network namespace issue when used with Gitlab runner

[imperialguy]

Using podman 3.4 on rhel8

Scenario 1
I have a shell script (let's call it prepare.sh) with just the following line in it:

podman run -d docker.io/library/alpine:latest

The above script is called by: a.) shell; b.) a Gitlab runner (example) as well. And it works perfectly fine in both cases.

Scenario 2
I make the following minor change:

podman run -d --entrypoint /sbin/init docker.io/library/alpine:latest

or

podman run -d -it docker.io/library/alpine:latest

In other words, the above two modified commands are calling for a perpetual running of the container.

Now, when the above script is called by a.) local shell - it works perfectly fine; b.) But, when it is called by a Gitlab runner, and then when the Gitlab runner does execute that prepare.sh script, the following happens:

ERRO[0000] error joining network namespace for container f62f7b061205dce066156cf4ea90f3a90ab417b45cf8b3a5f745ffd6fca13fe7: error retrieving network namespace at /tmp/custom-executor886608394/podman-run-1001/netns/cni-e606bcb7-7156-feeb-0866-15415ebab091: failed to Statfs "/tmp/custom-executor886608394/podman-run-1001/netns/cni-e606bcb7-7156-feeb-0866-15415ebab091": no such file or directory

The problem is, for some reason when the same script is called by the Gitlab runner, podman is looking for a network namespace inside a different folder.

So, in the above case, it's looking for a network namespace inside /tmp/custom-executor886608394/podman-run-1001/netns instead of just/tmp/podman-run-1001/netns which is the default location for that user.

Also, both the Gitlab runner and manual execution are using the same user.

Not sure if I can really blame it on Gitlab runner, because it ran Scenario 1 without any hiccups. The only modification I made in Scenario 2 is the addition of --entrypoint or -it flags, and all of a sudden it leads to the network namespace error i.e., only when called by the Gitlab runner.

Any ideas/thoughts?

[imperialguy]

Resolved #14192