* Stefan Monnier <jwvfsg4v66f.fsf-monnier+emacs@gnu.org> :
Wrote on Mon, 03 Oct 2022 19:44:29 -0400:
> My Gnus session occasionally gets into a state where it insists that my
> mail server's DNS name isn't found.  I haven't yet managed to get close
> to the source of the problem, in large part because I don't know how to
> reproduce it.  It seems to be associated with network failures (or
> captive portals, which are basically purposefully broken networks).
>
> That a DNS lookup would fail while the network is having problem is of
> course normal, but the problem is that the failure seems to persist "for
> ever", or more specifically until I restart Emacs.  E.g. the old Gnus
> sessions will keep giving me a hostname error while at the very same
> time all other processes on the system (Firefox, `host`, a fresh new
> Emacs, you name it) have no problem finding the IP address of (and even
> connecting to) the very same host.
>
> The last two times this occurred I tried to look a bit more into it, but
> I realized that I don't know enough about how this works to know where
> to look.
>
> The error itself is signaled by `make-network-process` and I suspect the
> error is actually raised by the following piece of code (tho I couldn't
> confirm it because I somehow ended up killing the process while trying to
> attach GDB to it):
>
>       msg = network_lookup_address_info_1 (host, portstring, &hints, &res);
>       if (!EQ (msg, Qt))
> 	error ("%s", SSDATA (msg));
>
> Looking at `network_lookup_address_info_1` it seems it doesn't do much
> more than call `getaddrinfo` (I couldn't copy&paste the actual error
> message but it basically contains my host name plus some error message
> in French, which seems consistent with an error message taken straight
> from `gai_strerror`).
>
> Does anyone here have an idea why `getaddrinfo` might return an outdated
> error, or whether it could come from elsewhere?  Or how/where I might
> find a kind of DNS cache that would be process-local rather
> than systemwide?

* commit:   commit 93bf7d52841c60ffc10e0c9c789a7987812ce55e
|            Author:     Paul Eggert <eggert@cs.ucla.edu>
|            AuthorDate: Mon Feb 29 09:39:45 2016 -0800
|            Commit:     Paul Eggert <eggert@cs.ucla.edu>
|            CommitDate: Mon Feb 29 09:40:58 2016 -0800
|
|  Stop calling res_init
|
|  Emacs shouldn’t need to call res_init any more, now that nscd or
|  equivalent is everywhere.  On modern systems, calling res_init simply
|  slows Emacs down.  On ancient systems lacking nscd Emacs will still
|  work well enough with this change; it’s just that it won’t respond to
|  changes in /etc/resolv.conf


Perhaps the expected results result from changes to resolv.conf, and you
aren't running glibc's nscd (I don't because it generates a lot of bogus
traffic)

The attached patch reverts that commit and reinstates res_init, maybe
you could try it to see if it helps