Routes with no Oifs appearing.

[terrydooher]

Hi all!

We have a use case along the lines of:

+-------------+        +-------------+           +------------+
| Client      |        | Server B    |           | Server A   |
|             |    eth1| 20.04       |eth0       |            |
| Proc Y      +--------+             +-----------+ Proc X     |
|             |        | smcroute    |           |            |
|             |        | 2.4.2-4     |           |            |
|             |        |             |           |            |
+-------------+        +-------------+           +------------+
 Proc Y                 Proc A                    Prox X
                        Proc B

Processes A. B, Z and Y all send and receive mcast. Server B runs smcroute and currently (for some code legacy reasons) the rules are installed with direct calls to smcroute, eg:

smcroute -a eth0 192.168.1.1 239.255.0.1 eth1
smcroute -a eth1 192.168.1.1 239.255.0.1 eth0

With the script containing them invoked by placing it in systemd's /etc/networkd-dispatcher/routable.d folder.

There's an issue that happens every 4 or 5 boots of Server B, that brings the route up with no Oifs on some of the routes, eg:

(192.168.1.1,239.255.0.1)    Iif: eth0  State: resolved

Here, Client behind eth doesn't receive the group messages.
Running sudo systemctl restart smcroute and then re-running the above script resolves this:

(192.168.1.1,239.255.0.1)     Iif: eth0 Oifs: eth1  State: resolved

...and traffic flows

My question is: Is my script racing some other part of the system by trying to run via the routable interface hook? Or should this work and we're just hitting a bug that has perhaps been fixed in 2.4.3 (I'm not sure if the fix detailed there is for this issue; All the IPs in our stack are static)

Thanks!

[troglobit]

If you are running a version < 2.4.3, I suggest you upgrade to the latest version. One particular bug you might run into is an old requirement that interfaces needed to have an IP address before SMCRoute started.

[troglobit]

routable != multicast routable.

[terrydooher]

No argument from me on that front :) This is more about what systemd regards as "routable" and hence where our script is being triggered.

We've backported 2.5.5-1 from 22.04 and it's running in a test environment. Will update here once we've put it through its paces. If nothing else the addition of /etc/smcroute.d support is going to really help with our deployment + packaging.

[troglobit]

Great to hear, good luck!

[terrydooher]

Hi @troglobit - Update as promised: We have rolled out the backported 2.5.5 package across most of our Server Bs and this issue has evaporated.

Thanks again for your help :)

[troglobit]

This is great news, thank you for reporting back! :-)