Clean up task dependency and scheduler metrics (#10340)

src/ray/raylet/scheduling_queue.cc

@@ -504,13 +504,13 @@ std::string SchedulingQueue::DebugString() const {
 }
 
 void SchedulingQueue::RecordMetrics() const {
-  for (size_t i = 0; i < static_cast<int>(ray::raylet::TaskState::kNumTaskQueues); i++) {
-    TaskState task_state = static_cast<TaskState>(i);
-    stats::SchedulingQueueStats().Record(
-        static_cast<double>(GetTaskQueue(task_state)->GetTasks().size()),
-        {{stats::ValueTypeKey,
-          std::string("num_") + GetTaskStateString(task_state) + "_tasks"}});
-  }
+  stats::NumPlaceableTasks().Record(
+      GetTaskQueue(TaskState::PLACEABLE)->GetTasks().size());
+  stats::NumPlaceableTasks().Record(GetTaskQueue(TaskState::WAITING)->GetTasks().size());
+  stats::NumPlaceableTasks().Record(GetTaskQueue(TaskState::READY)->GetTasks().size());
+  stats::NumPlaceableTasks().Record(GetTaskQueue(TaskState::RUNNING)->GetTasks().size());
+  stats::NumPlaceableTasks().Record(
+      GetTaskQueue(TaskState::INFEASIBLE)->GetTasks().size());
 }
 
 }  // namespace raylet

src/ray/raylet/task_dependency_manager.cc

@@ -506,16 +506,10 @@ std::string TaskDependencyManager::DebugString() const {
 }
 
 void TaskDependencyManager::RecordMetrics() const {
-  stats::TaskDependencyManagerStats().Record(
-      task_dependencies_.size(), {{stats::ValueTypeKey, "num_task_dependencies"}});
-  stats::TaskDependencyManagerStats().Record(
-      required_tasks_.size(), {{stats::ValueTypeKey, "num_required_tasks"}});
-  stats::TaskDependencyManagerStats().Record(
-      required_objects_.size(), {{stats::ValueTypeKey, "num_required_objects"}});
-  stats::TaskDependencyManagerStats().Record(
-      local_objects_.size(), {{stats::ValueTypeKey, "num_local_objects"}});
-  stats::TaskDependencyManagerStats().Record(
-      pending_tasks_.size(), {{stats::ValueTypeKey, "num_pending_tasks"}});
+  stats::NumSubscribedTasks().Record(task_dependencies_.size());
+  stats::NumRequiredTasks().Record(required_tasks_.size());
+  stats::NumRequiredObjects().Record(required_objects_.size());
+  stats::NumPendingTasks().Record(pending_tasks_.size());
 }
 
 bool TaskDependencyManager::GetOwnerAddress(const ObjectID &object_id,

src/ray/stats/metric_defs.h

@@ -89,14 +89,38 @@ static Gauge ObjectManagerProfileEvents("object_manager_num_buffered_profile_eve
                                         "Number of locally-buffered profile events.",
                                         "events");
 
-static Gauge TaskDependencyManagerStats("task_dependency_manager_stats",
-                                        "Stat the metric values of task dependency.",
-                                        "pcs", {ValueTypeKey});
+static Gauge NumSubscribedTasks(
+    "num_subscribed_tasks",
+    "The number of tasks that are subscribed to object dependencies.", "tasks");
 
-static Gauge SchedulingQueueStats("scheduling_queue_stats",
-                                  "Stats the metric values of scheduling queue.", "pcs",
-                                  {ValueTypeKey});
+static Gauge NumRequiredTasks("num_required_tasks",
+                              "The number of tasks whose output object(s) are "
+                              "required by another subscribed task.",
+                              "tasks");
 
-static Gauge ConnectionPoolStats("connection_pool_stats",
-                                 "Stats the connection pool metrics.", "pcs",
-                                 {ValueTypeKey});
+static Gauge NumRequiredObjects(
+    "num_required_objects",
+    "The number of objects that are required by a subscribed task.", "objects");
+
+static Gauge NumPendingTasks("num_pending_tasks",
+                             "The number of tasks that are pending execution.", "tasks");
+
+static Gauge NumPlaceableTasks(
+    "num_placeable_tasks",
+    "The number of tasks in the scheduler that are in the 'placeable' state.", "tasks");
+
+static Gauge NumWaitingTasks(
+    "num_waiting_tasks",
+    "The number of tasks in the scheduler that are in the 'waiting' state.", "tasks");
+
+static Gauge NumReadyTasks(
+    "num_ready_tasks",
+    "The number of tasks in the scheduler that are in the 'ready' state.", "tasks");
+
+static Gauge NumRunningTasks(
+    "num_running_tasks",
+    "The number of tasks in the scheduler that are in the 'running' state.", "tasks");
+
+static Gauge NumInfeasibleTasks(
+    "num_infeasible_tasks",
+    "The number of tasks in the scheduler that are in the 'infeasible' state.", "tasks");

src/ray/stats/tag_defs.h

@@ -38,6 +38,4 @@ static const TagKeyType DriverPidKey = TagKeyType::Register("DriverPid");
 
 static const TagKeyType ResourceNameKey = TagKeyType::Register("ResourceName");
 
-static const TagKeyType ValueTypeKey = TagKeyType::Register("ValueType");
-
 static const TagKeyType ActorIdKey = TagKeyType::Register("ActorId");