From 4a800115f95512e67074b723432888f35becd1df Mon Sep 17 00:00:00 2001
From: Lei Zhang <27994433+SWJTU-ZhangLei@users.noreply.github.com>
Date: Mon, 28 Oct 2024 15:08:11 +0800
Subject: [PATCH] [fix](be) Fix creating hdfs connection coredump within
 bthread

* https://brpc.apache.org/docs/server/basics/
* According to the brpc doc, `JNI code checks stack layout and cannot be run in bthreads`
  so create a pthread for creating hdfs connection if necessary.
---
 be/src/io/hdfs_util.cpp | 21 ++++++++++++++++++++-
 1 file changed, 20 insertions(+), 1 deletion(-)

diff --git a/be/src/io/hdfs_util.cpp b/be/src/io/hdfs_util.cpp
index 6c1bbf80a1526f1..f78c292169e7cbd 100644
--- a/be/src/io/hdfs_util.cpp
+++ b/be/src/io/hdfs_util.cpp
@@ -20,7 +20,10 @@
 #include <bvar/latency_recorder.h>
 #include <gen_cpp/cloud.pb.h>
 
+#include <future>
 #include <ostream>
+#include <thread>
+#include <utility>
 
 #include "common/logging.h"
 #include "io/fs/err_utils.h"
@@ -30,7 +33,7 @@
 namespace doris::io {
 namespace {
 
-Status create_hdfs_fs(const THdfsParams& hdfs_params, const std::string& fs_name, hdfsFS* fs) {
+Status _create_hdfs_fs(const THdfsParams& hdfs_params, const std::string& fs_name, hdfsFS* fs) {
     HDFSCommonBuilder builder;
     RETURN_IF_ERROR(create_hdfs_builder(hdfs_params, fs_name, &builder));
     hdfsFS hdfs_fs = hdfsBuilderConnect(builder.get());
@@ -41,6 +44,22 @@ Status create_hdfs_fs(const THdfsParams& hdfs_params, const std::string& fs_name
     return Status::OK();
 }
 
+inline bool is_bthread() {
+    return (bthread_self() != 0);
+}
+
+Status create_hdfs_fs(const THdfsParams& hdfs_params, const std::string& fs_name, hdfsFS* fs) {
+    if (is_bthread()) {
+        Status st;
+        std::thread t([&] { st = _create_hdfs_fs(hdfs_params, fs_name, fs); });
+        if (t.joinable()) {
+            t.join();
+        }
+        return st;
+    }
+    return _create_hdfs_fs(hdfs_params, fs_name, fs);
+}
+
 uint64_t hdfs_hash_code(const THdfsParams& hdfs_params, const std::string& fs_name) {
     uint64_t hash_code = 0;
     // The specified fsname is used first.