countme: Force disable Count Me logic in DNF

[travier]

Make sure that we do not use the internal Count Me logic in DNF in
rpm-ostree as we have our own external implementation that is aware of
the different behavior regarding repo handling.

See also the discussions in:

• RFE: Add an option/interface to disable Count Me globally rpm-software-management/libdnf#1174
• Allow access to metalink/countme fields for external DNF Count Me implementations rpm-software-management/libdnf#1068
• docs: Update instructions to disable countme #2671

[travier]

Needs proper testing but as I'm wondering where I should factorize setting this option, I opened a PR to discuss that.

[cgwalters]

I think we can likely just change this in the core in rpmostree_context_setup() right?

[jlebon]

Cool! I think with this we can drop the sed instructions in the countme docs now too, right?

[jlebon]

Minor/optional: should this be in rpmostree_context_new_system instead? This is where we disable other global-ish librpm/libdnf bits we know we never want upfront.

[travier]

Updated. This still needs testing and I'm wondering how I'll to do that 🤔

[jlebon]

Updated. This still needs testing and I'm wondering how I'll to do that thinking

Hmm probably not worth trying to get a CI test for this. For manual testing, it looks like libdnf already logs debug output when it sets LRO_ONETIMEFLAG, so you should be able to use that: https://github.com/rpm-software-management/libdnf/blob/9a0e17562b19586b3ffa70fa93eb961b558794c7/libdnf/repo/Repo.cpp#L1132.

Edit: You can see those log messages with G_MESSAGES_DEBUG:

[root@cosa-devsh ~]# cat /etc/systemd/system/rpm-ostreed.service.d/override.conf
[Service]
Environment=G_MESSAGES_DEBUG=all
[root@cosa-devsh ~]# rpm-ostree install ltrace
...
[root@cosa-devsh ~]# journalctl --grep countme
-- Logs begin at Wed 2021-04-14 15:06:42 UTC, end at Wed 2021-04-14 15:13:12 UTC. --
Apr 14 15:08:32 cosa-devsh rpm-ostree[1189]: countme: no event for updates: budget to spend: 2
Apr 14 15:08:38 cosa-devsh rpm-ostree[1189]: countme: no event for fedora: budget to spend: 1

(Obviously with this PR, you shouldn't see any countme messages in the journal output.)

[cgwalters]

Fine by me to merge as is if you've done the manual testing.

jlebon's suggestion is one approach; another may be for us to test via a custom webserver. But IMO not a blocker to merge this.

[travier]

Did the testing and this one is working as expected. Had countme mentions before, none after:

-- Logs begin at Fri 2021-04-16 12:13:55 UTC, end at Fri 2021-04-16 17:38:31 UTC. --
Apr 16 12:15:42 cosa-devsh rpm-ostree[1383]: countme: no event for updates: budget to spend: 3
Apr 16 12:15:45 cosa-devsh rpm-ostree[1383]: countme: no event for fedora: budget to spend: 1
Apr 16 13:15:18 cosa-devsh rpm-ostree[1751]: countme: no event for updates: budget to spend: 2
Apr 16 13:15:20 cosa-devsh rpm-ostree[1751]: countme: event triggered for fedora: bucket 1
Apr 16 13:15:20 cosa-devsh rpm-ostree[1751]: Downloading: https://mirrors.fedoraproject.org/metalink?repo=fedora-33&arch=x86_64&countme=1
Apr 16 13:15:20 cosa-devsh rpm-ostree[1751]: Transfer finished: https://mirrors.fedoraproject.org/metalink?repo=fedora-33&arch=x86_64 (Effective url: https://>
Apr 16 13:24:50 cosa-devsh rpm-ostree[2306]: countme: no event for updates: budget to spend: 1
Apr 16 13:24:52 cosa-devsh rpm-ostree[2306]: countme: no event for fedora: window already counted
-- Reboot --
-- Reboot --

Used a combination of:

rpm-ostree cleanup -m
rpm-ostree refresh-md
rpm-ostree install foo ...