cargo.ayedo.cloud/ayedo/hcloud/infrastructure
Inventory-driven Hetzner Cloud provisioning. After create, patches ansible_host,
IPs and (optionally) ansible_user / ansible_ssh_port into workspace inventory.yml.
Does not use the hcloud dynamic inventory plugin.
Actions
| Action | Description |
|---|---|
install |
SSH key → PG → network → firewall → VMs → patch inventory → wait for SSH |
ensure-ssh-key |
Create/update the Hetzner project SSH key from artifacts/secrets/id_rsa.pub |
uninstall |
Delete VMs; delete only managed network/firewall/PG/SSH key |
patch-inventory |
Refresh IPs (+ SSH connection vars) from Hetzner into inventory.yml |
show-plan |
Print resolved plan (no writes) |
polycrate run infrastructure show-plan
polycrate run infrastructure ensure-ssh-key # optional standalone
polycrate run infrastructure install
polycrate run hardening install # optional, idempotent
polycrate run k8s install
SSH bootstrap (sysops / 4422)
Default aligns with cargo.ayedo.cloud/ayedo/linux/hardening:
ssh:
manage: true
bootstrap_enabled: true
bootstrap_user: sysops
port: 4422
wait: true
wait_timeout: 300
public_key_file: artifacts/secrets/id_rsa.pub
On first VM create, cloud-init:
- creates
sysopswith sudo + authorized key - sets sshd to port
4422,PermitRootLogin no - install waits until that port accepts connections
user_datais applied by Hetzner only at create time. Existing servers are not re-bootstrapped unless rebuilt.
Override with raw defaults.user_data if needed (wins over generated bootstrap).
Network / firewall / SSH key reuse
| Field | Meaning |
|---|---|
*.enabled |
Attach resource to VMs |
*.manage |
true = create/update/delete; false = reuse existing by name |
*.name |
Hetzner resource name (empty → workspace.name) |
network:
enabled: true
manage: false
name: customer-prod-net
firewall:
enabled: true
manage: false
name: customer-prod-fw
ssh:
manage: false
key_name: customer-shared-key
ensure-ssh-key always pushes the local public key (forces manage for that action).
Inventory (k8s contract)
Aligned with cargo.ayedo.cloud/ayedo/linux/k8s-* inventory mode (cluster_ip_detection.enabled: false):
| Field | Where | Meaning |
|---|---|---|
ansible_host |
host | SSH target from Polycrate (public | private) |
public_ip / public_interface |
host / all.vars | Node external IP / NIC |
internal_ip / internal_interface |
host / all.vars | Private cluster plane (also written as private_ip) |
primary_ip / primary_interface |
all.vars | API/join endpoint for peers + workers |
primary_ip resolution (inventory.manage_primary_ip: true):
inventory.primary_ip— explicit override (VIP / managed / external CP)inventory.primary_host— use that host'sinternal_ip- else:
internal_ipof the first host ininventory.controlplane_group
inventory:
controlplane_group: k8s_controlplane
manage_primary_ip: true
# primary_ip: "10.11.3.99" # managed/external API
# primary_host: festo-controlplane-1
internal_interface: enp7s0
public_interface: eth0
primary_interface: enp7s0
Hosts live in inventory.yml (source of truth). The block also patches:
ansible_user/ansible_ssh_portwheninventory.patch_ssh_connectionand bootstrap enabled
Per-host overrides: hcloud_server_type, hcloud_image, hcloud_location,
hcloud_enable_ipv4, hcloud_enable_ipv6, hcloud_labels.