[Dashboard] Optimize dashboard datacenter (#11391)

* Optimize dashboard datacenter

* Fix tests

* Fix tests

* Fix

* Fix CI

* python/build-wheel-macos.sh

Co-authored-by: 刘宝 <po.lb@antfin.com>
Co-authored-by: Max Fitton <maxfitton@anyscale.com>

python/build-wheel-macos.sh

@@ -13,6 +13,7 @@ MACPYTHON_URL=https://www.python.org/ftp/python
 MACPYTHON_PY_PREFIX=/Library/Frameworks/Python.framework/Versions
 DOWNLOAD_DIR=python_downloads
 
+NODE_VERSION="14"
 PY_VERSIONS=("3.6.1"
              "3.7.0"
              "3.8.2")
@@ -36,6 +37,7 @@ mkdir -p .whl
 
 # Use the latest version of Node.js in order to build the dashboard.
 source "$HOME"/.nvm/nvm.sh
+nvm install $NODE_VERSION
 nvm use node
 
 # Build the dashboard so its static assets can be included in the wheel.

python/ray/dashboard/dashboard.py

@@ -81,7 +81,7 @@ class DashboardController(BaseDashboardController):
     def _construct_raylet_info(self):
         D = self.raylet_stats.get_raylet_stats()
         workers_info_by_node = {
-            data["nodeId"]: data.get("workersStats")
+            data["nodeId"]: data.get("coreWorkersStats")
             for data in D.values()
         }
 

python/ray/dashboard/node_stats.py

@@ -159,20 +159,18 @@ class NodeStats(threading.Thread):
                 actors[actor_id].update(self._addr_to_extra_info_dict[addr])
 
             for node_id, workers_info in workers_info_by_node.items():
-                for worker_info in workers_info:
-                    if "coreWorkerStats" in worker_info:
-                        core_worker_stats = worker_info["coreWorkerStats"]
-                        addr = (core_worker_stats["ipAddress"],
-                                str(core_worker_stats["port"]))
-                        if addr in self._addr_to_actor_id:
-                            actor_info = actors[self._addr_to_actor_id[addr]]
-                            format_reply_id(core_worker_stats)
-                            actor_info.update(core_worker_stats)
-                            actor_info["averageTaskExecutionSpeed"] = round(
-                                actor_info["numExecutedTasks"] /
-                                (now - actor_info["timestamp"] / 1000), 2)
-                            actor_info["nodeId"] = node_id
-                            actor_info["pid"] = worker_info["pid"]
+                for core_worker_stats in workers_info:
+                    addr = (core_worker_stats["ipAddress"],
+                            str(core_worker_stats["port"]))
+                    if addr in self._addr_to_actor_id:
+                        actor_info = actors[self._addr_to_actor_id[addr]]
+                        format_reply_id(core_worker_stats)
+                        actor_info.update(core_worker_stats)
+                        actor_info["averageTaskExecutionSpeed"] = round(
+                            actor_info["numExecutedTasks"] /
+                            (now - actor_info["timestamp"] / 1000), 2)
+                        actor_info["nodeId"] = node_id
+                        actor_info["pid"] = core_worker_stats["pid"]
 
             def _update_from_actor_tasks(task, task_spec_type,
                                          invalid_state_type):
@@ -183,8 +181,9 @@ class NodeStats(threading.Thread):
                 elif invalid_state_type == "infeasibleActor":
                     task["state"] = -2
                 else:
-                    raise ValueError(f"Invalid argument"
-                                     "invalid_state_type={invalid_state_type}")
+                    raise ValueError(
+                        "Invalid argument"
+                        f"invalid_state_type={invalid_state_type}")
                 task["actorTitle"] = task["functionDescriptor"][
                     "pythonFunctionDescriptor"]["className"]
                 format_reply_id(task)

python/ray/tests/test_metrics.py

@@ -4,6 +4,7 @@ import requests
 import time
 
 import ray
+from ray.core.generated import common_pb2
 from ray.core.generated import node_manager_pb2
 from ray.core.generated import node_manager_pb2_grpc
 from ray.test_utils import (RayTestTimeoutException,
@@ -36,7 +37,10 @@ def test_worker_stats(shutdown_only):
 
     reply = try_get_node_stats()
     # Check that there is one connected driver.
-    drivers = [worker for worker in reply.workers_stats if worker.is_driver]
+    drivers = [
+        worker for worker in reply.core_workers_stats
+        if worker.worker_type == common_pb2.DRIVER
+    ]
     assert len(drivers) == 1
     assert os.getpid() == drivers[0].pid
 
@@ -58,11 +62,10 @@ def test_worker_stats(shutdown_only):
     worker_pid = ray.get(f.remote())
     reply = try_get_node_stats()
     target_worker_present = False
-    for worker in reply.workers_stats:
-        stats = worker.core_worker_stats
+    for stats in reply.core_workers_stats:
         if stats.webui_display[""] == '{"message": "test", "dtype": "text"}':
             target_worker_present = True
-            assert worker.pid == worker_pid
+            assert stats.pid == worker_pid
         else:
             assert stats.webui_display[""] == ""  # Empty proto
     assert target_worker_present
@@ -72,11 +75,10 @@ def test_worker_stats(shutdown_only):
     worker_pid = ray.get(a.f.remote())
     reply = try_get_node_stats()
     target_worker_present = False
-    for worker in reply.workers_stats:
-        stats = worker.core_worker_stats
+    for stats in reply.core_workers_stats:
         if stats.webui_display[""] == '{"message": "test", "dtype": "text"}':
             target_worker_present = True
-            assert worker.pid == worker_pid
+            assert stats.pid == worker_pid
         else:
             assert stats.webui_display[""] == ""  # Empty proto
     assert target_worker_present
@@ -89,15 +91,15 @@ def test_worker_stats(shutdown_only):
                 "Timed out while waiting for worker processes")
 
         # Wait for the workers to start.
-        if len(reply.workers_stats) < num_cpus + 1:
+        if len(reply.core_workers_stats) < num_cpus + 1:
             time.sleep(1)
             reply = try_get_node_stats()
             continue
 
         # Check that the rest of the processes are workers, 1 for each CPU.
-        assert len(reply.workers_stats) == num_cpus + 1
+        assert len(reply.core_workers_stats) == num_cpus + 1
         # Check that all processes are Python.
-        pids = [worker.pid for worker in reply.workers_stats]
+        pids = [worker.pid for worker in reply.core_workers_stats]
         processes = [
             p.info["name"] for p in psutil.process_iter(attrs=["pid", "name"])
             if p.info["pid"] in pids

python/ray/tests/test_multi_tenancy.py

@@ -8,6 +8,7 @@ import pytest
 
 import ray
 import ray.test_utils
+from ray.core.generated import common_pb2
 from ray.core.generated import node_manager_pb2, node_manager_pb2_grpc
 from ray.test_utils import (wait_for_condition, wait_for_pid_to_exit,
                             run_string_as_driver,
@@ -22,8 +23,8 @@ def get_workers():
     stub = node_manager_pb2_grpc.NodeManagerServiceStub(channel)
     return [
         worker for worker in stub.GetNodeStats(
-            node_manager_pb2.GetNodeStatsRequest()).workers_stats
-        if not worker.is_driver
+            node_manager_pb2.GetNodeStatsRequest()).core_workers_stats
+        if worker.worker_type != common_pb2.DRIVER
     ]