[fix](clone) Increase robustness for clone (apache#36642)

Ma1oneZhang · Jun 21, 2024 · 73f0101 · 73f0101
1 parent 86dddbe
commit 73f0101
Show file tree

Hide file tree

Showing 3 changed files with 26 additions and 4 deletions.
diff --git a/be/src/io/cache/fs_file_cache_storage.cpp b/be/src/io/cache/fs_file_cache_storage.cpp
@@ -463,7 +463,12 @@ void FSFileCacheStorage::load_cache_info_into_memory(BlockFileCache* _mgr) const
             if (key_prefix_it->path().filename().native().size() != KEY_PREFIX_LENGTH) {
                 LOG(WARNING) << "Unknown directory " << key_prefix_it->path().native()
                              << ", try to remove it";
-                std::filesystem::remove(key_prefix_it->path());
+                std::error_code ec;
+                std::filesystem::remove(key_prefix_it->path(), ec);
+                if (ec) {
+                    LOG(WARNING) << "failed to remove=" << key_prefix_it->path()
+                                 << " msg=" << ec.message();
+                }
                 continue;
             }
             std::filesystem::directory_iterator key_it {key_prefix_it->path(), ec};

diff --git a/be/src/olap/single_replica_compaction.cpp b/be/src/olap/single_replica_compaction.cpp
@@ -582,7 +582,12 @@ Status SingleReplicaCompaction::_finish_clone(const string& clone_dir,
     }
     // clear clone dir
     std::filesystem::path clone_dir_path(clone_dir);
-    std::filesystem::remove_all(clone_dir_path);
+    std::error_code ec;
+    std::filesystem::remove_all(clone_dir_path, ec);
+    if (ec) {
+        LOG(WARNING) << "failed to remove=" << clone_dir_path << " msg=" << ec.message();
+        return Status::IOError("failed to remove {}, due to {}", clone_dir, ec.message());
+    }
     LOG(INFO) << "finish to clone data, clear downloaded data. res=" << res
               << ", tablet=" << _tablet->tablet_id() << ", clone_dir=" << clone_dir;
     return res;

diff --git a/be/src/olap/task/engine_clone_task.cpp b/be/src/olap/task/engine_clone_task.cpp
@@ -621,7 +621,13 @@ Status EngineCloneTask::_download_files(DataDir* data_dir, const std::string& re
 /// 2. Call _finish_xx_clone() to revise the tablet meta.
 Status EngineCloneTask::_finish_clone(Tablet* tablet, const std::string& clone_dir, int64_t version,
                                       bool is_incremental_clone) {
-    Defer remove_clone_dir {[&]() { std::filesystem::remove_all(clone_dir); }};
+    Defer remove_clone_dir {[&]() {
+        std::error_code ec;
+        std::filesystem::remove_all(clone_dir, ec);
+        if (ec) {
+            LOG(WARNING) << "failed to remove=" << clone_dir << " msg=" << ec.message();
+        }
+    }};
 
     // check clone dir existed
     bool exists = true;
@@ -652,7 +658,13 @@ Status EngineCloneTask::_finish_clone(Tablet* tablet, const std::string& clone_d
     bool contain_binlog = false;
     RowsetBinlogMetasPB rowset_binlog_metas_pb;
     if (binlog_metas_file_exists) {
-        auto binlog_meta_filesize = std::filesystem::file_size(binlog_metas_file);
+        std::error_code ec;
+        auto binlog_meta_filesize = std::filesystem::file_size(binlog_metas_file, ec);
+        if (ec) {
+            LOG(WARNING) << "get file size error" << ec.message();
+            return Status::IOError("can't retrive file_size of {}, due to {}", binlog_metas_file,
+                                   ec.message());
+        }
         if (binlog_meta_filesize > 0) {
             contain_binlog = true;
             RETURN_IF_ERROR(read_pb(binlog_metas_file, &rowset_binlog_metas_pb));