Aggiunge registry nodi stream e provisioning Hetzner/lab per lo scale-out.

Prepara l'architettura multi-nodo (MediaMTX+ffmpeg) con assignment URL per sessione, admin di provision/drain e provider Cloud/DNS astratti verso mltv-stream.net. Co-authored-by: Cursor <cursoragent@cursor.com>
2026-08-09 19:52:47 +02:00
parent 45bdda7c95
commit c3878cdc6d
48 changed files with 1980 additions and 25 deletions
@@ -0,0 +1,35 @@
+# frozen_string_literal: true
+
+module Streams
+  module CloudProviders
+    class Error < StandardError; end
+
+    # Descrittore restituito da create_node / list.
+    Instance = Struct.new(
+      :id, :name, :public_ip, :private_ip, :status, :raw,
+      keyword_init: true
+    )
+
+    class Base
+      def create_node(name:, labels: {})
+        raise NotImplementedError
+      end
+
+      def destroy_node(instance_id)
+        raise NotImplementedError
+      end
+
+      def list_nodes(labels: {})
+        raise NotImplementedError
+      end
+
+      def wait_until_running(instance_id, timeout: 120)
+        raise NotImplementedError
+      end
+
+      def public_ip(instance_id)
+        raise NotImplementedError
+      end
+    end
+  end
+end
@@ -0,0 +1,183 @@
+# frozen_string_literal: true
+
+require "faraday"
+
+module Streams
+  module CloudProviders
+    # Hetzner Cloud — create/destroy server per nodi stream.
+    #
+    # ENV:
+    #   HCLOUD_TOKEN (obbligatorio)
+    #   HCLOUD_LOCATION (default fsn1)
+    #   HCLOUD_SERVER_TYPE (default cpx21)
+    #   HCLOUD_IMAGE (default debian-12)
+    #   HCLOUD_SSH_KEY (nome chiave, default matchlivetv-stream)
+    #   HCLOUD_NETWORK_ID (opzionale, private network / WireGuard prep)
+    #   HCLOUD_USER_DATA_FILE (opzionale, cloud-init path)
+    class Hetzner < Base
+      API = "https://api.hetzner.cloud/v1"
+
+      def initialize(token: ENV.fetch("HCLOUD_TOKEN"), conn: nil)
+        @token = token
+        @conn = conn
+      end
+
+      def create_node(name:, labels: {})
+        body = {
+          name: name,
+          server_type: ENV.fetch("HCLOUD_SERVER_TYPE", "cpx21"),
+          image: ENV.fetch("HCLOUD_IMAGE", "debian-12"),
+          location: ENV.fetch("HCLOUD_LOCATION", "fsn1"),
+          start_after_create: true,
+          labels: default_labels.merge(stringify_labels(labels)),
+          ssh_keys: [ENV.fetch("HCLOUD_SSH_KEY", "matchlivetv-stream")],
+          public_net: {
+            enable_ipv4: true,
+            enable_ipv6: false
+          }
+        }
+        network_id = ENV["HCLOUD_NETWORK_ID"].presence
+        body[:networks] = [network_id.to_i] if network_id
+        user_data = cloud_init_user_data
+        body[:user_data] = user_data if user_data.present?
+
+        data = post("/servers", body)
+        server = data["server"] || {}
+        action = data["action"]
+        wait_action!(action) if action
+        instance = wait_until_running(server["id"].to_s)
+        instance.name = name
+        instance
+      end
+
+      def destroy_node(instance_id)
+        delete("/servers/#{instance_id}")
+        true
+      end
+
+      def list_nodes(labels: {})
+        params = {}
+        label_selector = labels.map { |k, v| "#{k}=#{v}" }.join(",")
+        params[:label_selector] = label_selector if label_selector.present?
+        params[:label_selector] ||= "matchlivetv=true,role=stream-node"
+
+        data = get("/servers", params)
+        Array(data["servers"]).map { |s| instance_from_server(s) }
+      end
+
+      def wait_until_running(instance_id, timeout: 180)
+        deadline = Time.now + timeout
+        loop do
+          data = get("/servers/#{instance_id}")
+          server = data["server"]
+          status = server["status"]
+          if status == "running"
+            return instance_from_server(server)
+          end
+          raise Error, "Timeout attesa server Hetzner #{instance_id} (status=#{status})" if Time.now >= deadline
+
+          sleep 3
+        end
+      end
+
+      def public_ip(instance_id)
+        wait_until_running(instance_id).public_ip
+      end
+
+      private
+
+      def default_labels
+        {
+          "matchlivetv" => "true",
+          "role" => "stream-node",
+          "env" => ENV.fetch("STREAM_NODE_ENV", "prod")
+        }
+      end
+
+      def stringify_labels(labels)
+        labels.to_h.transform_keys(&:to_s).transform_values(&:to_s)
+      end
+
+      def instance_from_server(server)
+        public_ip = server.dig("public_net", "ipv4", "ip")
+        private_ip = Array(server["private_net"]).first&.dig("ip")
+        Instance.new(
+          id: server["id"].to_s,
+          name: server["name"],
+          public_ip: public_ip,
+          private_ip: private_ip.presence || public_ip,
+          status: server["status"],
+          raw: server
+        )
+      end
+
+      def cloud_init_user_data
+        path = ENV["HCLOUD_USER_DATA_FILE"].presence
+        return File.read(path) if path && File.file?(path)
+
+        ENV["HCLOUD_USER_DATA"].presence
+      end
+
+      def conn
+        @conn ||= Faraday.new(url: API) do |f|
+          f.request :json
+          f.response :json, content_type: /\bjson$/
+          f.adapter Faraday.default_adapter
+        end
+      end
+
+      def auth_headers
+        { "Authorization" => "Bearer #{@token}" }
+      end
+
+      def get(path, params = {})
+        response = conn.get(path) do |req|
+          req.headers.update(auth_headers)
+          req.params.update(params)
+        end
+        unwrap!(response)
+      end
+
+      def post(path, body)
+        response = conn.post(path) do |req|
+          req.headers.update(auth_headers)
+          req.body = body
+        end
+        unwrap!(response)
+      end
+
+      def delete(path)
+        response = conn.delete(path) do |req|
+          req.headers.update(auth_headers)
+        end
+        return {} if response.status == 204
+
+        unwrap!(response)
+      end
+
+      def unwrap!(response)
+        unless response.success?
+          raise Error, "Hetzner Cloud API #{response.status}: #{response.body.inspect}"
+        end
+
+        response.body.is_a?(Hash) ? response.body : {}
+      end
+
+      def wait_action!(action, timeout: 120)
+        return unless action.is_a?(Hash) && action["id"]
+
+        deadline = Time.now + timeout
+        id = action["id"]
+        loop do
+          data = get("/actions/#{id}")
+          status = data.dig("action", "status")
+          return if status == "success"
+          raise Error, "Hetzner action #{id} failed: #{data.inspect}" if status == "error"
+          raise Error, "Timeout action Hetzner #{id}" if Time.now >= deadline
+
+          sleep 2
+        end
+      end
+    end
+  end
+end
@@ -0,0 +1,45 @@
+# frozen_string_literal: true
+
+module Streams
+  module CloudProviders
+    # Lab senza API Proxmox: simula create/destroy e riusa MediaMTX home per i path.
+    # Utile per testare registry, assignment e admin senza secondi host.
+    class LocalLab < Base
+      def create_node(name:, labels: {})
+        Instance.new(
+          id: "sim-#{name}",
+          name: name,
+          public_ip: labels[:public_ip].presence || "127.0.0.1",
+          private_ip: labels[:private_ip].presence || "127.0.0.1",
+          status: "running",
+          raw: { simulated: true, labels: labels }
+        )
+      end
+
+      def destroy_node(instance_id)
+        true
+      end
+
+      def list_nodes(labels: {})
+        StreamNode.where(provider: "local", role: "lab").map do |node|
+          Instance.new(
+            id: node.provider_instance_id,
+            name: node.slug,
+            public_ip: node.metadata["public_ip"],
+            private_ip: node.metadata["private_ip"],
+            status: node.status == "ready" ? "running" : node.status,
+            raw: node.metadata
+          )
+        end
+      end
+
+      def wait_until_running(instance_id, timeout: 120)
+        Instance.new(id: instance_id, name: instance_id, status: "running")
+      end
+
+      def public_ip(instance_id)
+        "127.0.0.1"
+      end
+    end
+  end
+end
@@ -0,0 +1,162 @@
+# frozen_string_literal: true
+
+require "faraday"
+
+module Streams
+  module CloudProviders
+    # Clone/start/stop di VM template su Proxmox VE (API token).
+    #
+    # ENV richiesti:
+    #   PROXMOX_API_URL, PROXMOX_TOKEN_ID, PROXMOX_TOKEN_SECRET,
+    #   PROXMOX_NODE, PROXMOX_TEMPLATE_VMID
+    class ProxmoxLab < Base
+      def initialize(
+        api_url: ENV.fetch("PROXMOX_API_URL"),
+        token_id: ENV.fetch("PROXMOX_TOKEN_ID"),
+        token_secret: ENV.fetch("PROXMOX_TOKEN_SECRET"),
+        node: ENV.fetch("PROXMOX_NODE"),
+        template_vmid: ENV.fetch("PROXMOX_TEMPLATE_VMID"),
+        verify_ssl: ENV.fetch("PROXMOX_VERIFY_SSL", "false") == "true"
+      )
+        @api_url = api_url.to_s.chomp("/")
+        @token_id = token_id
+        @token_secret = token_secret
+        @node = node
+        @template_vmid = template_vmid.to_i
+        @verify_ssl = verify_ssl
+      end
+
+      def create_node(name:, labels: {})
+        newid = next_vmid
+        post("/nodes/#{@node}/qemu/#{@template_vmid}/clone", {
+          newid: newid,
+          name: name,
+          full: 1,
+          target: @node
+        })
+        post("/nodes/#{@node}/qemu/#{newid}/status/start", {})
+        wait_until_running(newid.to_s)
+        ip = public_ip(newid.to_s)
+        Instance.new(
+          id: newid.to_s,
+          name: name,
+          public_ip: ip,
+          private_ip: ip,
+          status: "running",
+          raw: { node: @node, vmid: newid, labels: labels }
+        )
+      end
+
+      def destroy_node(instance_id)
+        vmid = instance_id.to_i
+        begin
+          post("/nodes/#{@node}/qemu/#{vmid}/status/stop", { timeout: 30 })
+        rescue Error
+          # già spenta
+        end
+        sleep 2
+        delete("/nodes/#{@node}/qemu/#{vmid}", { purge: 1 })
+        true
+      end
+
+      def list_nodes(labels: {})
+        items = get("/nodes/#{@node}/qemu")
+        Array(items).filter_map do |row|
+          name = row["name"].to_s
+          next unless name.start_with?("mltv-stream-") || name.start_with?("ingest-")
+
+          Instance.new(
+            id: row["vmid"].to_s,
+            name: name,
+            public_ip: nil,
+            private_ip: nil,
+            status: row["status"],
+            raw: row
+          )
+        end
+      end
+
+      def wait_until_running(instance_id, timeout: 180)
+        deadline = Time.now + timeout
+        loop do
+          status = get("/nodes/#{@node}/qemu/#{instance_id}/status/current")
+          return Instance.new(id: instance_id.to_s, status: "running", raw: status) if status["status"] == "running"
+          raise Error, "Timeout attesa VM #{instance_id}" if Time.now >= deadline
+
+          sleep 3
+        end
+      end
+
+      def public_ip(instance_id)
+        agent = get("/nodes/#{@node}/qemu/#{instance_id}/agent/network-get-interfaces")
+        interfaces = agent.is_a?(Hash) ? agent["result"] : nil
+        Array(interfaces).each do |iface|
+          Array(iface["ip-addresses"]).each do |addr|
+            ip = addr["ip-address"].to_s
+            next if ip.blank? || ip.start_with?("127.") || ip.include?(":")
+
+            return ip
+          end
+        end
+        ENV["STREAM_LAB_FALLBACK_IP"].presence || "127.0.0.1"
+      rescue Error
+        ENV["STREAM_LAB_FALLBACK_IP"].presence || "127.0.0.1"
+      end
+
+      private
+
+      def next_vmid
+        used = Array(get("/cluster/resources", type: "vm")).map { |r| r["vmid"].to_i }
+        candidate = ENV.fetch("PROXMOX_VMID_START", "9100").to_i
+        candidate += 1 while used.include?(candidate)
+        candidate
+      end
+
+      def conn
+        @conn ||= Faraday.new(url: "#{@api_url}/api2/json") do |f|
+          f.request :url_encoded
+          f.response :json, content_type: /\bjson$/
+          f.adapter Faraday.default_adapter
+          f.ssl[:verify] = @verify_ssl
+        end
+      end
+
+      def auth_headers
+        { "Authorization" => "PVEAPIToken=#{@token_id}=#{@token_secret}" }
+      end
+
+      def get(path, params = {})
+        response = conn.get(path) do |req|
+          req.headers.update(auth_headers)
+          req.params.update(params)
+        end
+        unwrap!(response)
+      end
+
+      def post(path, body = {})
+        response = conn.post(path) do |req|
+          req.headers.update(auth_headers)
+          req.body = body
+        end
+        unwrap!(response)
+      end
+
+      def delete(path, params = {})
+        response = conn.delete(path) do |req|
+          req.headers.update(auth_headers)
+          req.params.update(params)
+        end
+        unwrap!(response)
+      end
+
+      def unwrap!(response)
+        unless response.success?
+          raise Error, "Proxmox API #{response.status}: #{response.body.inspect}"
+        end
+
+        body = response.body
+        body.is_a?(Hash) && body.key?("data") ? body["data"] : body
+      end
+    end
+  end
+end