[branch-2.1] Picks "[opt](merge-on-write) Skip the alignment process …

…of some rowsets in partial update #38487" (#38682)

## Proposed changes

picks #38487

be/src/olap/partial_update_info.h

@@ -25,10 +25,10 @@ struct PartialUpdateInfo {
     void init(const TabletSchema& tablet_schema, bool partial_update,
               const std::set<string>& partial_update_cols, bool is_strict_mode,
               int64_t timestamp_ms, const std::string& timezone,
-              const std::string& auto_increment_column) {
+              const std::string& auto_increment_column, int64_t cur_max_version = -1) {
         is_partial_update = partial_update;
         partial_update_input_columns = partial_update_cols;
-
+        max_version_in_flush_phase = cur_max_version;
         this->timestamp_ms = timestamp_ms;
         this->timezone = timezone;
         missing_cids.clear();
@@ -91,6 +91,7 @@ struct PartialUpdateInfo {
 
 public:
     bool is_partial_update {false};
+    int64_t max_version_in_flush_phase {-1};
     std::set<std::string> partial_update_input_columns;
     std::vector<uint32_t> missing_cids;
     std::vector<uint32_t> update_cids;

be/src/olap/rowset/rowset.h

@@ -166,6 +166,7 @@ class Rowset : public std::enable_shared_from_this<Rowset> {
     bool is_segments_overlapping() const { return rowset_meta()->is_segments_overlapping(); }
     KeysType keys_type() { return _schema->keys_type(); }
     RowsetStatePB rowset_meta_state() const { return rowset_meta()->rowset_state(); }
+    bool produced_by_compaction() const { return rowset_meta()->produced_by_compaction(); }
 
     // remove all files in this rowset
     // TODO should we rename the method to remove_files() to be more specific?

be/src/olap/rowset/rowset_meta.h

@@ -242,6 +242,12 @@ class RowsetMeta {
         return num_segments() > 1 && is_singleton_delta() && segments_overlap() != NONOVERLAPPING;
     }
 
+    bool produced_by_compaction() const {
+        return has_version() &&
+               (start_version() < end_version() ||
+                (start_version() == end_version() && segments_overlap() == NONOVERLAPPING));
+    }
+
     // get the compaction score of this rowset.
     // if segments are overlapping, the score equals to the number of segments,
     // otherwise, score is 1.

be/src/olap/rowset_builder.cpp

@@ -118,7 +118,7 @@ void RowsetBuilder::_garbage_collection() {
 
 Status RowsetBuilder::init_mow_context(std::shared_ptr<MowContext>& mow_context) {
     std::lock_guard<std::shared_mutex> lck(tablet()->get_header_lock());
-    int64_t cur_max_version = tablet()->max_version_unlocked().second;
+    _max_version_in_flush_phase = tablet()->max_version_unlocked().second;
     std::vector<RowsetSharedPtr> rowset_ptrs;
     // tablet is under alter process. The delete bitmap will be calculated after conversion.
     if (tablet()->tablet_state() == TABLET_NOTREADY) {
@@ -130,12 +130,12 @@ Status RowsetBuilder::init_mow_context(std::shared_ptr<MowContext>& mow_context)
         }
         _rowset_ids.clear();
     } else {
-        RETURN_IF_ERROR(tablet()->all_rs_id(cur_max_version, &_rowset_ids));
+        RETURN_IF_ERROR(tablet()->all_rs_id(_max_version_in_flush_phase, &_rowset_ids));
         rowset_ptrs = tablet()->get_rowset_by_ids(&_rowset_ids);
     }
     _delete_bitmap = std::make_shared<DeleteBitmap>(tablet()->tablet_id());
-    mow_context = std::make_shared<MowContext>(cur_max_version, _req.txn_id, _rowset_ids,
-                                               rowset_ptrs, _delete_bitmap);
+    mow_context = std::make_shared<MowContext>(_max_version_in_flush_phase, _req.txn_id,
+                                               _rowset_ids, rowset_ptrs, _delete_bitmap);
     return Status::OK();
 }
 
@@ -402,7 +402,8 @@ void BaseRowsetBuilder::_build_current_tablet_schema(int64_t index_id,
                                table_schema_param->partial_update_input_columns(),
                                table_schema_param->is_strict_mode(),
                                table_schema_param->timestamp_ms(), table_schema_param->timezone(),
-                               table_schema_param->auto_increment_coulumn());
+                               table_schema_param->auto_increment_coulumn(),
+                               _max_version_in_flush_phase);
 }
 
 } // namespace doris

be/src/olap/rowset_builder.h

@@ -107,6 +107,7 @@ class BaseRowsetBuilder {
     std::unique_ptr<CalcDeleteBitmapToken> _calc_delete_bitmap_token;
     // current rowset_ids, used to do diff in publish_version
     RowsetIdUnorderedSet _rowset_ids;
+    int64_t _max_version_in_flush_phase {-1};
 
     std::shared_ptr<PartialUpdateInfo> _partial_update_info;
 

be/src/olap/tablet.cpp

@@ -3506,7 +3506,9 @@ Status Tablet::update_delete_bitmap(TabletTxnInfo* txn_info, int64_t txn_id) {
     // When the new segment flush fails or the rowset build fails, the deletion marker for the
     // duplicate key of the original segment should not remain in `txn_info->delete_bitmap`,
     // so we need to make a copy of `txn_info->delete_bitmap` and make changes on it.
-    if (txn_info->partial_update_info && txn_info->partial_update_info->is_partial_update) {
+    bool is_partial_update =
+            txn_info->partial_update_info && txn_info->partial_update_info->is_partial_update;
+    if (is_partial_update) {
         delete_bitmap = std::make_shared<DeleteBitmap>(*(txn_info->delete_bitmap));
     }
 
@@ -3540,6 +3542,37 @@ Status Tablet::update_delete_bitmap(TabletTxnInfo* txn_info, int64_t txn_id) {
     }
     auto t3 = watch.get_elapse_time_us();
 
+    // If a rowset is produced by compaction before the commit phase of the partial update load
+    // and is not included in txn_info->rowset_ids, we can skip the alignment process of that rowset
+    // because data remains the same before and after compaction. But we still need to calculate the
+    // the delete bitmap for that rowset.
+    std::vector<RowsetSharedPtr> rowsets_skip_alignment;
+    if (is_partial_update) {
+        int64_t max_version_in_flush_phase =
+                txn_info->partial_update_info->max_version_in_flush_phase;
+        DCHECK(max_version_in_flush_phase != -1);
+        std::vector<RowsetSharedPtr> remained_rowsets;
+        for (const auto& rowset : specified_rowsets) {
+            if (rowset->end_version() <= max_version_in_flush_phase &&
+                rowset->produced_by_compaction()) {
+                rowsets_skip_alignment.emplace_back(rowset);
+            } else {
+                remained_rowsets.emplace_back(rowset);
+            }
+        }
+        if (!rowsets_skip_alignment.empty()) {
+            specified_rowsets = std::move(remained_rowsets);
+        }
+    }
+
+    if (!rowsets_skip_alignment.empty()) {
+        auto token = _engine.calc_delete_bitmap_executor()->create_token();
+        // set rowset_writer to nullptr to skip the alignment process
+        RETURN_IF_ERROR(calc_delete_bitmap(rowset, segments, rowsets_skip_alignment, delete_bitmap,
+                                           cur_version - 1, token.get(), nullptr));
+        RETURN_IF_ERROR(token->wait());
+    }
+
     auto token = _engine.calc_delete_bitmap_executor()->create_token();
     RETURN_IF_ERROR(calc_delete_bitmap(rowset, segments, specified_rowsets, delete_bitmap,
                                        cur_version - 1, token.get(), rowset_writer.get()));

be/src/olap/task/engine_publish_version_task.cpp

@@ -110,6 +110,20 @@ Status EnginePublishVersionTask::execute() {
             std::this_thread::sleep_for(std::chrono::milliseconds(wait));
         }
     });
+    DBUG_EXECUTE_IF("EnginePublishVersionTask::execute.enable_spin_wait", {
+        auto token = dp->param<std::string>("token", "invalid_token");
+        while (DebugPoints::instance()->is_enable("EnginePublishVersionTask::execute.block")) {
+            auto block_dp = DebugPoints::instance()->get_debug_point(
+                    "EnginePublishVersionTask::execute.block");
+            if (block_dp) {
+                auto pass_token = block_dp->param<std::string>("pass_token", "");
+                if (pass_token == token) {
+                    break;
+                }
+            }
+            std::this_thread::sleep_for(std::chrono::milliseconds(50));
+        }
+    });
     std::unique_ptr<ThreadPoolToken> token =
             StorageEngine::instance()->tablet_publish_txn_thread_pool()->new_token(
                     ThreadPool::ExecutionMode::CONCURRENT);

regression-test/data/fault_injection_p0/partial_update/test_partial_update_compaction_with_higher_version.out

@@ -0,0 +1,11 @@
+-- This file is automatically generated. You should know what you did if you want to edit this
+-- !sql --
+1	1	1	1	1
+2	2	2	2	2
+3	3	3	3	3
+
+-- !sql --
+1	999	999	666	666
+2	888	888	2	2
+3	777	777	555	555
+

regression-test/data/fault_injection_p0/partial_update/test_partial_update_conflict_skip_compaction.out

@@ -0,0 +1,11 @@
+-- This file is automatically generated. You should know what you did if you want to edit this
+-- !sql --
+1	1	1	1	1
+2	2	2	2	2
+3	3	3	3	3
+
+-- !sql --
+1	999	999	666	666
+2	888	888	2	2
+3	777	777	555	555
+

regression-test/data/fault_injection_p0/partial_update/test_partial_update_skip_compaction.out

@@ -0,0 +1,11 @@
+-- This file is automatically generated. You should know what you did if you want to edit this
+-- !sql --
+1	1	1	1	1
+2	2	2	2	2
+3	3	3	3	3
+
+-- !sql --
+1	999	999	1	1
+2	888	888	2	2
+3	777	777	3	3
+