Debug log.

jvm-packages/xgboost4j-spark/src/main/scala/ml/dmlc/xgboost4j/scala/spark/XGBoost.scala

@@ -565,6 +565,7 @@ object XGBoost extends XGBoostStageLevel {
         logger.error("the job was aborted due to ", t)
         throw t
     } finally {
+      Console.println("[td]: Call tracker shutdown.")
       optionalCachedRDD.foreach(_.unpersist())
     }
   }

jvm-packages/xgboost4j-spark/src/test/scala/ml/dmlc/xgboost4j/scala/spark/XGBoostGeneralSuite.scala

@@ -372,5 +372,4 @@ class XGBoostGeneralSuite extends AnyFunSuite with TmpFolderPerSuite with PerTes
       xgb.fit(trainingDF)
     }
   }
-
 }

src/collective/comm.cc

@@ -223,6 +223,7 @@ Comm* RabitComm::MakeCUDAVar(Context const*, std::shared_ptr<Coll>) const {
 
 [[nodiscard]] Result RabitComm::Bootstrap(std::chrono::seconds timeout, std::int32_t retry,
                                           std::string task_id) {
+  std::cout << "[comm]: bootstrap:" << getpid() << std::endl;
   TCPSocket tracker;
   std::int32_t world{-1};
   auto rc = ConnectTrackerImpl(this->TrackerInfo(), timeout, retry, task_id, &tracker, this->Rank(),
@@ -351,6 +352,7 @@ RabitComm::~RabitComm() noexcept(false) {
 }
 
 [[nodiscard]] Result RabitComm::Shutdown() {
+  std::cout << "[comm]: shutdown" << std::endl;
   if (!this->IsDistributed()) {
     return Success();
   }

src/collective/tracker.cc

@@ -307,6 +307,7 @@ Result RabitTracker::Bootstrap(std::vector<WorkerProxy>* p_workers) {
             // The worker can still send shutdown after call to `std::exit`.
             continue;
           }
+          std::cout << "[tracker]: shutdown" << std::endl;
           state.Shutdown();
           continue;
         }
@@ -344,6 +345,7 @@ Result RabitTracker::Bootstrap(std::vector<WorkerProxy>* p_workers) {
 }
 
 [[nodiscard]] Result RabitTracker::Stop() {
+  std::cout << "[tracker]: stop" << std::endl;
   if (!this->Ready()) {
     return Success();
   }