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>
159 lines
4.6 KiB
Ruby
159 lines
4.6 KiB
Ruby
# 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
|