Type checking interface for veth breaks container startup on 1.15.0 and 1.16.0 images, causes issues w/ ipvlan and macvlan networks.

[MxD-js]

Hi, This line here causes ipvlan network issues, docker sets the network type toipvlan in the container and this check fails and never moves past this loop when a image is restricted to a ipvlan type network per kasm documentation. Removing the veth check or adding ipvlan to the type check resolves this issue. This works fine if the image is on the kasm_default_network because docker sets the interface type to vethbut breaks when the container is restricted to ipvlan or macvlan networks.

Curious why the strict check for veth?

workspaces-core-images/src/common/startup_scripts/vnc_startup.sh

Line 446 in 41de61b

interfaces=$(ip link show type veth | awk -F: '/^[0-9]+: / {print $2}' | awk '{print $1}' | sed 's/@.*//')

issue thread for context and resolution.
kasmtech/workspaces-issues#615

[jcbshw]

Thanks for the discovery. I was also experiencing this after setting up IPVlan using the official Kasm documentation.

Making the changes locally to vnc_startup.sh resolves the issue.

[Marcussjogren]

I also have this issue, please come on and fix this issue guys! :)

[yusiwen]

I've encountered this today as well. My containers do not use virtual bridges via veth pairs, they're connected to LAN directly via MacVLAN devices. So the startup script falls into an infinite loop.

Currently the temporarily workaround is to comment this line in vnc_startup.sh when building my own image from this, like:

RUN sed -i -e 's/^wait_for_network_devices/#wait_for_network_devices/g' /dockerstartup/vnc_startup.sh

[Marcussjogren]

A simple suggestion to the developers could be to just remove the loopback interface from the search and wait for any other network interface to start up.

The line above, which someone resolved by adding "type veth type ipvlan" does not work properly on pods with virtual ethernets. So the best solution would be to exclude loopback which according to the comments in the code is the reason to why this line even existed in the first place.

ip link show | awk '! /lo:/' | awk -F: '/^[0-9]+: / {print $2}' | awk '{print $1}' | sed 's/@.*//'

[Marcussjogren]

HI Guys,

Just wanted to share my final workaround for this problem.

For each workspace that I want VLAN-attached, I have added a "File Mapping" under workspace settings. I have taken the full text from vnc_startup.sh and pasted it (dont do file upload because it seem to suck management wise).
Then I edited the network discovery-line above with:

ip link show | awk '! /lo:/' | awk -F: '/^[0-9]+: / {print $2}' | awk '{print $1}' | sed 's/@.*//'

And entered "/dockerstartup/vnc_startup.sh" as the mounting point. This overwrites the file from the container build with my version.

After that, I can change network in all possible ways without having to touch anything again.

The script now looks like this:

function wait_for_network_devices() {
        while true; do
                interfaces=$(ip link show | awk '! /lo:/' | awk -F: '/^[0-9]+: / {print $2}' | awk '{print $1}' | sed 's/@.*//')