[lxc-users] Eth0 not present on boot

[Guido Jäkel]

Dear Peter,

I also rename the veth to the name of the container and without this, as the default a random name will be used. You'll see why with the later you don't notice an issue:

If a container goes down, the veth did not vanish as long as there is a tcp connection using it. And in case of a shutdown scenario, there's a good chance to interrupt communication processes in the middle of conversation, e.g. during delivery of content. Then, an outerside client will wait for output until an application level timeout or because of TCP handshake timeouts. This typically will take about 5min.

In my experience, sometimes you even cant "delete the link" as described by Fayar, but i was able to rename the interface "away" (ip link set dev $DEV name $DEV._away). It will vanish later but the restarting container may use the name again for a new veth.


greetings

Guido

On 23.09.2015 03:24, Peter Steele wrote:
> On 09/22/2015 08:08 AM, Guido Jäkel wrote:
>> * Do you use  lxc.network.veth.pair  to name the hosts side of the veth?
> Yes. I rename the veth interfaces to match the names of the containers.
>> * Was the Container up and running "just before" and you (re)start it within less than 5min?
>>
> Yes. When the problem occurs, a reboot has just been issued on the container (using the "reboot" command). When the container restarts, its eth0 is missing. Another reboot and the eth0 interface reappears. Curiously, it appears to happen much more frequently on some hardware than it does on others.
> 
>