Re: [PATCH RFC 0/6] spec: Decompose the daemon subpackage

On Wed, Nov 23, 2022 at 04:11:49PM -0700, Jim Fehlig wrote: With this refactoring the two big questions - What is the desired end state - Can we ensure a clean upgrade path Let me ignore everything except the QEMU driver and the nodedev driver, for sake of illustration. Today we've got a few install approaches recommended - Ask for 'libvirt', which gives you - libvirt-daemon - libvirt-daemon-driver-qemu - libvirt-daemon-driver-nodedev All the libvirt pieces, but not the hypervisor itself - Ask for 'libvirt-daemon-kvm', which also gives you - libvirt-daemon - libvirt-daemon-driver-qemu - libvirt-daemon-driver-nodedev - qemu-kvm All the recommended libvirt pieces for QEMU, including the hypervisor itself - Ask for 'libvirt-daemon-driver-qemu', 'libvirt-daemon-driver-nodedev', which also gives you - libvirt-daemon The bare minimum libvirt pieces, but not the hypervisor itself In terms of package contents - libvirt-daemon-driver-qemu /usr/lib64/libvirt/connection-driver/libvirt_driver_qemu.so /usr/sbin/virtqemud - libvirt-daemon /usr/sbin/libvirtd /usr/sbin/virtlogd /usr/sbin/virtlockd /usr/sbin/virtproxyd Our goal is the ability to install only the module daemon, without libvirtd. This implies that we need to cut the dependency from libvirt-daemon-driver-qemu, to libvirt-daemon. I didn't do that originally because it would break apps that installed libvirt-daemon-driver-qemu, expecting to get libvirtd. We also needed it for the virtlogd/virtproxyd dep. That decision, however, was made when libvirtd was still the default active daemon. Since we changed the systemd presets in Fedora/RHEL to activate modular daemons, we don't need the deps to pull in libvirtd by default. We do not want to force virlockd as that's an opt-in feature, and also don't want to force virtproxyd as that's only needed for back compat with old remote libvirt clients, or for exposing over TCP. virtlogd is more difficult, it is possible to turn that off and if people do that they might prefer not to have it installed. Or they may want to use it, but have it installed in a different container from virtqemud. So on balance I thin we need to sever the dependancy there too. So I'd say we need to have - libvirt-daemon-lock /usr/sbin/virtlockd - libvirt-daemon-log /usr/sbin/virtlogd - libvirt-daemon-proxy /usr/sbin/virtproxyd Probably a 'libvirt-daemon-plugin-lockd' for the lock-driver/lockd.so as that really ought to be an optional component too. A 'libvirt-daemon-client' package for 'virt-admin' The libvirt-daemon package would need to hold /usr/sbin/libvirtd and would also need dependencies for the upgrade path on libvirt-daemon-lock libvirt-daemon-log libvirt-daemon-proxy libvirt-daemon-pugin-lockd libvirt-daemon-client I'm not sure if we'll still need a 'libvirt-daemon-core' too as your patches suggest, for some of the common files, or if they all end up in one of the other sub-RPMs above. The libvirt-daemon-driver-XXX packages would then have *no* dependency on anything. If you are asking for libvirt-daemon-driver-XXXX packages of any kind, you're responsible for pickin the exact set of pieces you want to have present. Installing 'libvirt-daemon' will give you libvirtd, virtproxyd, virtlockd etc, and you can ask for libvirt-daemon-driver-XXX on top. The 'libvirt-daemon-kvm' would give you the sensible set of pieces, *not* included libvirt-daemon any more though on distro versions which have switched to module daemons. The only thing we can't achieve this way is to install libvirtd and QEMU, without having the module daemons present. I'm not sure that matters though, if we aim to discontinue shipping libvirtd long term. With regards, Daniel -- |: https://berrange.com -o- https://www.flickr.com/photos/dberrange :| |: https://libvirt.org -o- https://fstop138.berrange.com :| |: https://entangle-photo.org -o- https://www.instagram.com/dberrange :|