Aggiunge l'autoscaler streaming con warm spare e kill-switch.

Scala i nodi overflow quando gli slot calano, mantiene una spare a caldo sotto carico e spegne gli idle, disattivabile con STREAM_AUTOSCALE_ENABLED. Co-authored-by: Cursor <cursoragent@cursor.com>
2026-08-09 19:56:58 +02:00
parent 0b55d5a8fa
commit 318a319608
10 changed files with 354 additions and 1 deletions
@@ -9,6 +9,7 @@ module Admin
      @cloud_provider = ENV.fetch("STREAM_CLOUD_PROVIDER", "local_lab")
      @lab_hosts = lab_dns_snippet
      @hetzner_configured = ENV["HCLOUD_TOKEN"].present?
+      @autoscale_metrics = Streams::Autoscaler.metrics
    end

    def create
@@ -0,0 +1,33 @@
+# frozen_string_literal: true
+
+module Streams
+  class AutoscalerJob
+    include Sidekiq::Job
+
+    sidekiq_options retry: 1, queue: "default"
+
+    INTERVAL_SECS = ENV.fetch("STREAM_AUTOSCALE_INTERVAL_SECS", "60").to_i
+    REDIS_CHAIN_KEY = "streams:autoscaler:chain"
+
+    def self.ensure_chain
+      return unless redis
+      return if redis.get(REDIS_CHAIN_KEY)
+
+      redis.set(REDIS_CHAIN_KEY, "1", ex: INTERVAL_SECS * 2)
+      perform_in(INTERVAL_SECS)
+    end
+
+    def self.redis
+      @redis ||= Redis.new(url: ENV.fetch("REDIS_URL", "redis://localhost:6379/0"))
+    rescue Redis::CannotConnectError
+      nil
+    end
+
+    def perform
+      Streams::Autoscaler.reconcile!
+    ensure
+      self.class.redis&.set(REDIS_CHAIN_KEY, "1", ex: INTERVAL_SECS * 2)
+      self.class.perform_in(INTERVAL_SECS)
+    end
+  end
+end
@@ -0,0 +1,158 @@
+# frozen_string_literal: true
+
+module Streams
+  # Scale-out / warm spare / scale-in dei nodi overflow (lab o Hetzner).
+  # Kill-switch: STREAM_AUTOSCALE_ENABLED!=1 → no-op.
+  class Autoscaler
+    Result = Struct.new(:actions, :metrics, :skipped, :error, keyword_init: true)
+
+    LOCK_KEY = "streams:autoscaler:lock"
+
+    class << self
+      def enabled?
+        ENV["STREAM_AUTOSCALE_ENABLED"] == "1"
+      end
+
+      def soft_free_slots
+        ENV.fetch("STREAM_AUTOSCALE_SOFT_FREE_SLOTS", "2").to_i
+      end
+
+      def warm_spare_min
+        ENV.fetch("STREAM_AUTOSCALE_WARM_SPARE", "1").to_i
+      end
+
+      def idle_minutes
+        ENV.fetch("STREAM_AUTOSCALE_IDLE_MINUTES", "30").to_i
+      end
+
+      def max_overflow_nodes
+        ENV.fetch("STREAM_AUTOSCALE_MAX_NODES", "5").to_i
+      end
+
+      def kind
+        ENV.fetch("STREAM_AUTOSCALE_KIND", "lab") # lab|cloud
+      end
+
+      def reconcile!(provisioner: nil)
+        return Result.new(skipped: true, actions: [], metrics: metrics) unless enabled?
+
+        redis = Redis.new(url: ENV.fetch("REDIS_URL", "redis://localhost:6379/0"))
+        unless redis.set(LOCK_KEY, worker_id, nx: true, ex: 55)
+          return Result.new(skipped: true, actions: [], metrics: metrics, error: "locked")
+        end
+
+        begin
+          new(provisioner: provisioner).reconcile!
+        ensure
+          redis.del(LOCK_KEY)
+        end
+      end
+
+      def metrics
+        NodeRegistry.ensure_home_from_env!
+        nodes = StreamNode.ready.to_a
+        overflow = StreamNode.where.not(slug: NodeRegistry::HOME_SLUG)
+          .where(status: %w[ready draining provisioning]).to_a
+        {
+          free_slots: nodes.sum(&:free_slots),
+          ready_nodes: nodes.size,
+          spare_ready: nodes.count { |n| n.slug != NodeRegistry::HOME_SLUG && n.active_publishers.zero? },
+          overflow_nodes: overflow.size,
+          soft_free_slots: soft_free_slots,
+          warm_spare_min: warm_spare_min,
+          max_overflow_nodes: max_overflow_nodes,
+          enabled: enabled?,
+          kind: kind
+        }
+      end
+
+      def worker_id
+        ENV.fetch("HOSTNAME", "autoscaler")
+      end
+    end
+
+    def initialize(provisioner: nil)
+      @provisioner = provisioner || NodeProvisioner.new
+    end
+
+    def reconcile!
+      actions = []
+      m = self.class.metrics
+
+      if need_capacity?(m) && can_provision?(m)
+        provision_overflow!
+        actions << :scale_out
+        m = self.class.metrics
+      end
+
+      if warm_spare_desired?(m) && m[:spare_ready] < self.class.warm_spare_min && can_provision?(m)
+        provision_overflow!
+        actions << :warm_spare
+        m = self.class.metrics
+      end
+
+      scale_in_candidates.each do |node|
+        next if keep_as_warm_spare?(node)
+
+        @provisioner.decommission!(node)
+        actions << :"scale_in_#{node.slug}"
+        m = self.class.metrics
+      rescue NodeProvisioner::BusyError, NodeProvisioner::Error => e
+        Rails.logger.warn("[Streams::Autoscaler] scale-in #{node.slug}: #{e.message}")
+      end
+
+      Rails.logger.info("[Streams::Autoscaler] actions=#{actions.inspect} metrics=#{m.inspect}")
+      Result.new(actions: actions, metrics: m, skipped: false)
+    end
+
+    private
+
+    def need_capacity?(m)
+      m[:free_slots] <= self.class.soft_free_slots
+    end
+
+    def warm_spare_desired?(m)
+      return false if self.class.warm_spare_min <= 0
+
+      need_capacity?(m) || overflow_in_use?
+    end
+
+    def overflow_in_use?
+      StreamNode.ready.where.not(slug: NodeRegistry::HOME_SLUG).any? { |n| n.active_publishers.positive? }
+    end
+
+    def can_provision?(m)
+      m[:overflow_nodes] < self.class.max_overflow_nodes
+    end
+
+    def provision_overflow!
+      case self.class.kind
+      when "cloud" then @provisioner.provision_cloud!
+      else @provisioner.provision_lab!
+      end
+    end
+
+    def scale_in_candidates
+      StreamNode.where.not(slug: NodeRegistry::HOME_SLUG)
+        .where(status: %w[ready draining])
+        .order(:created_at)
+        .select { |n| n.active_publishers.zero? && idle_long_enough?(n) }
+    end
+
+    def idle_long_enough?(node)
+      idle_since(node) <= self.class.idle_minutes.minutes.ago
+    end
+
+    def idle_since(node)
+      last_end = node.stream_sessions.where(status: %w[ended error]).maximum(:ended_at)
+      last_end || node.created_at
+    end
+
+    def keep_as_warm_spare?(node)
+      return false unless warm_spare_desired?(self.class.metrics)
+
+      spares = StreamNode.ready.where.not(slug: NodeRegistry::HOME_SLUG).select { |n| n.active_publishers.zero? }
+      spares.size <= self.class.warm_spare_min && spares.map(&:id).include?(node.id)
+    end
+  end
+end
@@ -4,6 +4,20 @@
  <%= t("admin.stream_nodes.providers", cloud: @cloud_provider, dns: @dns_provider) %>
 </p>

+<p class="admin-muted">
+  <%= t(
+        "admin.stream_nodes.autoscale",
+        enabled: (@autoscale_metrics[:enabled] ? "ON" : "OFF"),
+        free: @autoscale_metrics[:free_slots],
+        soft: @autoscale_metrics[:soft_free_slots],
+        spare: @autoscale_metrics[:spare_ready],
+        warm: @autoscale_metrics[:warm_spare_min],
+        overflow: @autoscale_metrics[:overflow_nodes],
+        max: @autoscale_metrics[:max_overflow_nodes],
+        kind: @autoscale_metrics[:kind]
+      ) %>
+</p>
+
 <p>
  <%= button_to t("admin.stream_nodes.provision_lab"), admin_stream_nodes_path, method: :post, params: { kind: "lab" }, class: "admin-btn" %>
  <% if @hetzner_configured %>