[fix](move-memtable) fix move memtable core when use multi table load (apache#35458)

sollhui · web-flow · commit 327cbf8e967e · 2024-06-05T09:11:43.000+08:00
## Proposed changes Move memtable core when use multi table load: ``` 0x51f000c73860 is located 3040 bytes inside of 3456-byte region [0x51f000c72c80,0x51f000c73a00) freed by thread T4867 (FragmentMgrThre) here: #0 0x558f6ad7f43d in operator delete(void*) (/mnt/hdd01/STRESS_ENV/be/lib/doris_be+0x22eec43d) (BuildId: b46f73d1f76dfcd6) #1 0x558f6e6cea2c in std::__new_allocator<doris::PTabletID>::deallocate(doris::PTabletID*, unsigned long) /mnt/disk2/xujianxu/ldb_toolchain/bin/../lib/gcc/x86_64-linux-gnu/13/../../../../include/c++/13/bits/new_allocator.h:168:2 #2 0x558f6e6ce9e7 in std::allocator<doris::PTabletID>::deallocate(doris::PTabletID*, unsigned long) /mnt/disk2/xujianxu/ldb_toolchain/bin/../lib/gcc/x86_64-linux-gnu/13/../../../../include/c++/13/bits/allocator.h:210:25 #3 0x558f6e6ce9e7 in std::allocator_traits<std::allocator<doris::PTabletID>>::deallocate(std::allocator<doris::PTabletID>&, doris::PTabletID*, unsigned long) /mnt/disk2/xujianxu/ldb_toolchain/bin/../lib/gcc/x86_64-linux-gnu/13/../../../../include/c++/13/bits/alloc_traits.h:516:13 #4 0x558f6e6ce9e7 in std::_Vector_base<doris::PTabletID, std::allocator<doris::PTabletID>>::_M_deallocate(doris::PTabletID*, unsigned long) /mnt/disk2/xujianxu/ldb_toolchain/bin/../lib/gcc/x86_64-linux-gnu/13/../../../../include/c++/13/bits/stl_vector.h:387:4 #5 0x558f6e6d0780 in void std::vector<doris::PTabletID, std::allocator<doris::PTabletID>>::_M_range_insert<__gnu_cxx::__normal_iterator<doris::PTabletID const*, std::vector<doris::PTabletID, std::allocator<doris::PTabletID>>>>(__gnu_cxx::__normal_iterator<doris::PTabletID*, std::vector<doris::PTabletID, std::allocator<doris::PTabletID>>>, __gnu_cxx::__normal_iterator<doris::PTabletID const*, std::vector<doris::PTabletID, std::allocator<doris::PTabletID>>>, __gnu_cxx::__normal_iterator<doris::PTabletID const*, std::vector<doris::PTabletID, std::allocator<doris::PTabletID>>>, std::forward_iterator_tag) /mnt/disk2/xujianxu/ldb_toolchain/bin/../lib/gcc/x86_64-linux-gnu/13/../../../../include/c++/13/bits/vector.tcc:832:3 #6 0x558f6e6c54c5 in __gnu_cxx::__normal_iterator<doris::PTabletID*, std::vector<doris::PTabletID, std::allocator<doris::PTabletID>>> std::vector<doris::PTabletID, std::allocator<doris::PTabletID>>::insert<__gnu_cxx::__normal_iterator<doris::PTabletID const*, std::vector<doris::PTabletID, std::allocator<doris::PTabletID>>>, void>(__gnu_cxx::__normal_iterator<doris::PTabletID const*, std::vector<doris::PTabletID, std::allocator<doris::PTabletID>>>, __gnu_cxx::__normal_iterator<doris::PTabletID const*, std::vector<doris::PTabletID, std::allocator<doris::PTabletID>>>, __gnu_cxx::__normal_iterator<doris::PTabletID const*, std::vector<doris::PTabletID, std::allocator<doris::PTabletID>>>) /mnt/disk2/xujianxu/ldb_toolchain/bin/../lib/gcc/x86_64-linux-gnu/13/../../../../include/c++/13/bits/stl_vector.h:1483:4 #7 0x558f9b4b214f in doris::LoadStreamMap::save_tablets_to_commit(long, std::vector<doris::PTabletID, std::allocator<doris::PTabletID>> const&) /mnt/disk2/xujianxu/doris/be/src/vec/sink/load_stream_map_pool.cpp:90:13 apache#8 0x558f9b7258dd in doris::vectorized::VTabletWriterV2::_calc_tablets_to_commit() /mnt/disk2/xujianxu/doris/be/src/vec/sink/writer/vtablet_writer_v2.cpp:650:27 apache#9 0x558f9b7229f1 in doris::vectorized::VTabletWriterV2::close(doris::Status) /mnt/disk2/xujianxu/doris/be/src/vec/sink/writer/vtablet_writer_v2.cpp:547:9 ``` Multiple sinks with different table loads use the load id, causing confusion in the use of shared data structures between sinks.
diff --git a/be/src/io/file_factory.cpp b/be/src/io/file_factory.cpp
@@ -214,26 +214,6 @@ Status FileFactory::create_pipe_reader(const TUniqueId& load_id, io::FileReaderS
         *file_reader = stream_load_ctx->pipe;
     }
 
-    if (file_reader->get() == nullptr) {
-        return Status::OK();
-    }
-
-    auto multi_table_pipe = std::dynamic_pointer_cast<io::MultiTablePipe>(*file_reader);
-    if (multi_table_pipe == nullptr || runtime_state == nullptr) {
-        return Status::OK();
-    }
-
-    TUniqueId pipe_id;
-    if (runtime_state->enable_pipeline_x_exec()) {
-        pipe_id = io::StreamLoadPipe::calculate_pipe_id(runtime_state->query_id(),
-                                                        runtime_state->fragment_id());
-    } else {
-        pipe_id = runtime_state->fragment_instance_id();
-    }
-    *file_reader = multi_table_pipe->get_pipe(pipe_id);
-    LOG(INFO) << "create pipe reader for fragment instance: " << pipe_id
-              << " pipe: " << (*file_reader).get();
-
     return Status::OK();
 }
 
diff --git a/be/src/io/fs/multi_table_pipe.cpp b/be/src/io/fs/multi_table_pipe.cpp
@@ -61,9 +61,9 @@ Status MultiTablePipe::append_json(const char* data, size_t size) {
 }
 
 KafkaConsumerPipePtr MultiTablePipe::get_pipe_by_table(const std::string& table) {
-    auto pipe = _planned_pipes.find(table);
-    DCHECK(pipe != _planned_pipes.end());
-    return pipe->second;
+    auto pair = _planned_tables.find(table);
+    DCHECK(pair != _planned_tables.end());
+    return std::static_pointer_cast<io::KafkaConsumerPipe>(pair->second->pipe);
 }
 
 static std::string_view get_first_part(const char* dat, char delimiter) {
@@ -78,15 +78,15 @@ static std::string_view get_first_part(const char* dat, char delimiter) {
 }
 
 Status MultiTablePipe::finish() {
-    for (auto& pair : _planned_pipes) {
-        RETURN_IF_ERROR(pair.second->finish());
+    for (auto& pair : _planned_tables) {
+        RETURN_IF_ERROR(pair.second->pipe->finish());
     }
     return Status::OK();
 }
 
 void MultiTablePipe::cancel(const std::string& reason) {
-    for (auto& pair : _planned_pipes) {
-        pair.second->cancel(reason);
+    for (auto& pair : _planned_tables) {
+        pair.second->pipe->cancel(reason);
     }
 }
 
@@ -101,19 +101,29 @@ Status MultiTablePipe::dispatch(const std::string& table, const char* data, size
         return Status::InternalError("empty data");
     }
     KafkaConsumerPipePtr pipe = nullptr;
-    auto iter = _planned_pipes.find(table);
-    if (iter != _planned_pipes.end()) {
-        pipe = iter->second;
+    auto iter = _planned_tables.find(table);
+    if (iter != _planned_tables.end()) {
+        pipe = std::static_pointer_cast<io::KafkaConsumerPipe>(iter->second->pipe);
         RETURN_NOT_OK_STATUS_WITH_WARN((pipe.get()->*cb)(data, size),
                                        "append failed in planned kafka pipe");
     } else {
-        iter = _unplanned_pipes.find(table);
-        if (iter == _unplanned_pipes.end()) {
+        iter = _unplanned_tables.find(table);
+        if (iter == _unplanned_tables.end()) {
+            std::shared_ptr<StreamLoadContext> ctx =
+                    std::make_shared<StreamLoadContext>(doris::ExecEnv::GetInstance());
+            ctx->id = UniqueId::gen_uid();
             pipe = std::make_shared<io::KafkaConsumerPipe>();
-            LOG(INFO) << "create new unplanned pipe: " << pipe.get() << ", ctx: " << _ctx->brief();
-            _unplanned_pipes.emplace(table, pipe);
+            ctx->pipe = pipe;
+#ifndef BE_TEST
+            RETURN_NOT_OK_STATUS_WITH_WARN(
+                    doris::ExecEnv::GetInstance()->new_load_stream_mgr()->put(ctx->id, ctx),
+                    "put stream load ctx error");
+#endif
+            _unplanned_tables.emplace(table, ctx);
+            LOG(INFO) << "create new unplanned table ctx, table: " << table
+                      << "load id: " << ctx->id << ", txn id: " << _ctx->txn_id;
         } else {
-            pipe = iter->second;
+            pipe = std::static_pointer_cast<io::KafkaConsumerPipe>(iter->second->pipe);
         }
 
         // It is necessary to determine whether the sum of pipe_current_capacity and size is greater than pipe_max_capacity,
@@ -124,7 +134,7 @@ Status MultiTablePipe::dispatch(const std::string& table, const char* data, size
         auto pipe_current_capacity = pipe->current_capacity();
         auto pipe_max_capacity = pipe->max_capacity();
         if (_unplanned_row_cnt >= _row_threshold ||
-            _unplanned_pipes.size() >= _wait_tables_threshold ||
+            _unplanned_tables.size() >= _wait_tables_threshold ||
             pipe_current_capacity + size > pipe_max_capacity) {
             LOG(INFO) << fmt::format(
                                  "unplanned row cnt={} reach row_threshold={} or "
@@ -151,112 +161,106 @@ Status MultiTablePipe::dispatch(const std::string& table, const char* data, size
 
 #ifndef BE_TEST
 Status MultiTablePipe::request_and_exec_plans() {
-    if (_unplanned_pipes.empty()) {
+    if (_unplanned_tables.empty()) {
         return Status::OK();
     }
 
-    // get list of table names in unplanned pipes
-    std::vector<std::string> tables;
     fmt::memory_buffer log_buffer;
     log_buffer.clear();
-    fmt::format_to(log_buffer, "request plans for {} tables: [ ", _unplanned_pipes.size());
-    for (auto& pair : _unplanned_pipes) {
-        tables.push_back(pair.first);
+    fmt::format_to(log_buffer, "request plans for {} tables: [ ", _unplanned_tables.size());
+    for (auto& pair : _unplanned_tables) {
         fmt::format_to(log_buffer, "{} ", pair.first);
     }
     fmt::format_to(log_buffer, "]");
     LOG(INFO) << fmt::to_string(log_buffer);
 
-    TStreamLoadPutRequest request;
-    set_request_auth(&request, _ctx->auth);
-    request.db = _ctx->db;
-    request.table_names = tables;
-    request.__isset.table_names = true;
-    request.txnId = _ctx->txn_id;
-    request.formatType = _ctx->format;
-    request.__set_compress_type(_ctx->compress_type);
-    request.__set_header_type(_ctx->header_type);
-    request.__set_loadId(_ctx->id.to_thrift());
-    request.fileType = TFileType::FILE_STREAM;
-    request.__set_thrift_rpc_timeout_ms(config::thrift_rpc_timeout_ms);
-    request.__set_memtable_on_sink_node(_ctx->memtable_on_sink_node);
-    request.__set_user(_ctx->qualified_user);
-    request.__set_cloud_cluster(_ctx->cloud_cluster);
-    // no need to register new_load_stream_mgr coz it is already done in routineload submit task
-
-    // plan this load
-    ExecEnv* exec_env = doris::ExecEnv::GetInstance();
-    TNetworkAddress master_addr = exec_env->master_info()->network_address;
-    int64_t stream_load_put_start_time = MonotonicNanos();
-    RETURN_IF_ERROR(ThriftRpcHelper::rpc<FrontendServiceClient>(
-            master_addr.hostname, master_addr.port,
-            [&request, this](FrontendServiceConnection& client) {
-                client->streamLoadMultiTablePut(_ctx->multi_table_put_result, request);
-            }));
-    _ctx->stream_load_put_cost_nanos = MonotonicNanos() - stream_load_put_start_time;
-
-    Status plan_status(Status::create(_ctx->multi_table_put_result.status));
-    if (!plan_status.ok()) {
-        LOG(WARNING) << "plan streaming load failed. errmsg=" << plan_status << _ctx->brief();
-        return plan_status;
-    }
-
     Status st;
-    if (_ctx->multi_table_put_result.__isset.params &&
-        !_ctx->multi_table_put_result.__isset.pipeline_params) {
-        st = exec_plans(exec_env, _ctx->multi_table_put_result.params);
-    } else if (!_ctx->multi_table_put_result.__isset.params &&
-               _ctx->multi_table_put_result.__isset.pipeline_params) {
-        st = exec_plans(exec_env, _ctx->multi_table_put_result.pipeline_params);
-    } else {
-        return Status::Aborted("too many or too few params are set in multi_table_put_result.");
-    }
+    for (auto& pair : _unplanned_tables) {
+        TStreamLoadPutRequest request;
+        set_request_auth(&request, _ctx->auth);
+        std::vector<std::string> tables;
+        tables.push_back(pair.first);
+        request.db = _ctx->db;
+        request.table_names = tables;
+        request.__isset.table_names = true;
+        request.txnId = _ctx->txn_id;
+        request.formatType = _ctx->format;
+        request.__set_compress_type(_ctx->compress_type);
+        request.__set_header_type(_ctx->header_type);
+        request.__set_loadId((pair.second->id).to_thrift());
+        request.fileType = TFileType::FILE_STREAM;
+        request.__set_thrift_rpc_timeout_ms(config::thrift_rpc_timeout_ms);
+        request.__set_memtable_on_sink_node(_ctx->memtable_on_sink_node);
+        request.__set_user(_ctx->qualified_user);
+        request.__set_cloud_cluster(_ctx->cloud_cluster);
+        // no need to register new_load_stream_mgr coz it is already done in routineload submit task
+
+        // plan this load
+        ExecEnv* exec_env = doris::ExecEnv::GetInstance();
+        TNetworkAddress master_addr = exec_env->master_info()->network_address;
+        int64_t stream_load_put_start_time = MonotonicNanos();
+        RETURN_IF_ERROR(ThriftRpcHelper::rpc<FrontendServiceClient>(
+                master_addr.hostname, master_addr.port,
+                [&request, this](FrontendServiceConnection& client) {
+                    client->streamLoadMultiTablePut(_ctx->multi_table_put_result, request);
+                }));
+        _ctx->stream_load_put_cost_nanos = MonotonicNanos() - stream_load_put_start_time;
+
+        Status plan_status(Status::create(_ctx->multi_table_put_result.status));
+        if (!plan_status.ok()) {
+            LOG(WARNING) << "plan streaming load failed. errmsg=" << plan_status << _ctx->brief();
+            return plan_status;
+        }
 
+        if (_ctx->multi_table_put_result.__isset.params &&
+            !_ctx->multi_table_put_result.__isset.pipeline_params) {
+            st = exec_plans(exec_env, _ctx->multi_table_put_result.params);
+        } else if (!_ctx->multi_table_put_result.__isset.params &&
+                   _ctx->multi_table_put_result.__isset.pipeline_params) {
+            st = exec_plans(exec_env, _ctx->multi_table_put_result.pipeline_params);
+        } else {
+            return Status::Aborted("too many or too few params are set in multi_table_put_result.");
+        }
+        if (!st.ok()) {
+            return st;
+        }
+    }
+    _unplanned_tables.clear();
     return st;
 }
 
 template <typename ExecParam>
 Status MultiTablePipe::exec_plans(ExecEnv* exec_env, std::vector<ExecParam> params) {
     // put unplanned pipes into planned pipes and clear unplanned pipes
-    for (auto& pipe : _unplanned_pipes) {
-        _ctx->table_list.push_back(pipe.first);
-        _planned_pipes.emplace(pipe.first, pipe.second);
+    for (auto& pair : _unplanned_tables) {
+        _ctx->table_list.push_back(pair.first);
+        _planned_tables.emplace(pair.first, pair.second);
     }
     LOG(INFO) << fmt::format("{} tables plan complete, planned table cnt={}, returned plan cnt={}",
-                             _unplanned_pipes.size(), _planned_pipes.size(), params.size())
+                             _unplanned_tables.size(), _planned_tables.size(), params.size())
               << ", ctx: " << _ctx->brief();
-    _unplanned_pipes.clear();
 
     for (auto& plan : params) {
         DBUG_EXECUTE_IF("MultiTablePipe.exec_plans.failed",
                         { return Status::Aborted("MultiTablePipe.exec_plans.failed"); });
         if (!plan.__isset.table_name ||
-            _planned_pipes.find(plan.table_name) == _planned_pipes.end()) {
+            _unplanned_tables.find(plan.table_name) == _unplanned_tables.end()) {
             return Status::Aborted("Missing vital param: table_name");
         }
 
-        if constexpr (std::is_same_v<ExecParam, TExecPlanFragmentParams>) {
-            RETURN_IF_ERROR(
-                    put_pipe(plan.params.fragment_instance_id, _planned_pipes[plan.table_name]));
-            LOG(INFO) << "fragment_instance_id=" << print_id(plan.params.fragment_instance_id)
-                      << " table=" << plan.table_name << ", ctx: " << _ctx->brief();
-        } else if constexpr (std::is_same_v<ExecParam, TPipelineFragmentParams>) {
-            auto pipe_id = calculate_pipe_id(plan.query_id, plan.fragment_id);
-            RETURN_IF_ERROR(put_pipe(pipe_id, _planned_pipes[plan.table_name]));
-            LOG(INFO) << "pipe_id=" << pipe_id << ", table=" << plan.table_name
-                      << ", ctx: " << _ctx->brief();
-        } else {
-            LOG(WARNING) << "illegal exec param type, need `TExecPlanFragmentParams` or "
-                            "`TPipelineFragmentParams`, will crash"
-                         << ", ctx: " << _ctx->brief();
-            CHECK(false);
-        }
-
         _inflight_cnt++;
 
         RETURN_IF_ERROR(exec_env->fragment_mgr()->exec_plan_fragment(
-                plan, [this](RuntimeState* state, Status* status) {
+                plan, [this, plan](RuntimeState* state, Status* status) {
                     DCHECK(state);
+                    auto pair = _planned_tables.find(plan.table_name);
+                    if (pair == _planned_tables.end()) {
+                        LOG(WARNING) << "failed to get ctx, table: " << plan.table_name;
+                    } else {
+                        doris::ExecEnv::GetInstance()->new_load_stream_mgr()->remove(
+                                pair->second->id);
+                    }
+
                     {
                         std::lock_guard<std::mutex> l(_tablet_commit_infos_lock);
                         _tablet_commit_infos.insert(_tablet_commit_infos.end(),
@@ -300,12 +304,12 @@ Status MultiTablePipe::exec_plans(ExecEnv* exec_env, std::vector<ExecParam> para
 #else
 Status MultiTablePipe::request_and_exec_plans() {
     // put unplanned pipes into planned pipes
-    for (auto& pipe : _unplanned_pipes) {
-        _planned_pipes.emplace(pipe.first, pipe.second);
+    for (auto& pipe : _unplanned_tables) {
+        _planned_tables.emplace(pipe.first, pipe.second);
     }
     LOG(INFO) << fmt::format("{} tables plan complete, planned table cnt={}",
-                             _unplanned_pipes.size(), _planned_pipes.size());
-    _unplanned_pipes.clear();
+                             _unplanned_tables.size(), _planned_tables.size());
+    _unplanned_tables.clear();
     return Status::OK();
 }
 
@@ -330,35 +334,6 @@ void MultiTablePipe::_handle_consumer_finished() {
     _ctx->promise.set_value(_status); // when all done, finish the routine load task
 }
 
-Status MultiTablePipe::put_pipe(const TUniqueId& pipe_id,
-                                std::shared_ptr<io::StreamLoadPipe> pipe) {
-    std::lock_guard<std::mutex> l(_pipe_map_lock);
-    auto it = _pipe_map.find(pipe_id);
-    if (it != std::end(_pipe_map)) {
-        return Status::InternalError("id already exist");
-    }
-    _pipe_map.emplace(pipe_id, pipe);
-    return Status::OK();
-}
-
-std::shared_ptr<io::StreamLoadPipe> MultiTablePipe::get_pipe(const TUniqueId& pipe_id) {
-    std::lock_guard<std::mutex> l(_pipe_map_lock);
-    auto it = _pipe_map.find(pipe_id);
-    if (it == std::end(_pipe_map)) {
-        return {};
-    }
-    return it->second;
-}
-
-void MultiTablePipe::remove_pipe(const TUniqueId& pipe_id) {
-    std::lock_guard<std::mutex> l(_pipe_map_lock);
-    auto it = _pipe_map.find(pipe_id);
-    if (it != std::end(_pipe_map)) {
-        _pipe_map.erase(it);
-        VLOG_NOTICE << "remove stream load pipe: " << pipe_id;
-    }
-}
-
 template Status MultiTablePipe::exec_plans(ExecEnv* exec_env,
                                            std::vector<TExecPlanFragmentParams> params);
 template Status MultiTablePipe::exec_plans(ExecEnv* exec_env,
diff --git a/be/src/io/fs/multi_table_pipe.h b/be/src/io/fs/multi_table_pipe.h
@@ -60,13 +60,6 @@ class MultiTablePipe : public KafkaConsumerPipe {
 
     void cancel(const std::string& reason) override;
 
-    // register <instance id, pipe> pair
-    Status put_pipe(const TUniqueId& pipe_id, std::shared_ptr<io::StreamLoadPipe> pipe);
-
-    std::shared_ptr<io::StreamLoadPipe> get_pipe(const TUniqueId& pipe_id);
-
-    void remove_pipe(const TUniqueId& pipe_id);
-
 private:
     // parse table name from data
     std::string parse_dst_table(const char* data, size_t size);
@@ -82,8 +75,8 @@ class MultiTablePipe : public KafkaConsumerPipe {
     void _handle_consumer_finished();
 
 private:
-    std::unordered_map<std::string /*table*/, KafkaConsumerPipePtr> _planned_pipes;
-    std::unordered_map<std::string /*table*/, KafkaConsumerPipePtr> _unplanned_pipes;
+    std::unordered_map<std::string /*table*/, std::shared_ptr<StreamLoadContext>> _planned_tables;
+    std::unordered_map<std::string /*table*/, std::shared_ptr<StreamLoadContext>> _unplanned_tables;
     std::atomic<uint64_t> _unplanned_row_cnt {0}; // trigger plan request when exceed threshold
     // inflight count, when it is zero, means consume and all plans is finished
     std::atomic<uint64_t> _inflight_cnt {1};
