[pmg-devel] [PATCH pmg-api] pmgtunnel: adapt to read_etc_network_interfaces using run_command

[Stoiko Ivanov]

the changes in pve-common:
346ff97 ("inotify/interfaces: use 'ip link' instead of /proc/net/dev")
made the /etc/network/interfaces parser call `ip link -details` via
run_command. This was the first use of a external command apart from
the ssh processes for the tunnels.

pmgtunnel changed the SIGCHLD handler before entering it's main-loop,
which in turn made run_command return an error on each invocation:
`command 'ip -details -json link show' failed: failed to execute`

The issue is not directly reproducible, as /etc/network/interfaces is
only read if the node-name does not resolv to a non-loopback ip (via
/etc/hosts or DNS).

reported in our community-forum:
https://forum.proxmox.com/threads/.169854/

Signed-off-by: Stoiko Ivanov <s.ivanov at proxmox.com>
---
minimally tested on a PMG Cluster I have for testing, where I reproduced
this by commenting each node's entry in their /etc/hosts and stopping the
DNS-Server used by the nodes.
 src/PMG/Service/pmgtunnel.pm | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/src/PMG/Service/pmgtunnel.pm b/src/PMG/Service/pmgtunnel.pm
index 513bba0..7b9fa28 100644
--- a/src/PMG/Service/pmgtunnel.pm
+++ b/src/PMG/Service/pmgtunnel.pm
@@ -180,7 +180,10 @@ sub run {
         $next_update = time() + $updatetime;
 
         eval {
+            # reset SIGCHLD handler as ClusterConfig::new uses run_command (for reading ip link)
+            $SIG{CHLD} = 'DEFAULT';
             my $cinfo = PMG::ClusterConfig->new(); # reload
+            $SIG{CHLD} = \&finish_children;
             $self->purge_tunnels($cinfo);
             $self->start_tunnels($cinfo);
         };
-- 
2.39.5