Louis Lagendijk writes:
> Running diff on them shows pretty much the only differences you
> would
> expect: NAME, UUID, HWADDR, IPADDR*, PREFIX*, GATEWAY, and DNS* are
> different. In terms of everything else, the config is the samen
what happens if you change this to yes? Does nm-online then wait for
the connection to be up?
Well, I could, perhaps, test that.
However, /usr/share/doc/initscripts/sysconfig.txt claims that the
IPV4_FAILURE_FATAL setting is used only with BOOTPROTO=dhcp, and these
interfaces are all static IP addresses, so I wouldn't expect that any
difference would result from this.
Still, it's possible that NM hijacked something that's documented to be used
only with DHCP, and recycled it for something completely unrelated. It's
possible, I guess. But, I've already screwed around with this for much, much
longer than I really wanted. There's definitely a breakdown somewhere, with
either systemd or networkmanager, when it comes to deliver what network-
online.target promised to deliver. If someone who's more knowledgeable with
these packages can figure out the fail, and a tentative fix is proposed,
I'll be interested in poking at it with a stick, try to test it, and see if
it works.
But it looks to me like this simply cannot be made to work, in the current
state of the world and changing random options in random configuration files
will not be very productive. For now, the most productive path forward is to
use a hacky workaround script that does make it possible to actually start
services that require a working IP address only after an IP address is,
well, actually working.