-dashboard
cockpit-machines-ovirt
cockpit-networkmanager
cockpit-ovirt-dashboard
dracut-fips
firewalld
glusterfs-rdma
ipa-client
katello-agent
mailx
net-snmp
net-snmp-utils
ovirt-host-dependencies = 4.2.3-1.el7
ovirt-hosted-engine-setup
ovirt-provider-ovn-driver
postfix
python-firewall
rng-tools
rpmlib(CompressedFileNames) <= 3.0.4-1
rpmlib(FileDigests) <= 4.6.0-1
rpmlib(PayloadFilesHavePrefix) <= 4.0-1
screen
sysstat
tcpdump
vdsm-hook-ethtool-options
vdsm-hook-fcoe
vdsm-hook-openstacknet
vdsm-hook-vfio-mdev
vdsm-hook-vhostmd
vdsm-hook-vmfex-dev

Many of thee are useless depending of your setup.

Indeed, but there is obviously a trade-off here.

If we can solve a bug that affects, say, 30% of the setups, by requiring a package that takes 300KB disk space, and has (hopefully) no security/administrative/etc. implications, I'd say it's worth adding.

rng-tools, for example, but my hardware don't provides any of the needed random generators.
screen, tcpdump ? I'm quite surprised.
glusterfs-rdma, I'm not doing glusterfs, and I don't have any hardware to do rdma

I am pretty certain that each was added with a good reason. Sometimes you can find the reason in the git commit, or in the linked bug.

If you disagree about a specific item, after accepting the general reasoning above, feel free to open a bug.

If you disagree with the entire reasoning, it means, practically, that you want more than one ovirt-node, perhaps many more, which will require much more work. Would you volunteer to maintain the various flavors? ovirt-node-gluster-postfix, ovirt-node-ovn-postfix, ovirt-node-gluster-ovn-postfix, ovirt-node-postfix, ovirt-node-sendmail, ...

While I do not follow closely maintenance of ovirt-host, I was involved in deciding we do want it, and so far think it was a good decision. Before that, we had partial lists of requirements duplicated over ovirt-host-deploy, ovirt-hosted-engine-setup, ovirt-node (and then ovirt-release, which new ovirt-node uses), and always forgot to update some of them when adding/updating stuff. Current state, of having everything in a single (rather large, admittedly) ovirt-host meta-package is much better.

Didi