Multiple DHCP network interfaces fail to initialize

[benok]

When configuring two or more network interfaces to use DHCP, the setup becomes very unstable:

rancher:
  network:
    interfaces:
      eth0:
        dhcp: true
      eth1:
        dhcp: true

Looking at the logs from sudo system-docker logs network, this appears to be caused by conflicts when multiple dhcpcd processes start simultaneously(by goroutine):

[            ] netconf:info: Apply Network Config
[            ] netconf:info: Running DHCP on eth1: dhcpcd -MA4 --timeout 10 -w --debug eth1
[            ] netconf:info: Running DHCP on eth0: dhcpcd -MA4 --timeout 10 -w --debug eth0
main: pidfile_lock: /var/run/dhcpcd/pid: Operation canceled
dhcpcd-10.0.1 starting
[            ] netconf:error: Failed to run dhcpcd for eth1: exit status 1
spawned manager process on PID 526

This issue appears to be related to rancher#2986.

[olljanat]

Yes, this is old known issue (which I have solved by having static IP in eth1) but very good finding that it actually caused by multiple dhcpcd processes starting same time.

Now when I read dhcpcd documentation it seems that challenge is related to starting of manager process:

which causes both of those trying to start manager process which would listen /var/run/dhcpcd/pid

If you go inside of network container and kill dhcpcd process the file /var/run/dhcpcd/pid actually disappears.

And if you then run command dhcpcd -MA4 -e force_hostname=true --timeout 10 -w --debug eth0 it will print spawned manager process on PID ... and all later runs of the command will simply print sending commands to dhcpcd process.

However, on both of those cases those command finishes very quickly so I think that proper fix to this issue would be refactor logic in https://github.com/burmilla/os/blob/v2.0.3/pkg/netconf/netconf_linux.go#L252-L265 on way that it will not use goroutines but simply loop interfaces one by one like it was the case before rancher#1869

[olljanat]

Fix included to v2.0.4.
Thanks!