https://bugzilla.redhat.com/show_bug.cgi?id=963952
The gist of this bug is, if the user installs Fedora and doesn't have (wired) ethernet connected at the time of installation, onboot=no and therefore on any subsequent boot, the network does not come up. A separate GNOME bug is that even if the user manually turns on that wired connection, it isn't a sticky setting. So networking is off again the next reboot. The trick is to find the totally obscure and buried option for the wired device to connect automatically.
This bug doesn't piss me off the way it did back in 2013 when I filed it, because I know the work around. But this is still a Top 10 Most Asinine Bug. Amusingly, wireless connections don't have this behavior, even though they're in the realm of 8000% more risky even when making a manual+conscious connection. Whereas when I take a physical plug and stick it into the jack, it's a considerable WTF moment when that connection doesn't work until I intervene further, and doesn't work yet again on a reboot until I intervene yet again.
I can kinda sorta almost understand server folks differing on this (even though I find it specious for various reasons) but Workstation really shouldn't behave this way. It is not an easy work around, at all, when the user doesn't don't know about it. And in no sane world should they need to know about it. And even when they do know about it, it's always a *sigh* "oh yes, that" moment. So from every perspective I see it as a negative and no benefit.
Thanks,
On Thu, 2015-05-28 at 09:37 -0600, Chris Murphy wrote:
https://bugzilla.redhat.com/show_bug.cgi?id=963952
The gist of this bug is, if the user installs Fedora and doesn't have (wired) ethernet connected at the time of installation, onboot=no and therefore on any subsequent boot, the network does not come up. A separate GNOME bug is that even if the user manually turns on that wired connection, it isn't a sticky setting. So networking is off again the next reboot. The trick is to find the totally obscure and buried option for the wired device to connect automatically.
This bug doesn't piss me off the way it did back in 2013 when I filed it, because I know the work around. But this is still a Top 10 Most Asinine Bug. Amusingly, wireless connections don't have this behavior, even though they're in the realm of 8000% more risky even when making
WiFi doesn't have this behavior because it's a 100% conscious decision to click on the WiFi you want to connect to. Wired (by default on Workstation/Desktop where NetworkManager-config-server is *not* installed) automatically connects without asking you when it gets a carrier.
a manual+conscious connection. Whereas when I take a physical plug and stick it into the jack, it's a considerable WTF moment when that connection doesn't work until I intervene further, and doesn't work yet again on a reboot until I intervene yet again.
I can kinda sorta almost understand server folks differing on this (even though I find it specious for various reasons) but Workstation really shouldn't behave this way. It is not an easy work around, at all, when the user doesn't don't know about it. And in no sane world should they need to know about it. And even when they do know about it, it's always a *sigh* "oh yes, that" moment. So from every perspective I see it as a negative and no benefit.
On the anaconda side, perhaps the best course of action is to just not have Dracut write ifcfg files for wired network interfaces at all when networking is not configured, but rely on having NetworkManager-config-server installed where people want the "no network by default" behavior. That means that Workstation would have the autoconnect wired behavior (because it shouldn't be installing NM-config-server) but other variants wouldn't.
NM-config-server adds some config snippets that suppress the autoconnect DHCP thing that NM does.
Dan
desktop@lists.stg.fedoraproject.org