Alza soft capacity fase A e aggiunge quiet hours CPX notturne.

Home a 4 e cloud a 6 con MAX_NODES=12 (~76 soft); di notte (02–07 Europe/Rome) niente scale-out/warm-spare e sweeper che chiude i CPX idle. Co-authored-by: Cursor <cursoragent@cursor.com>
2026-09-05 09:48:50 +02:00
parent d52434fb2e
commit 7c7b2bf14c
16 changed files with 513 additions and 25 deletions
@@ -19,6 +19,11 @@ RSpec.describe Streams::Autoscaler do
    redis.del(Streams::DnsProviders::Lab::REDIS_KEY)
    StreamSession.where.not(stream_node_id: nil).update_all(stream_node_id: nil, status: "ended", ended_at: Time.current)
    StreamNode.where.not(slug: Streams::NodeRegistry::HOME_SLUG).delete_all
+    ENV["STREAM_AUTOSCALE_QUIET_HOURS"] = "off"
+  end
+
+  after do
+    ENV.delete("STREAM_AUTOSCALE_QUIET_HOURS")
  end

  it "is a no-op when disabled" do
@@ -261,4 +266,34 @@ RSpec.describe Streams::Autoscaler do
      expect(node.reload.status).to eq("ready")
    end
  end
+
+  it "blocks scale-out and warm spare during quiet hours" do
+    with_env(
+      "STREAM_AUTOSCALE_ENABLED" => "1",
+      "STREAM_AUTOSCALE_QUIET_HOURS" => "02:00-07:00",
+      "STREAM_AUTOSCALE_SOFT_FREE_SLOTS" => "2",
+      "STREAM_AUTOSCALE_WARM_SPARE" => "1",
+      "STREAM_AUTOSCALE_KIND" => "lab",
+      "STREAM_AUTOSCALE_MAX_NODES" => "5",
+      "STREAM_NODE_HOME_MAX_PUBLISHERS" => "1",
+      "MEDIAMTX_API_URL" => "http://mtx-home:9997",
+      "MEDIAMTX_RTMP_URL" => "rtmp://home.example:1935",
+      "HLS_PUBLIC_URL" => "https://home.example/hls"
+    ) do
+      allow(Streams::QuietHours).to receive(:active?).and_return(true)
+      home = Streams::NodeRegistry.ensure_home_from_env!
+      user = User.create!(email: "qh@example.com", name: "Q", password: "Password123", role: "coach")
+      club = Club.create!(name: "QH", sport: "volleyball")
+      team = club.teams.create!(name: "T", sport: "volleyball", slug: "qh-t")
+      match = team.matches.create!(opponent_name: "X", scheduled_at: 1.hour.from_now)
+      StreamSession.create!(match: match, user: user, platform: "matchlivetv", status: "live", stream_node: home)
+
+      provisioner = instance_double(Streams::NodeProvisioner)
+      expect(provisioner).not_to receive(:provision_lab!)
+
+      result = described_class.reconcile!(provisioner: provisioner)
+      expect(result.actions).to include(:blocked_capacity)
+      expect(result.metrics[:quiet_hours]).to eq(true)
+    end
+  end
 end
@@ -0,0 +1,114 @@
+# frozen_string_literal: true
+
+require "rails_helper"
+
+RSpec.describe Streams::NightCloudSweeper do
+  def with_env(vars)
+    previous = vars.keys.index_with { |k| ENV[k] }
+    vars.each { |k, v| ENV[k] = v }
+    yield
+  ensure
+    previous.each { |k, v| v.nil? ? ENV.delete(k) : ENV[k] = v }
+  end
+
+  def cloud_node!(slug: "ingest-cloud-01")
+    StreamNode.create!(
+      slug: slug,
+      hostname: "#{slug}.mltv-stream.net",
+      role: "cloud",
+      status: "ready",
+      provider: "hetzner",
+      provider_instance_id: "42",
+      rtmp_base_url: "rtmp://#{slug}:1935",
+      hls_base_url: "https://#{slug}/hls",
+      api_base_url: "http://10.0.0.9:9997",
+      max_publishers: 6,
+      max_relays: 6,
+      metadata: { "public_ip" => "1.2.3.4" }
+    )
+  end
+
+  before do
+    StreamSession.where.not(stream_node_id: nil).update_all(stream_node_id: nil, status: "ended", ended_at: Time.current)
+    StreamNode.where(role: "cloud").delete_all
+  end
+
+  it "skips outside quiet hours" do
+    with_env(
+      "STREAM_NIGHT_SWEEP_ENABLED" => "1",
+      "STREAM_AUTOSCALE_QUIET_HOURS" => "02:00-07:00",
+      "STREAM_AUTOSCALE_QUIET_TZ" => "Europe/Rome"
+    ) do
+      allow(Streams::QuietHours).to receive(:active?).and_return(false)
+      cloud_node!
+      result = described_class.sweep!
+      expect(result.skipped).to eq(true)
+      expect(StreamNode.where(role: "cloud").count).to eq(1)
+    end
+  end
+
+  it "decommissions idle cloud nodes during quiet hours" do
+    with_env("STREAM_NIGHT_SWEEP_ENABLED" => "1") do
+      allow(Streams::QuietHours).to receive(:active?).and_return(true)
+      node = cloud_node!
+      provisioner = instance_double(Streams::NodeProvisioner)
+      allow(provisioner).to receive(:drain!) do |n|
+        n.update!(status: "draining")
+        n
+      end
+      allow(provisioner).to receive(:decommission!) do |n|
+        n.destroy!
+        true
+      end
+
+      result = described_class.sweep!(provisioner: provisioner)
+      expect(result.skipped).to eq(false)
+      expect(result.actions).to include(:"decommission_#{node.slug}")
+      expect(StreamNode.find_by(id: node.id)).to be_nil
+    end
+  end
+
+  it "alerts and keeps nodes with active sessions" do
+    with_env("STREAM_NIGHT_SWEEP_ENABLED" => "1") do
+      allow(Streams::QuietHours).to receive(:active?).and_return(true)
+      node = cloud_node!
+      user = User.create!(email: "night@example.com", name: "N", password: "Password123", role: "coach")
+      club = Club.create!(name: "Night", sport: "volleyball")
+      team = club.teams.create!(name: "T", sport: "volleyball", slug: "night-t")
+      match = team.matches.create!(opponent_name: "X", scheduled_at: 1.hour.from_now)
+      StreamSession.create!(match: match, user: user, platform: "matchlivetv", status: "live", stream_node: node)
+
+      allow(Ops::IncidentRecorder).to receive(:record)
+      provisioner = instance_double(Streams::NodeProvisioner)
+      expect(provisioner).not_to receive(:decommission!)
+
+      result = described_class.sweep!(provisioner: provisioner)
+      expect(result.actions).to include(:"alert_active_#{node.slug}")
+      expect(Ops::IncidentRecorder).to have_received(:record).with(hash_including(
+        fingerprint: "stream_overflow:night_active:#{node.slug}"
+      ))
+      expect(node.reload.status).to eq("ready")
+    end
+  end
+
+  it "does not touch lab nodes" do
+    with_env("STREAM_NIGHT_SWEEP_ENABLED" => "1") do
+      allow(Streams::QuietHours).to receive(:active?).and_return(true)
+      StreamNode.create!(
+        slug: "ingest-lab-night-01",
+        hostname: "lab-night.local",
+        role: "lab",
+        status: "ready",
+        provider: "local",
+        rtmp_base_url: "rtmp://lab:1935",
+        hls_base_url: "https://lab/hls",
+        api_base_url: "http://lab:9997",
+        max_publishers: 2,
+        max_relays: 2
+      )
+      result = described_class.sweep!
+      expect(result.actions).to eq([])
+      expect(StreamNode.find_by(slug: "ingest-lab-night-01")).to be_present
+    end
+  end
+end
@@ -0,0 +1,40 @@
+# frozen_string_literal: true
+
+require "rails_helper"
+
+RSpec.describe Streams::QuietHours do
+  def with_env(vars)
+    previous = vars.keys.index_with { |k| ENV[k] }
+    vars.each { |k, v| ENV[k] = v }
+    yield
+  ensure
+    previous.each { |k, v| v.nil? ? ENV.delete(k) : ENV[k] = v }
+  end
+
+  it "is active inside the Rome window" do
+    with_env(
+      "STREAM_AUTOSCALE_QUIET_HOURS" => "02:00-07:00",
+      "STREAM_AUTOSCALE_QUIET_TZ" => "Europe/Rome"
+    ) do
+      now = Time.find_zone!("Europe/Rome").local(2026, 9, 5, 3, 30)
+      expect(described_class.active?(now: now)).to eq(true)
+    end
+  end
+
+  it "is inactive outside the window" do
+    with_env(
+      "STREAM_AUTOSCALE_QUIET_HOURS" => "02:00-07:00",
+      "STREAM_AUTOSCALE_QUIET_TZ" => "Europe/Rome"
+    ) do
+      now = Time.find_zone!("Europe/Rome").local(2026, 9, 5, 10, 0)
+      expect(described_class.active?(now: now)).to eq(false)
+    end
+  end
+
+  it "can be disabled with off" do
+    with_env("STREAM_AUTOSCALE_QUIET_HOURS" => "off") do
+      now = Time.find_zone!("Europe/Rome").local(2026, 9, 5, 3, 0)
+      expect(described_class.active?(now: now)).to eq(false)
+    end
+  end
+end