On Sat, Mar 27, 2021 at 01:19:01AM +0000, Gerard Ryan wrote:
On Fri, 2021-03-26 at 08:42 -0700, Adam Williamson wrote:
> On Fri, 2021-03-26 at 09:16 +0000, Zbigniew Jędrzejewski-Szmek wrote:
> > Hi,
> >
> > we have been trying to figure out the issue where resolved
> > sometimes
> > does not resolve certain names [1], e.g. 'google.com'.
> > Unfortunately,
> > the issue is only reproducible for some people (most likely it
> > depends
> > on the dns server or other network topology details…).
> >
> > One of the patches that seem problematic [2] was included in F33
> > and
> > then reverted. But it is still present in the systemd main branch.
> > Before tagging the next release and pushing it to F34 and rawhide,
> > we
> > would like to solve this issue (or verify that it does not occur
> > anymore).
> >
> > I prepared a copr build of latest system git [3,4] to make this
> > easy
> > to test.
> >
> > The ask: if you could reproduce the issue before, please test if it
> > still occurs with the copr build. Just "yes"/"no" is
already
> > useful.
I was one of the people who experienced the problem with 246.12-1.fc33
on F33. With that version, I was able to resolve
google.com and
fedoraproject.org just fine, but it seemed I wasn't able to resolve
most other domains that I tried (including
duckduckgo.com and
redhat.com). Just highlighting that as it sounds like other folks had
issues with
google.com but not other domains.
With 248~rc4.git20210326.b240c08-1.fc33 from your copr build on F33, my
wired ethernet (eno1) doesn't work, so it seems I've got other issues
with that one.
That's probably unrelated. This shouldn't happen, so maybe file
a bug. Maybe it's something about systemd-networkd or NetworkManager
not starting (whatever you use)?
With 248~rc4.git20210326.b240c08-1.fc34 from your copr build on F34
on
that same machine (different ostree), I don't appear to have any
networking issues -- I'm able to connect to the network and resolve all
of the domains that I tried.
Let me know if you'd like me to try anything in particular in either of
those scenarios: I've got each one available as a pinned ostree
deployment in silverblue, so it's easy to get back to them.
When sending out the email yesterday morning, I was trying to pin down
the versions where this happens. There has been at least one report
that this also happens with v246-stable and v247-stable, but no clear
confirmation that it happens with v248-rc tip.
Second goal is to find a reproducer and get a backtrace from when the
issue occurs.
Zbyszek