Ubuntu user does not exist in container at startup of container

[maratstepanov]

At https://github.com/flongo82/raspberry_virtualization/blob/daemon/server.js#L395:

exec(`lxc exec ${name} -- usermod -a -G gpio ubuntu`, (error, stdout, stderr) => {

we get error:

An error has occured while adding ubuntu user to gpio group: Error: Command failed: lxc exec test1 -- usermod -a -G gpio ubuntu
usermod: user 'ubuntu' does not exist

If we get into the container, we can see that ubuntu user exists.

[flongo82]

Yes I know this issue. This is due to the fact that the cloud-init script that creates the ubuntu user has not finished yet when you try to add the ubuntu user to the gpio group.

We should add a certain delay between the creation of the container and the adding of the ubuntu user to the gpio group. Some kind of sleep is easily implementable in Node.js.

Let's put 1 second of sleep and see if it is enough.

[maratstepanov]

Okay, I'll try it. There is pull request with resolving this issue other way - please ignore it.

[maratstepanov]

Yes, 1 second of delay did work.

[maratstepanov]

Sorry, it did not work actually (action was performing outside the container).
It takes about 5 seconds for ubuntu user to appear in container.
We can add user to group in parallel function, with adding timeout of 5 seconds in order to avoid blocking all other nested operations. What do you think?

[flongo82]

It is a good idea, but I have the feeling that some other operations later need the ubuntu user to be in that group. Can you check this?

[maratstepanov]

All operations inside lxc containter are being performed as root, if I don't mistake. I did this conclusion because when I perform lxc exec test1 -- bash, it shows me root's prompt: root@test1:~#
After few retries, I found optimal timeout, when it always works: 15 seconds.

[flongo82]

This means that it takes at least 15 seconds for the daemon to create the virtual rasp and reply to the user. Does this break HTTP timeouts?

By the way, I think that the time the init-cloud script needs to finish strongly depends on the status of the physical rasp. If the rasp is overloaded it is probable that this timeout will not be okay anymore.

I would keep it as it is but keep this issue open to remember that there could be this problem in the future.

[maratstepanov]

Does this break HTTP timeouts?
No, server answers immediately, and task is being performed in background

I think that best solution is to periodically check for user existence and add it to group as soon as it appears. If you don't mind, I'll do it, but with low priority.

[maratstepanov]

Latest commit's code waits 15 seconds (in the parrallel thread) and creates user.
I have tried performing check, but somewhy it blocks server for check period. Will be investigating later, when more critical issues are resolved.