Race when updating hosts file?

[rmja]

I am observing a peculiar issue when running kubefwd with multiple services.

Im an running on Windows with Docker Desktop.

Docker Desktop adds the following entry in the hosts file during installation "127.0.0.1 kubernetes.docker.internal" and it is this host that is specified in the cluster section in the .kube\config file (seen by kubectl config view). The result is that the port-forwarding commands are sent to https://kubernetes.docker.internal:6443/....

However, if kubefwd forwards multiple services, it happens at random that the following error is returned when forwarding services:

ERROR: error upgrading connection: error sending request: Post https://kubernetes.docker.internal:6443/api/v1/namespaces/default/pods/mypod-6dd464d68b-mhk29/portforward?timeout=32s: dial tcp: lookup kubernetes.docker.internal: no such host

Whether the error is shown for a pod or not seems random, however if a selector is specified so that only one service is selected, then the error never shows.

If I update the .kube\config file to have server: https://localhost:6443 then forwarding starts correctly everytime.

I have tried removing the call to pfo.Hostfile.Save() and it seems to solve the problem.

[cjimti]

If I update the .kube\config file to have server: https://localhost:6443 then forwarding starts correctly everytime.

That is strange since localhost and kubernetes.docker.internal are both aliases in /etc/hosts unless windows handle this differently. If kubefwd errors on kubernetes.docker.internal and not localhost then I can't see how it could be the /etc/hosts writes.

I have not personally run into this problem on my mac. I don't typically run local clusters but I do use my /etc/hosts for a ton of other aliases and have not experienced interference. If any other windows users have this issue I would be curious.e.

Using 127.0.0.1 to connect to your Docker Desktop is going to be your best option for the time being if the /etc/hosts writes are causing DNS lookup to fail.

[rmja]

@cjimti This bug has reappeared after upgrading from 1.9.1 to 1.11.1.

[rmja]

More specifically, 1.10.0 works, 1.11.0 does not.

[rmja]

@cjimti Could you please re-open this issue?

[cjimti]

Is there a reason to use kubernetes.docker.internal rather than localhost? I don't really know of a work-around for your situation since kubefwd must be able to write to /etc/hosts and if other processes like Docker Desktop are trying to add entries at the same time it will be a problem.

Is there a solution that you have in mind or something I am missing?

[rmja]

The thing is that #76 fixed the issue, at least for me on Windows. But I can see that it has been replaced with a different mechanism in the newer versions. I can simply stay on 1.10.0 for the time being...

[cjimti]

I'll re-open the ticket and explore any possible solutions but can't promise anything in the short term.

Is there an option to prevent Docker Desktop from adding kubernetes.docker.internal 127.0.0.1 every time? If so you can add it once manually and never have to worry about it.