provisioning-catalog/providers/hetzner/runner.nu

# Hetzner Cloud ComputeProvider runner.
#
# Invocation (via runner-dispatch):
#   <json_input> | nu runner.nu <command>
#
# Commands: spawn | destroy | status | describe
# Input:    JSON record on stdin matching the interface contract
# Output:   JSON record on stdout (spawn/status/describe); silent (destroy)
# Errors:   human-readable message to stderr, non-zero exit

# ── Server type resolution ────────────────────────────────────────────────────

def select-server-type [cpu: int, memory_gb: int]: nothing -> string {
    let types = [
        { name: "cax11", cpu: 2,  memory_gb: 4  }
        { name: "cax21", cpu: 4,  memory_gb: 8  }
        { name: "cax31", cpu: 8,  memory_gb: 16 }
        { name: "cax41", cpu: 16, memory_gb: 32 }
    ]
    let fit = ($types | where { |t| $t.cpu >= $cpu and $t.memory_gb >= $memory_gb })
    if ($fit | is-empty) {
        error make { msg: $"No CAX type satisfies cpu=($cpu) memory_gb=($memory_gb). Maximum: cax41 16cpu/32GB" }
    }
    ($fit | first).name
}

# ── Cloud-init generation ─────────────────────────────────────────────────────
# Generates a cloud-config that configures systemd-resolved to use a private DNS
# server. Written to a temp file and passed to hcloud via --user-data-from-file.

def generate-cloud-init [dns_ip: string]: nothing -> string {
    $"#cloud-config
write_files:
  - path: /etc/systemd/resolved.conf.d/10-private-dns.conf
    content: |
      [Resolve]
      DNS=($dns_ip)
      FallbackDNS=1.1.1.1 8.8.8.8
    owner: root:root
    permissions: '0644'
runcmd:
  - systemctl restart systemd-resolved
  - ip route replace 10.200.2.0/24 via 10.0.8.1 || true
  - grep -qF 'reg.librecloud.online' /etc/hosts || echo '10.200.2.253 reg.librecloud.online termas.librecloud.online' >> /etc/hosts
  - grep -qF 'daoreg.librecloud.online' /etc/hosts || echo '10.0.8.6 daoreg.librecloud.online daotermas.librecloud.online' >> /etc/hosts
"
}

# ── Image resolution ──────────────────────────────────────────────────────────
# image_ref containing '=' is a label selector; otherwise treated as ID/name.

def resolve-image [image_ref: string]: nothing -> string {
    if ($image_ref | str contains "=") {
        let r = (do { ^hcloud image list --selector $image_ref --type snapshot --output json } | complete)
        if $r.exit_code != 0 {
            error make { msg: $"hcloud image list failed: ($r.stderr | str trim)" }
        }
        let imgs = ($r.stdout | from json)
        if ($imgs | is-empty) {
            error make { msg: $"No snapshot found with selector: ($image_ref)" }
        }
        ($imgs | sort-by created | last).id | into string
    } else {
        $image_ref
    }
}

# ── SSH port readiness ────────────────────────────────────────────────────────
# Polls TCP port every 5 seconds up to max_attempts times.

def wait-for-port [ip: string, port: int, max_attempts: int]: nothing -> nothing {
    mut attempt = 0
    mut open = false
    while $attempt < $max_attempts and (not $open) {
        let r = (do { ^nc -z -w 3 $ip $port } | complete)
        if $r.exit_code == 0 {
            $open = true
        } else {
            sleep 5sec
            $attempt = $attempt + 1
        }
    }
    if not $open {
        error make { msg: $"TCP port ($port) on ($ip) unreachable after ($max_attempts) attempts (~($max_attempts * 5)s)" }
    }
}

# ── spawn ─────────────────────────────────────────────────────────────────────

def cmd-spawn [input: record]: nothing -> record {
    let server_type = (select-server-type $input.cpu $input.memory_gb)
    let image_id    = (resolve-image $input.image_ref)

    let ts          = (date now | format date "%Y%m%d%H%M%S")
    let rand        = (random int 1000..9999)
    let server_name = $"runner-($ts)-($rand)"

    mut args = [
        "server" "create"
        "--name"     $server_name
        "--type"     $server_type
        "--image"    $image_id
        "--location" $input.location
        "--ssh-key"  $input.ssh_key_ref
        "--output"   "json"
    ]

    let network      = (try { $input.network_ref  } catch { null })
    let firewall     = (try { $input.firewall_ref } catch { null })
    let internal_dns = (try { $input.internal_dns } catch { null })

    if ($network != null) {
        $args = ($args | append ["--network" $network])
    }
    if ($firewall != null) {
        $args = ($args | append ["--firewall" $firewall])
    }

    let cloud_init_file = if ($internal_dns != null) {
        let f = (mktemp --suffix .yaml)
        (generate-cloud-init $internal_dns) | save --force $f
        $f
    } else {
        null
    }

    if ($cloud_init_file != null) {
        $args = ($args | append ["--user-data-from-file" $cloud_init_file])
    }

    let create_args = $args
    let r = (do { ^hcloud ...$create_args } | complete)

    if ($cloud_init_file != null) {
        ^rm -f $cloud_init_file
    }

    if $r.exit_code != 0 {
        error make { msg: $"hcloud server create failed: ($r.stderr | str trim)" }
    }

    let created   = ($r.stdout | from json)
    let server_id = ($created.server.id | into string)
    let ip        = $created.server.public_net.ipv4.ip

    wait-for-port $ip 22 36

    {
        handle:     ({ name: $server_name, id: $server_id } | to json)
        ssh_host:   $ip
        ssh_port:   22
        expires_at: null
    }
}

# ── destroy ───────────────────────────────────────────────────────────────────

def cmd-destroy [input: record]: nothing -> nothing {
    let h = ($input.handle | from json)
    let r = (do { ^hcloud server delete $h.name } | complete)
    if $r.exit_code != 0 {
        error make { msg: $"hcloud server delete failed: ($r.stderr | str trim)" }
    }
}

# ── status ────────────────────────────────────────────────────────────────────

def cmd-status [input: record]: nothing -> record {
    let h = ($input.handle | from json)
    let r = (do { ^hcloud server describe $h.name --output json } | complete)
    if $r.exit_code != 0 {
        if ($r.stderr | str contains "not found") {
            return { running: false, ssh_host: null }
        }
        error make { msg: $"hcloud server describe failed: ($r.stderr | str trim)" }
    }
    let srv     = ($r.stdout | from json)
    let running = ($srv.status == "running")
    let ip      = (try { $srv.public_net.ipv4.ip } catch { null })
    { running: $running, ssh_host: $ip }
}

# ── describe ──────────────────────────────────────────────────────────────────

def cmd-describe []: nothing -> record {
    {
        locations: ["fsn1" "nbg1" "hel1" "ash" "hil"]
        server_types: [
            { name: "cax11", cpu: 2,  memory_gb: 4,  disk_gb: 40  }
            { name: "cax21", cpu: 4,  memory_gb: 8,  disk_gb: 80  }
            { name: "cax31", cpu: 8,  memory_gb: 16, disk_gb: 160 }
            { name: "cax41", cpu: 16, memory_gb: 32, disk_gb: 320 }
        ]
        image_query: "hcloud image list --type snapshot --selector app=buildkit-runner-golden --output json"
    }
}

# ── entry point ───────────────────────────────────────────────────────────────

def main [command: string] {
    let input = (^cat | from json)

    match $command {
        "spawn"    => { cmd-spawn    $input | to json | print }
        "destroy"  => { cmd-destroy  $input }
        "status"   => { cmd-status   $input | to json | print }
        "describe" => { cmd-describe        | to json | print }
        _          => {
            error make { msg: $"Unknown command: ($command). Allowed: spawn destroy status describe" }
        }
    }
}