HLS su edge e monitoraggio ops con latenza p95.

Evita di saturare Puma in diretta, separa i check Rails/pubblico e traccia la latenza /up per alert più affidabili.

Co-authored-by: Cursor <cursoragent@cursor.com>

backend/app/services/ops/up_latency_tracker.rb

@@ -0,0 +1,44 @@
+# frozen_string_literal: true
+
+module Ops
+  # Campioni latenza GET /up (edge → rails) per calcolo p95.
+  class UpLatencyTracker
+    REDIS_KEY = "ops:up_latency_ms"
+    MAX_SAMPLES = -> { ENV.fetch("OPS_UP_LATENCY_SAMPLES", "60").to_i }
+
+    class << self
+      def record(latency_ms)
+        return unless redis
+
+        redis.pipelined do |pipe|
+          pipe.lpush(REDIS_KEY, latency_ms.to_i)
+          pipe.ltrim(REDIS_KEY, 0, MAX_SAMPLES.call - 1)
+        end
+      end
+
+      def samples
+        return [] unless redis
+
+        redis.lrange(REDIS_KEY, 0, -1).map(&:to_i)
+      end
+
+      def sample_count
+        samples.size
+      end
+
+      def p95
+        sorted = samples.sort
+        return nil if sorted.empty?
+
+        index = [(sorted.size * 0.95).ceil - 1, 0].max
+        sorted[index]
+      end
+
+      def redis
+        @redis ||= Redis.new(url: ENV.fetch("REDIS_URL", "redis://localhost:6379/0"))
+      rescue Redis::CannotConnectError
+        nil
+      end
+    end
+  end
+end