Allow the node manager port and object manager port to be set through… (#3130)

* Allow the node manager port and object manager port to be set through ray start.

* Linting

* Fix Java test

* Address comments.

src/ray/object_manager/object_manager.h

@@ -29,6 +29,10 @@
 namespace ray {
 
 struct ObjectManagerConfig {
+  /// The port that the object manager should use to listen for connections
+  /// from other object managers. If this is 0, the object manager will choose
+  /// its own port.
+  int object_manager_port;
   /// The time in milliseconds to wait before retrying a pull
   /// that fails due to client id lookup.
   uint pull_timeout_ms;

src/ray/raylet/main.cc

@@ -20,19 +20,21 @@ int main(int argc, char *argv[]) {
                                          ray::RayLogLevel::INFO,
                                          /*log_dir=*/"");
   ray::RayLog::InstallFailureSignalHandler();
-  RAY_CHECK(argc == 11 || argc == 12);
+  RAY_CHECK(argc == 13 || argc == 14);
 
   const std::string raylet_socket_name = std::string(argv[1]);
   const std::string store_socket_name = std::string(argv[2]);
-  const std::string node_ip_address = std::string(argv[3]);
-  const std::string redis_address = std::string(argv[4]);
-  int redis_port = std::stoi(argv[5]);
-  int num_initial_workers = std::stoi(argv[6]);
-  int maximum_startup_concurrency = std::stoi(argv[7]);
-  const std::string static_resource_list = std::string(argv[8]);
-  const std::string python_worker_command = std::string(argv[9]);
-  const std::string java_worker_command = std::string(argv[10]);
-  const std::string redis_password = (argc == 12 ? std::string(argv[11]) : "");
+  int object_manager_port = std::stoi(argv[3]);
+  int node_manager_port = std::stoi(argv[4]);
+  const std::string node_ip_address = std::string(argv[5]);
+  const std::string redis_address = std::string(argv[6]);
+  int redis_port = std::stoi(argv[7]);
+  int num_initial_workers = std::stoi(argv[8]);
+  int maximum_startup_concurrency = std::stoi(argv[9]);
+  const std::string static_resource_list = std::string(argv[10]);
+  const std::string python_worker_command = std::string(argv[11]);
+  const std::string java_worker_command = std::string(argv[12]);
+  const std::string redis_password = (argc == 14 ? std::string(argv[13]) : "");
 
   // Configuration for the node manager.
   ray::raylet::NodeManagerConfig node_manager_config;
@@ -51,6 +53,7 @@ int main(int argc, char *argv[]) {
       ray::raylet::ResourceSet(std::move(static_resource_conf));
   RAY_LOG(DEBUG) << "Starting raylet with static resource configuration: "
                  << node_manager_config.resource_config.ToString();
+  node_manager_config.node_manager_port = node_manager_port;
   node_manager_config.num_initial_workers = num_initial_workers;
   node_manager_config.num_workers_per_process =
       RayConfig::instance().num_workers_per_process();
@@ -76,6 +79,7 @@ int main(int argc, char *argv[]) {
 
   // Configuration for the object manager.
   ray::ObjectManagerConfig object_manager_config;
+  object_manager_config.object_manager_port = object_manager_port;
   object_manager_config.store_socket_name = store_socket_name;
   object_manager_config.pull_timeout_ms =
       RayConfig::instance().object_manager_pull_timeout_ms();

src/ray/raylet/node_manager.h

@@ -23,15 +23,23 @@ namespace ray {
 namespace raylet {
 
 struct NodeManagerConfig {
+  /// The node's resource configuration.
   ResourceSet resource_config;
+  /// The port to use for listening to incoming connections. If this is 0 then
+  /// the node manager will choose its own port.
+  int node_manager_port;
+  /// The initial number of workers to create.
   int num_initial_workers;
+  /// The number of workers per process.
   int num_workers_per_process;
   /// The maximum number of workers that can be started concurrently by a
   /// worker pool.
   int maximum_startup_concurrency;
   /// The commands used to start the worker process, grouped by language.
   std::unordered_map<Language, std::vector<std::string>> worker_commands;
+  /// The time between heartbeats in milliseconds.
   uint64_t heartbeat_period_ms;
+  /// the maximum lineage size.
   uint64_t max_lineage_size;
   /// The store socket name.
   std::string store_socket_name;

src/ray/raylet/raylet.cc

@@ -24,10 +24,13 @@ Raylet::Raylet(boost::asio::io_service &main_service, const std::string &socket_
       acceptor_(main_service, boost::asio::local::stream_protocol::endpoint(socket_name)),
       socket_(main_service),
       object_manager_acceptor_(
-          main_service, boost::asio::ip::tcp::endpoint(boost::asio::ip::tcp::v4(), 0)),
+          main_service,
+          boost::asio::ip::tcp::endpoint(boost::asio::ip::tcp::v4(),
+                                         object_manager_config.object_manager_port)),
       object_manager_socket_(main_service),
-      node_manager_acceptor_(
-          main_service, boost::asio::ip::tcp::endpoint(boost::asio::ip::tcp::v4(), 0)),
+      node_manager_acceptor_(main_service, boost::asio::ip::tcp::endpoint(
+                                               boost::asio::ip::tcp::v4(),
+                                               node_manager_config.node_manager_port)),
       node_manager_socket_(main_service) {
   // Start listening for clients.
   DoAccept();