35network-manager won't setup network if other modules don't add any "initqueue/finished" hook

[ryncsn]

Previously when using 35network-legacy, it will add initqueue/finished hooks to wait for the interface to be configured. However 35network-manager won't do so, and if other modules didn't add any initqueue/finished hook, initqueue won't run, and the nm-run.sh won't have a chance to run either.

Is 35network-manager expecting other module to install initqueue/finished to wait for network to get ready, or perhaps 35network-manager should at lease ensure nm-run.sh got executed for at least once?

[ryncsn]

There is a workaround for this problem in kdump, by install another initqueue hook to keep the initqueue running until network is ready:

See: https://lists.fedorahosted.org/archives/list/kexec@lists.fedoraproject.org/thread/ZAYFCE3UOH7UTCBFWZ6KGR5VI625IGZT/

[cgwalters]

xref coreos/fedora-coreos-config#259
and in particular https://mail.gnome.org/archives/networkmanager-list/2020-January/msg00000.html
where I argue that we should be running NM as a regular systemd unit.

[haraldh]

@lkundrak !

[LorbusChris]

Any updates on this?

[jomeier]

@cgwalters
Hi. Any news on this one? Greetz, Josef

[jomeier]

@haraldh
Hi Harald. @LorbusChris mentioned that you could come to the rescue here.

Could you give priority to this issue, please? The OKD community needs you 🕺🏻

[LorbusChris]

I think I meant @lkundrak might be able to do that.
Anyway, now we've pinged 'em all! 👅
And as @jomeier said, we over at Fedora CoreOS and OKD would appreciate a fix for this issue =)

[LorbusChris]

https://bugzilla.redhat.com/show_bug.cgi?id=1809641 is a related BZ for this

[tyll]

Hi, The NetworkManager team is happy to help. What is the exact request?

@thom311 or @bengal please take a look.

https://bugzilla.redhat.com/show_bug.cgi?id=1809641 is a related BZ for this

This is not reported against NetworkManager and also does not include the NM team in CC. Should it?

[LorbusChris]

Thanks @tyll! The root cause for that BZ is this issue here :)
We need this to be solved to unblock coreos/fedora-coreos-config#259 (and subsequently a few other things)

[bengal]

@LorbusChris I'm looking into this.

[jlebon]

To be clear, https://bugzilla.redhat.com/show_bug.cgi?id=1809641 isn't an issue with NM in the initrd, it's with the current legacy network dracut module FCOS is still using. We have a hunch that switching to NM will resolve this too, but it's not sure.

IOW, we should focus on getting the original problem filed here fixed so we can unblock FCOS moving to NM in the initrd (coreos/fedora-coreos-config#259). And that may fix that RHBZ at the same time. If not, that's a separate issue to fix.