[Core] Make worker_register_timeout_seconds configurable (#9221)

python/ray/includes/ray_config.pxd

@@ -47,6 +47,8 @@ cdef extern from "ray/common/ray_config.h" nogil:
 
         int64_t kill_worker_timeout_milliseconds() const
 
+        int64_t worker_register_timeout_seconds() const
+
         int64_t max_time_for_handler_milliseconds() const
 
         int64_t max_time_for_loop() const

python/ray/includes/ray_config.pxi

@@ -56,7 +56,8 @@ cdef class Config:
 
     @staticmethod
     def raylet_client_connect_timeout_milliseconds():
-        return RayConfig.instance().raylet_client_connect_timeout_milliseconds()
+        return (RayConfig.instance()
+                .raylet_client_connect_timeout_milliseconds())
 
     @staticmethod
     def raylet_fetch_timeout_milliseconds():
@@ -80,6 +81,10 @@ cdef class Config:
     def kill_worker_timeout_milliseconds():
         return RayConfig.instance().kill_worker_timeout_milliseconds()
 
+    @staticmethod
+    def worker_register_timeout_seconds():
+        return RayConfig.instance().worker_register_timeout_seconds()
+
     @staticmethod
     def max_time_for_handler_milliseconds():
         return RayConfig.instance().max_time_for_handler_milliseconds()

src/ray/common/ray_config_def.h

@@ -192,6 +192,10 @@ RAY_CONFIG(size_t, raylet_max_active_object_ids, 0)
 /// the worker SIGKILL.
 RAY_CONFIG(int64_t, kill_worker_timeout_milliseconds, 100)
 
+/// The duration that we wait after the worekr is launched before the
+/// starting_worker_timeout_callback() is called.
+RAY_CONFIG(int64_t, worker_register_timeout_seconds, 30)
+
 /// This is a timeout used to cause failures in the plasma manager and raylet
 /// when certain event loop handlers take too long.
 RAY_CONFIG(int64_t, max_time_for_handler_milliseconds, 1000)

src/ray/raylet/worker_pool.cc

@@ -250,9 +250,9 @@ Process WorkerPool::StartWorkerProcess(const Language &language,
 
 void WorkerPool::MonitorStartingWorkerProcess(const Process &proc,
                                               const Language &language) {
-  constexpr static size_t worker_register_timeout_seconds = 30;
   auto timer = std::make_shared<boost::asio::deadline_timer>(
-      *io_service_, boost::posix_time::seconds(worker_register_timeout_seconds));
+      *io_service_, boost::posix_time::seconds(
+                        RayConfig::instance().worker_register_timeout_seconds()));
   // Capture timer in lambda to copy it once, so that it can avoid destructing timer.
   timer->async_wait(
       [timer, language, proc, this](const boost::system::error_code e) -> void {