[pve-devel] [RFC PATCH 1/2] frr: add networking.service as systemd dependency

[pve-devel] [RFC PATCH 1/2] frr: add networking.service as systemd dependency

[Gabriel Goller]

Add networking.service to the 'After' dependency directive. Guarantees that
the frr.service will start after the networking.service is done.

We had some issues with data races between FRR and ifupdown [0], mostly
around the dummy interface. At startup, FRR and by extension fabricd is
up faster than the dummy interface which means that the interface won't
be configured properly. This does not happen with 10.3.1 anymore due to
this patch [1], which fixes the problem in the fabricd daemon.

This is a cleaner solution which will hopefully fix similar issues in
the future.

[0]: https://lore.proxmox.com/pve-devel/575dc158-c628-4343-ab8e-c3ea75da1507@proxmox.com/
[1]: https://github.com/FRRouting/frr/pull/17083

Suggested-by: Stefan Hanreich <s.hanreich@proxmox.com>
Signed-off-by: Gabriel Goller <g.goller@proxmox.com>
---

Sending this as an RFC as this could have some implications that I'm not
yet aware of. Feedback is welcome! This won't be merged upstream as FRR
doesn't require ifupdown2.

 ...add-dependancy-to-networking.service.patch | 51 +++++++++++++++++++
 debian/patches/series                         |  1 +
 2 files changed, 52 insertions(+)
 create mode 100644 debian/patches/pve/0007-systemd-add-dependancy-to-networking.service.patch

diff --git a/debian/patches/pve/0007-systemd-add-dependancy-to-networking.service.patch b/debian/patches/pve/0007-systemd-add-dependancy-to-networking.service.patch
new file mode 100644
index 000000000000..e547b0374caa
--- /dev/null
+++ b/debian/patches/pve/0007-systemd-add-dependancy-to-networking.service.patch
@@ -0,0 +1,51 @@
+From 570265b39158cd7f72e69116ddeedd25fd91db6b Mon Sep 17 00:00:00 2001
+From: Gabriel Goller <g.goller@proxmox.com>
+Date: Thu, 26 Jun 2025 13:42:43 +0200
+Subject: [PATCH] systemd: add dependancy to networking.service
+
+Add networking.service to the 'After' dependency directive. Guarantees that
+the frr.service will start after the networking.service is done.
+
+We had some issues with data races between FRR and ifupdown, mostly around the
+dummy interface. At startup, FRR and by extension fabricd is up faster than the
+dummy interface, which means that the interface won't be configured properly.
+This does not happen with 10.3.1 anymore due to this patch [0], which fixes the
+problem in the fabricd daemon.
+
+This is a cleaner solution which will also fix errors appearing in the
+journal.
+
+[0]: https://github.com/FRRouting/frr/pull/17083
+
+Signed-off-by: Gabriel Goller <g.goller@proxmox.com>
+---
+ tools/frr.service.in  | 2 +-
+ tools/frr@.service.in | 2 +-
+ 2 files changed, 2 insertions(+), 2 deletions(-)
+
+Index: b/tools/frr.service.in
+===================================================================
+--- a/tools/frr.service.in	2025-06-26 13:42:43.835424384 +0100
++++ b/tools/frr.service.in	2025-06-26 13:42:43.833424382 +0100
+@@ -2,7 +2,7 @@
+ Description=FRRouting
+ Documentation=https://frrouting.readthedocs.io/en/latest/setup.html
+ Wants=network.target
+-After=network-pre.target systemd-sysctl.service
++After=network-pre.target systemd-sysctl.service networking.service
+ Before=network.target
+ OnFailure=heartbeat-failed@%n
+ 
+Index: b/tools/frr@.service.in
+===================================================================
+--- a/tools/frr@.service.in	2025-06-26 13:42:43.835424384 +0100
++++ b/tools/frr@.service.in	2025-06-26 13:42:43.833424382 +0100
+@@ -2,7 +2,7 @@
+ Description=FRRouting
+ Documentation=https://frrouting.readthedocs.io/en/latest/setup.html
+ Wants=network.target
+-After=network-pre.target systemd-sysctl.service
++After=network-pre.target systemd-sysctl.service networking.service
+ Before=network.target
+ OnFailure=heartbeat-failed@%n
+ 
diff --git a/debian/patches/series b/debian/patches/series
index c59ef31dfe6c..dd5b97af3086 100644
--- a/debian/patches/series
+++ b/debian/patches/series
@@ -4,3 +4,4 @@ pve/0003-tests-add-bgp-evpn-autort-test.patch
 pve/0004-zebra-add-ZEBRA_IF_DUMMY-flag-for-dummy-interfaces.patch
 pve/0005-fabricd-add-option-to-treat-dummy-interfaces-as-loop.patch
 pve/0006-fabricd-enable-dummy_as_loopback-option-per-default.patch
+pve/0007-systemd-add-dependancy-to-networking.service.patch
-- 
2.39.5



_______________________________________________
pve-devel mailing list
pve-devel@lists.proxmox.com
https://lists.proxmox.com/cgi-bin/mailman/listinfo/pve-devel

[pve-devel] [RFC PATCH 2/2] debian: bump version

[Gabriel Goller]

Bump to 10.3.1-1+pve2. Add patch that adds networking.service to
frr.service dependencies.

Signed-off-by: Gabriel Goller <g.goller@proxmox.com>
---
 debian/changelog | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/debian/changelog b/debian/changelog
index 2f19f309424a..ec46e0a7c221 100644
--- a/debian/changelog
+++ b/debian/changelog
@@ -1,3 +1,10 @@
+frr (10.3.1-1+pve2) UNRELEASED; urgency=medium
+
+  * add dependency on networking.service for frr.service to fix race
+    conditions where frr fails to configure interfaces that are not yet up.
+
+ -- Proxmox Support Team <support@proxmox.com>  Thu, 26 Jun 2025 14:45:23 +0200
+
 frr (10.3.1-1+pve1) trixie; urgency=medium
 
   * update upstream source to 10.3.1
-- 
2.39.5



_______________________________________________
pve-devel mailing list
pve-devel@lists.proxmox.com
https://lists.proxmox.com/cgi-bin/mailman/listinfo/pve-devel

Re: [pve-devel] [RFC PATCH 1/2] frr: add networking.service as systemd dependency

[Gabriel Goller]

ping

Hannes and I tried to fix the errors that appear when starting FRR
10.3.1 (even on a blank config). The errors look similar to these:

     Jul 22 08:36:09 pve1 zebra[845]: libyang Invalid boolean value "". (/frr-vrf:lib/vrf/state/active)
     Jul 22 08:36:09 pve1 zebra[845]: libyang Invalid type uint32 empty value. (/frr-vrf:lib/vrf/state/id)

(and some other ones).

For us this is partially fixed with this patch (which we obviously can't
upstream because frr doesn't rely on systemd) and the following PR:
https://github.com/FRRouting/frr/pull/19248.

This patch can be applied already, it also fixes some other errors like
frr starting before the dummy interfaces are created.

I'd like to wait a bit for some feedback from the maintainers before
backporting the PR though, as that's quite a tricky change and I don't
know if it is correct.


_______________________________________________
pve-devel mailing list
pve-devel@lists.proxmox.com
https://lists.proxmox.com/cgi-bin/mailman/listinfo/pve-devel

[pve-devel] applied: [RFC PATCH 1/2] frr: add networking.service as systemd dependency

[Thomas Lamprecht]

On Thu, 26 Jun 2025 15:12:12 +0200, Gabriel Goller wrote:
> Add networking.service to the 'After' dependency directive. Guarantees that
> the frr.service will start after the networking.service is done.
> 
> We had some issues with data races between FRR and ifupdown [0], mostly
> around the dummy interface. At startup, FRR and by extension fabricd is
> up faster than the dummy interface which means that the interface won't
> be configured properly. This does not happen with 10.3.1 anymore due to
> this patch [1], which fixes the problem in the fabricd daemon.
> 
> [...]

Applied, thanks!

[1/2] frr: add networking.service as systemd dependency
      commit: 84509be6283fb93c638bab0e958f275cd57a7171


_______________________________________________
pve-devel mailing list
pve-devel@lists.proxmox.com
https://lists.proxmox.com/cgi-bin/mailman/listinfo/pve-devel