[Core] Emit total lineage bytes metric (#46725)

Signed-off-by: Jiajun Yao <jeromeyjj@gmail.com>
ray-project · Jul 26, 2024 · f9b36da · f9b36da
1 parent 69f3218
commit f9b36da
Show file tree

Hide file tree

Showing 3 changed files with 13 additions and 0 deletions.
diff --git a/src/ray/core_worker/task_manager.cc b/src/ray/core_worker/task_manager.cc
@@ -1525,6 +1525,7 @@ void TaskManager::FillTaskInfo(rpc::GetCoreWorkerStatsReply *reply,
 
 void TaskManager::RecordMetrics() {
   absl::MutexLock lock(&mu_);
+  ray::stats::STATS_total_lineage_bytes.Record(total_lineage_footprint_bytes_);
   task_counter_.FlushOnChangeCallbacks();
 }
 

diff --git a/src/ray/stats/metric_defs.cc b/src/ray/stats/metric_defs.cc
@@ -357,6 +357,15 @@ DEFINE_stats(
     ("Type", "Name"),
     (),
     ray::stats::COUNT);
+
+/// Core Worker Task Manager
+DEFINE_stats(
+    total_lineage_bytes,
+    "Total amount of memory used to store task specs for lineage reconstruction.",
+    (),
+    (),
+    ray::stats::GAUGE);
+
 }  // namespace stats
 
 }  // namespace ray
diff --git a/src/ray/stats/metric_defs.h b/src/ray/stats/metric_defs.h
@@ -119,6 +119,9 @@ DECLARE_stats(gcs_actors_count);
 /// Memory Manager
 DECLARE_stats(memory_manager_worker_eviction_total);
 
+/// Core Worker Task Manager
+DECLARE_stats(total_lineage_bytes);
+
 /// The below items are legacy implementation of metrics.
 /// TODO(sang): Use DEFINE_stats instead.