leshikus/gh-runner

Scripts which create a docker container for a github runner

gh-runner's Issues

Consider using a Docker volume instead of `/home/{user}/.config/gh-runner/{runner}/creds/`

If you change your script to use a Docker volume, then the script can be used to restart a runner by different users.

When using HTTP/HTTPS proxy the script overrides ~/.docker/config.json on the host

This overwrites other options that present in the configuration file, such as auths key, so users have to login to all Docker registries again. If the script replaces ~/.docker/config.json to set HTTP/HTTPS proxy in the container, then there are two better options:

• Pass proxy to the container using environment variables (--env HTTP_PROXY and so on)
• Use a tool such as jq to create/update the existing file in place, for example:

[[ -f ~/.docker/config.json ]] || (echo "{}" > ~/.docker/config.json)
tmp="$(mktemp)"
jq ".proxies.default = {\"httpProxy\": \"$http_proxy\", \"httpsProxy\": \"$https_proxy\", \"noProxy\": \"$no_proxy\"}" ~/.docker/config.json > "$tmp"
mv "$tmp" ~/.docker/config.json

The script keeps `tail` running in background

Every time I run the script there is a forked tail process keeps running in the background:

tail -f /home/user/.config/gh-runner/intel-sandbox.x1.test-runner/log

After N runs there are N tail processes running.

Incorrect mount for Vagrant home

The scripts mounts Vagrant home to /runner/.vagrant.d:

mount_vagrant="-v $agent_dir/vagrant:/runner/.vagrant.d -v /var/run/libvirt/:/var/run/libvirt/"

The mount point needs to be /.vagrant.d, since VAGRANT_HOME is set to that value.

mount_vagrant="-v $agent_dir/vagrant:/.vagrant.d -v /var/run/libvirt/:/var/run/libvirt/"

Vagrant directory is per runner, not per host

I've noticed that script mounts $agent_dir/vagrant to /runner/.vagrant.d. That means that every runner will need to download all vagrant boxes. Also if we want to use a custom vagrant box, then we have to distribute such box across all runners on all hosts, rather than once per host.

While this is not a big issue at the moment, we need to find a good way to distribute custom vagrant boxes to all runners.