Datapath healing does not work when host interface goes down and up again about cmd-forwarder-vpp

[glazychev-art]

I don't think that

docker network disconnect kind kind-worker2
docker network connect kind kind-worker2

is the same as up/down interface. I did that and I got:
before:

77: eth2@if78: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP group default 
    link/ether 02:42:ac:12:00:05 brd ff:ff:ff:ff:ff:ff link-netnsid 0
    inet 172.18.0.5/16 brd 172.18.255.255 scope global eth2
       valid_lft forever preferred_lft forever
    inet6 fc00:f853:ccd:e793::5/64 scope global nodad 
       valid_lft forever preferred_lft forever
    inet6 fe80::42:acff:fe12:5/64 scope link 
       valid_lft forever preferred_lft forever

after:

79: eth3@if80: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP group default 
    link/ether 02:42:ac:12:00:05 brd ff:ff:ff:ff:ff:ff link-netnsid 0
    inet 172.18.0.5/16 brd 172.18.255.255 scope global eth3
       valid_lft forever preferred_lft forever
    inet6 fc00:f853:ccd:e793::5/64 scope global nodad 
       valid_lft forever preferred_lft forever
    inet6 fe80::42:acff:fe12:5/64 scope link 
       valid_lft forever preferred_lft forever

This is not the same interface. It has different name and different ID.
I think this is the same as changing the NIC, but setting the same address.

I think these commands are more correct:

docker exec kind-worker ip link set eth0 down
docker exec kind-worker ip link set eth0 up

As for the problem, it looks like this is the same - #927
We found some bugs and are now preparing fixes

[szvincze]

Is there a document or presentation material that would explain how the interface and the related events are handled?

Based on our observation the changes of the admin/operational state of the host interface reflected immediately by the forwarder. So, it would be interesting to know how the detection works. Is it event driven or done by a detection mechanism?

[szvincze]

That's very good news. Have you seen similar symptoms on the host interface like in case of #927?
I mean RX counters are not increasing or not printed out in vppctl show interface.

[glazychev-art]

Yes, I restarted the forwarders several times and the problem is reproduced.
But in fact, I see that TX is not increasing..

[NikitaSkrynnik]

Checked the setup several times and couldn't reproduce the bug