[coll] Keep the tracker alive during initialization error. (dmlc#10306)

trivialfis · web-flow · commit 966dc8178819 · 2024-05-23T11:13:59.000+08:00
diff --git a/src/collective/tracker.cc b/src/collective/tracker.cc
@@ -286,7 +286,8 @@ Result RabitTracker::Bootstrap(std::vector<WorkerProxy>* p_workers) {
 
       auto worker = WorkerProxy{n_workers_, std::move(sock), std::move(addr)};
       if (!worker.Status().OK()) {
-        return Fail("Failed to initialize worker proxy.", std::move(worker.Status()));
+        LOG(WARNING) << "Failed to initialize worker proxy." << worker.Status().Report();
+        continue;
       }
       switch (worker.Command()) {
         case proto::CMD::kStart: {

Original file line number	Diff line number	Diff line change
`@@ -286,7 +286,8 @@ Result RabitTracker::Bootstrap(std::vector<WorkerProxy>* p_workers) {`
`286`	`286`
`287`	`287`	`auto worker = WorkerProxy{n_workers_, std::move(sock), std::move(addr)};`
`288`	`288`	`if (!worker.Status().OK()) {`
`289`		`- return Fail("Failed to initialize worker proxy.", std::move(worker.Status()));`
	`289`	`+ LOG(WARNING) << "Failed to initialize worker proxy." << worker.Status().Report();`
	`290`	`+ continue;`
`290`	`291`	`}`
`291`	`292`	`switch (worker.Command()) {`
`292`	`293`	`case proto::CMD::kStart: {`