Parameter	Type	Description
`cloud`	string	Name of the cloud config from clouds.yaml to use
`clouds_config`	string	Optional. Path to clouds.yaml
`name`	string	Name of the Auto Scaling Group
`boot_time`	string	Optional. Maximum wait time for instance to boot up. During that time plugin check Cloud-Init signatures.
`server_spec`	object	Server spec used to create instances. See: Compute API

Default connector config

Parameter	Default
`os`	`linux`
`protocol`	`ssh`
`username`	`unset`
`use_static_credentials`	`true`

OpenStack setup

You should create a special user (recommended) and project (optional), then export clouds.yaml with credentials for that cloud.
You may create a tenant network for workers, in that case don't forget to add a router. In that case manager VM should have two ports: external and that tenant network, so it will be able to connect to the worker instances.
You should upload a special image with gitlab-runner and container runtime installed in it. For example we use Fedora 38 with Podman.
You should generate SSH keypair which will be used my manager instance to connect to workers. Public key must be added to Nova from the user.

Example runner config

concurrent = 16
check_interval = 0
shutdown_timeout = 0
log_level = "info"

[session_server]
session_timeout = 1800

[[runners]]
name = "manager"
url = "https://gitlab.com"
token = "token"
executor = "docker-autoscaler"
output_limit = 10240
shell = "bash"
environment = [
  "FF_NETWORK_PER_BUILD=1"
  ]

[runners.cache]
Type = "s3"
Shared = true

[runners.cache.s3]
ServerAddress = "s3.foo.bar"
AccessKey = "access"
SecretKey = "secret"
BucketName = "cache"

[runners.docker]
disable_entrypoint_overwrite = false
oom_kill_disable = false
disable_cache = true
shm_size = 0
network_mtu = 0
host = "unix:///run/user/1000/podman/podman.sock"
tls_verify = false
image = "quay.io/podman/stable"
privileged = true
pull_policy = ["always", "always"]

[runners.autoscaler]
capacity_per_instance = 1
max_use_count = 10
max_instances = 16
plugin = "fleeting-plugin-openstack"

[runners.autoscaler.plugin_config]
cloud = "runner"
# clouds_file = "/etc/openstack/clouds.yaml"
name = "podman-runners"
boot_time = "10m"

[runners.autoscaler.plugin_config.server_spec]
name = "podman-runner-%d"                                               # %d replaced with instance index
description = "GitLab CI Podman runners with autoscaling"
tags = ["GitLab", "CI", "Podman"]
imageRef = "d5460af5-83f3-47d7-9c4f-80294c66b267"                       # Fedora 38 + gitlab-runner
flavorRef = "4e9d4fa4-a703-4850-8bc1-58b5e139ab57"                      # xlarge flavor
key_name = "runner_key"                                                 # SSH public key for worker nodes
networks = [ { uuid = "f05e7f64-9e0f-4c5c-acb0-b636000d7301" } ]        # tenant network
security_groups = [ "cee22d91-bb9a-455d-be88-e911d3cb066a" ]            # allow SSH ingress from tenant network
scheduler_hints = { group = "a9c941cb-5b34-46e0-8fc6-7471e3b77c75" }    # [Soft-]Anti-Affinity group
# runcmd required for podman, see also gitlab docs
user_data = '''#cloud-config
package_update: true
package_upgrade: true
runcmd:
- systemctl --now enable podman.socket
- sudo -u fedora systemctl --user --now enable podman.socket
- loginctl enable-linger gitlab-runner
'''

[runners.autoscaler.connector_config]
username = "fedora"
password = ""
key_path = "/etc/gitlab-runner/id_rsa"
use_static_credentials = true
keepalive = "30s"
timeout = "0m"
use_external_addr = false

[[runners.autoscaler.policy]]
idle_count = 2
idle_time = "30m0s"
scale_factor = 0.0
scale_factor_limit = 0