Fix race create sessione su CPX: ready solo dopo MediaMTX.

I nodi cloud restano in provisioning finché :9997 risponde; retry su create_path e 503 retryable se l’ingest è ancora irraggiungibile.

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

backend/app/services/streams/node_provisioner.rb

@@ -80,11 +80,15 @@ module Streams
       urls = urls_for_node(role: role, home: home, hostname: hostname, simulated: simulated,
                            private_ip: private_ip, public_ip: ip, use_node_hostname: use_node_hostname)
 
-      StreamNode.create!(
+      # Simulated/lab che riusa MediaMTX home: subito ready. Cloud reale: provisioning
+      # finché :9997 risponde (evita allocate → Faraday Connection refused → 500).
+      initial_status = simulated ? "ready" : "provisioning"
+
+      node = StreamNode.create!(
         slug: slug,
         hostname: hostname,
         role: role,
-        status: "ready",
+        status: initial_status,
         provider: provider_name_for(cloud, role: role),
         provider_instance_id: instance.id,
         rtmp_base_url: urls.fetch(:rtmp_base_url),
@@ -102,6 +106,34 @@ module Streams
           "cloud_raw" => instance.raw
         }
       )
+
+      wait_until_mediamtx_ready!(node) unless simulated
+      node.reload
+    end
+
+    def wait_until_mediamtx_ready!(node, timeout: nil, interval: nil)
+      timeout ||= ENV.fetch("STREAM_NODE_READY_TIMEOUT_SECS", "180").to_i
+      interval ||= ENV.fetch("STREAM_NODE_READY_POLL_SECS", "3").to_f
+      deadline = Process.clock_gettime(Process::CLOCK_MONOTONIC) + timeout
+
+      loop do
+        if Streams::NodeHealth.promote_if_healthy!(node)
+          Rails.logger.info("[Streams::NodeProvisioner] #{node.slug} MediaMTX ready")
+          return node
+        end
+
+        remaining = deadline - Process.clock_gettime(Process::CLOCK_MONOTONIC)
+        if remaining <= 0
+          Rails.logger.warn(
+            "[Streams::NodeProvisioner] #{node.slug} ancora provisioning dopo #{timeout}s " \
+            "(api=#{node.api_base_url}); autoscaler continuerà a promuovere"
+          )
+          return node
+        end
+
+        sleep([interval, remaining].min)
+        node.reload
+      end
     end
 
     def urls_for_node(role:, home:, hostname:, simulated:, private_ip:, public_ip: nil, use_node_hostname:)