Completa il hardening autoscaler (Fase 4): budget, kill-switch e runbook.

Drain sicuro in scale-in, alert Ops su overflow e controlli admin senza abilitare il deploy in produzione.

Co-authored-by: Cursor <cursoragent@cursor.com>
This commit is contained in:
2026-08-09 20:00:15 +02:00
co-authored by Cursor
parent 318a319608
commit e5fc925bae
15 changed files with 286 additions and 14 deletions
@@ -15,6 +15,7 @@ RSpec.describe Streams::Autoscaler do
before do
redis.del(Streams::Autoscaler::LOCK_KEY)
redis.del(Streams::Autoscaler::KILL_SWITCH_KEY)
redis.del(Streams::DnsProviders::Lab::REDIS_KEY)
end
@@ -116,10 +117,55 @@ RSpec.describe Streams::Autoscaler do
)
provisioner = instance_double(Streams::NodeProvisioner)
allow(provisioner).to receive(:drain!)
expect(provisioner).to receive(:decommission!).with(idle)
result = described_class.reconcile!(provisioner: provisioner)
expect(result.actions.map(&:to_s)).to include("scale_in_ingest-lab-99")
end
end
it "skips when Redis kill-switch is engaged even if ENV is enabled" do
redis.del(described_class::KILL_SWITCH_KEY)
with_env("STREAM_AUTOSCALE_ENABLED" => "1") do
described_class.engage_kill_switch!
expect(described_class.enabled?).to eq(false)
result = described_class.reconcile!
expect(result.skipped).to eq(true)
ensure
described_class.clear_kill_switch!
end
end
it "blocks scale-out when next node would exceed monthly budget" do
with_env(
"STREAM_AUTOSCALE_ENABLED" => "1",
"STREAM_AUTOSCALE_SOFT_FREE_SLOTS" => "2",
"STREAM_AUTOSCALE_WARM_SPARE" => "0",
"STREAM_AUTOSCALE_KIND" => "lab",
"STREAM_AUTOSCALE_MAX_NODES" => "5",
"STREAM_AUTOSCALE_NODE_EUR_PER_HOUR" => "1",
"STREAM_AUTOSCALE_MONTHLY_BUDGET_EUR" => "10",
"STREAM_CLOUD_PROVIDER" => "local_lab",
"STREAM_DNS_PROVIDER" => "lab",
"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
home = Streams::NodeRegistry.ensure_home_from_env!
user = User.create!(email: "budget@example.com", name: "B", password: "Password123", role: "coach")
club = Club.create!(name: "Budget", sport: "volleyball")
team = club.teams.create!(name: "T", sport: "volleyball", slug: "budget-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(described_class.metrics[:within_budget]).to eq(true)
end
end
end