Storages: Fix cloning delta index when there are duplicated tuples (#9000)

dbms/src/Storages/DeltaMerge/Delta/Snapshot.cpp

@@ -98,8 +98,8 @@ size_t DeltaValueReader::readRows(MutableColumns & output_cols, size_t offset, s
     //
     // So here, we should filter out those out-of-range rows.
 
-    auto mem_table_rows_offset = delta_snap->getMemTableSetRowsOffset();
-    auto total_delta_rows = delta_snap->getRows();
+    const auto mem_table_rows_offset = delta_snap->getMemTableSetRowsOffset();
+    const auto total_delta_rows = delta_snap->getRows();
 
     auto persisted_files_start = std::min(offset, mem_table_rows_offset);
     auto persisted_files_end = std::min(offset + limit, mem_table_rows_offset);

dbms/src/Storages/DeltaMerge/DeltaIndex.h

@@ -84,17 +84,17 @@ class DeltaIndex
         }
     }
 
-    DeltaIndexPtr tryCloneInner(size_t placed_deletes_limit, const Updates * updates = nullptr)
+    DeltaIndexPtr tryCloneInner(size_t rows_limit, size_t placed_deletes_limit, const Updates * updates = nullptr)
     {
         DeltaTreePtr delta_tree_copy;
         size_t placed_rows_copy = 0;
         size_t placed_deletes_copy = 0;
-        // Make sure the delta index do not place more deletes than `placed_deletes_limit`.
-        // Because delete ranges can break MVCC view.
         {
             std::scoped_lock lock(mutex);
-            // Safe to reuse the copy of the existing DeltaIndex
-            if (placed_deletes <= placed_deletes_limit)
+            // Make sure the MVCC view will not be broken by the mismatch of delta index and snapshot:
+            // - First, make sure the delta index do not place more deletes than `placed_deletes_limit`.
+            // - Second, make sure the snapshot includes all duplicated tuples in the delta index.
+            if (placed_deletes <= placed_deletes_limit && delta_tree->maxDupTupleID() < static_cast<Int64>(rows_limit))
             {
                 delta_tree_copy = delta_tree;
                 placed_rows_copy = placed_rows;
@@ -186,8 +186,9 @@ class DeltaIndex
     {
         std::scoped_lock lock(mutex);
 
-        if ((maybe_advanced.placed_rows >= placed_rows && maybe_advanced.placed_deletes >= placed_deletes)
-            && !(maybe_advanced.placed_rows == placed_rows && maybe_advanced.placed_deletes == placed_deletes))
+        if ((maybe_advanced.placed_rows >= placed_rows && maybe_advanced.placed_deletes >= placed_deletes) // advance
+            // not excatly the same
+            && (maybe_advanced.placed_rows != placed_rows || maybe_advanced.placed_deletes != placed_deletes))
         {
             delta_tree = maybe_advanced.delta_tree;
             placed_rows = maybe_advanced.placed_rows;
@@ -197,14 +198,17 @@ class DeltaIndex
         return false;
     }
 
-    DeltaIndexPtr tryClone(size_t /*rows*/, size_t deletes) { return tryCloneInner(deletes); }
+    /**
+     * Try to get a clone of current instance.
+     * Return an empty DeltaIndex if `deletes < this->placed_deletes` because the advanced delta-index will break
+     * the MVCC view.
+     */
+    DeltaIndexPtr tryClone(size_t rows, size_t deletes) { return tryCloneInner(rows, deletes); }
 
     DeltaIndexPtr cloneWithUpdates(const Updates & updates)
     {
-        if (unlikely(updates.empty()))
-            throw Exception("Unexpected empty updates");
-
-        return tryCloneInner(updates.front().delete_ranges_offset, &updates);
+        RUNTIME_CHECK_MSG(!updates.empty(), "Unexpected empty updates");
+        return tryCloneInner(updates.front().rows_offset, updates.front().delete_ranges_offset, &updates);
     }
 };
 

dbms/src/Storages/DeltaMerge/DeltaPlace.h

@@ -248,7 +248,10 @@ bool placeInsert(const SkippableBlockInputStreamPtr & stable, //
             tuple_id = delta_value_space_offset + (offset + i);
 
         if (dup)
+        {
             delta_tree.addDelete(rid);
+            delta_tree.setMaxDupTupleID(tuple_id);
+        }
         delta_tree.addInsert(rid, tuple_id);
     }
 

dbms/src/Storages/DeltaMerge/DeltaTree.h

@@ -778,6 +778,7 @@ class DeltaTree
     size_t num_inserts = 0;
     size_t num_deletes = 0;
     size_t num_entries = 0;
+    Int64 max_dup_tuple_id = -1;
 
     std::unique_ptr<Allocator> allocator;
     size_t bytes = 0;
@@ -1016,6 +1017,8 @@ class DeltaTree
     {
         return num_deletes;
     }
+    Int64 maxDupTupleID() const { return max_dup_tuple_id; }
+    void setMaxDupTupleID(Int64 tuple_id) { max_dup_tuple_id = std::max(tuple_id, max_dup_tuple_id); }
 
     void addDelete(UInt64 rid);
     void addInsert(UInt64 rid, UInt64 tuple_id);
@@ -1032,6 +1035,7 @@ DT_CLASS::DeltaTree(const DT_CLASS::Self & o)
     , num_inserts(o.num_inserts)
     , num_deletes(o.num_deletes)
     , num_entries(o.num_entries)
+    , max_dup_tuple_id(o.max_dup_tuple_id)
     , allocator(std::make_unique<Allocator>())
 {
     // If exception is thrown before clear copying_nodes, all nodes will be destroyed.

dbms/src/Storages/DeltaMerge/ReadThread/UnorderedInputStream.h

@@ -27,6 +27,10 @@ extern const char pause_when_reading_from_dt_stream[];
 
 namespace DB::DM
 {
+namespace tests
+{
+class DeltaMergeStoreRWTest;
+}
 class UnorderedInputStream : public IProfilingBlockInputStream
 {
     static constexpr auto NAME = "UnorderedInputStream";
@@ -134,5 +138,7 @@ class UnorderedInputStream : public IProfilingBlockInputStream
     LoggerPtr log;
     int64_t ref_no;
     bool task_pool_added;
+
+    friend class tests::DeltaMergeStoreRWTest;
 };
 } // namespace DB::DM

dbms/src/Storages/DeltaMerge/Segment.cpp

@@ -2229,7 +2229,8 @@ std::pair<DeltaIndexPtr, bool> Segment::ensurePlace(const DMContext & dm_context
                                                     UInt64 max_version) const
 {
     auto delta_snap = delta_reader->getDeltaSnap();
-    // Clone a new delta index.
+    // Try to clone from the sahred delta index, if it fails to reuse the shared delta index,
+    // it will return an empty delta index and we should place it in the following branch.
     auto my_delta_index = delta_snap->getSharedDeltaIndex()->tryClone(delta_snap->getRows(), delta_snap->getDeletes());
     auto my_delta_tree = my_delta_index->getDeltaTree();
 

dbms/src/Storages/DeltaMerge/tests/gtest_dm_delta_merge_store.cpp

@@ -38,6 +38,7 @@
 #include <algorithm>
 #include <future>
 #include <iterator>
+#include <memory>
 #include <random>
 
 namespace DB
@@ -3470,6 +3471,185 @@ try
 CATCH
 
 
+void DeltaMergeStoreRWTest::dupHandleVersionAndDeltaIndexAdvancedThanSnapshot()
+{
+    auto table_column_defines = DMTestEnv::getDefaultColumns();
+    store = reload(table_column_defines);
+
+    auto create_block = [&](UInt64 beg, UInt64 end, UInt64 ts) {
+        auto block = DMTestEnv::prepareSimpleWriteBlock(beg, end, false, ts);
+        block.checkNumberOfRows();
+        return block;
+    };
+
+    auto write_block = [&](UInt64 beg, UInt64 end, UInt64 ts) {
+        auto block = create_block(beg, end, ts);
+        store->write(*db_context, db_context->getSettingsRef(), block);
+    };
+
+    auto create_stream = [&]() {
+        return store->read(
+            *db_context,
+            db_context->getSettingsRef(),
+            store->getTableColumns(),
+            {RowKeyRange::newAll(store->isCommonHandle(), store->getRowKeyColumnSize())},
+            /* num_streams= */ 1,
+            /* start_ts= */ std::numeric_limits<UInt64>::max(),
+            EMPTY_FILTER,
+            TRACING_NAME,
+            /* keep_order= */ false,
+            /* is_fast_scan= */ false,
+            DEFAULT_BLOCK_SIZE)[0];
+    };
+
+    auto count_rows = [](BlockInputStreamPtr stream) {
+        std::size_t count = 0;
+        stream->readPrefix();
+        for (;;)
+        {
+            auto block = stream->read();
+            if (!block)
+            {
+                break;
+            }
+            count += block.rows();
+        }
+        stream->readSuffix();
+        return count;
+    };
+
+    auto get_seg_read_task = [&](BlockInputStreamPtr stream) {
+        auto unordered_stream = std::dynamic_pointer_cast<UnorderedInputStream>(stream);
+        const auto & tasks = unordered_stream->task_pool->getTasks();
+        RUNTIME_CHECK(tasks.size() == 1, tasks.size());
+        return tasks.begin()->second;
+    };
+
+    auto clone_delta_index = [](SegmentReadTaskPtr seg_read_task) {
+        auto delta_snap = seg_read_task->read_snapshot->delta;
+        return delta_snap->getSharedDeltaIndex()->tryClone(delta_snap->getRows(), delta_snap->getDeletes());
+    };
+
+    auto check_delta_index
+        = [](DeltaIndexPtr delta_index, size_t expect_rows, size_t expect_deletes, Int64 expect_max_dup_tuple_id) {
+              auto [placed_rows, placed_deletes] = delta_index->getPlacedStatus();
+              ASSERT_EQ(placed_rows, expect_rows);
+              ASSERT_EQ(placed_deletes, expect_deletes);
+              ASSERT_EQ(delta_index->getDeltaTree()->maxDupTupleID(), expect_max_dup_tuple_id);
+          };
+
+    auto ensure_place = [&](SegmentReadTaskPtr seg_read_task) {
+        auto pk_ver_col_defs = std::make_shared<ColumnDefines>(
+            ColumnDefines{getExtraHandleColumnDefine(dm_context->is_common_handle), getVersionColumnDefine()});
+        auto delta_reader = std::make_shared<DeltaValueReader>(
+            *dm_context,
+            seg_read_task->read_snapshot->delta,
+            pk_ver_col_defs,
+            RowKeyRange::newAll(store->isCommonHandle(), store->getRowKeyColumnSize()));
+        return seg_read_task->segment->ensurePlace(
+            *dm_context,
+            seg_read_task->read_snapshot->stable,
+            delta_reader,
+            {RowKeyRange::newAll(store->isCommonHandle(), store->getRowKeyColumnSize())},
+            std::numeric_limits<UInt64>::max());
+    };
+
+    // Write [0, 128) with ts 1 for initializing stable.
+    write_block(0, 128, 1);
+    store->mergeDeltaAll(*db_context);
+
+    // Write [50, 60) with ts 2 for initializing delta.
+    write_block(50, 60, 2);
+
+    // Scan table normally.
+    {
+        auto stream = create_stream();
+        auto count = count_rows(stream);
+        ASSERT_EQ(count, 128);
+    }
+
+    // The snapshot does not include all the duplicated tuples of the delta index.
+    // This snapshot should rebuild delta index for itself.
+    // https://github.com/pingcap/tiflash/issues/8845
+    {
+        // Create snapshot but not place index
+        auto stream1 = create_stream();
+
+        // !!!Duplicated!!!: Write [50, 60) with ts 2
+        write_block(50, 60, 2);
+
+        // Place index with newest data.
+        auto stream2 = create_stream();
+        auto count2 = count_rows(stream2);
+        ASSERT_EQ(count2, 128);
+
+        // stream1 should not resue delta index of stream2
+
+        // Check cloning delta index
+        {
+            auto seg_read_task = get_seg_read_task(stream1);
+
+            // Shared delta index has been placed to the newest by `count_rows(stream2)`.
+            auto shared_delta_index = seg_read_task->read_snapshot->delta->getSharedDeltaIndex();
+            check_delta_index(shared_delta_index, 20, 0, 19);
+
+            // Cannot clone delta index because it contains duplicated records in the gap of snapshot and the shared delta index.
+            auto cloned_delta_index = clone_delta_index(seg_read_task);
+            check_delta_index(cloned_delta_index, 0, 0, -1);
+        }
+        // Check scanning result of stream1
+        auto count1 = count_rows(stream1);
+        ASSERT_EQ(count1, count2);
+    }
+
+    // Make sure shared delta index can be reused by new snapshot
+    {
+        auto stream = create_stream();
+        auto seg_read_task = get_seg_read_task(stream);
+        auto cloned_delta_index = clone_delta_index(seg_read_task);
+        check_delta_index(cloned_delta_index, 20, 0, 19);
+    }
+
+    // The snapshot includes all the duplicated tuples of the delta index.
+    // Delta index can be reused safely.
+    {
+        write_block(70, 80, 2);
+        auto stream = create_stream();
+        auto seg_read_task = get_seg_read_task(stream);
+        auto shared_delta_index = seg_read_task->read_snapshot->delta->getSharedDeltaIndex();
+        check_delta_index(shared_delta_index, 20, 0, 19);
+        auto cloned_delta_index = clone_delta_index(seg_read_task);
+        check_delta_index(cloned_delta_index, 20, 0, 19);
+        auto [placed_delta_index, fully_indexed] = ensure_place(seg_read_task);
+        ASSERT_TRUE(fully_indexed);
+        check_delta_index(placed_delta_index, 30, 0, 19);
+        auto count = count_rows(stream);
+        ASSERT_EQ(count, 128);
+    }
+
+    {
+        write_block(75, 85, 2);
+        auto stream = create_stream();
+        auto seg_read_task = get_seg_read_task(stream);
+        auto shared_delta_index = seg_read_task->read_snapshot->delta->getSharedDeltaIndex();
+        check_delta_index(shared_delta_index, 30, 0, 19);
+        auto cloned_delta_index = clone_delta_index(seg_read_task);
+        check_delta_index(cloned_delta_index, 30, 0, 19);
+        auto [placed_delta_index, fully_indexed] = ensure_place(seg_read_task);
+        ASSERT_TRUE(fully_indexed);
+        check_delta_index(placed_delta_index, 40, 0, 34);
+        auto count = count_rows(stream);
+        ASSERT_EQ(count, 128);
+    }
+}
+
+TEST_P(DeltaMergeStoreRWTest, DupHandleVersionAndDeltaIndexAdvancedThanSnapshot)
+try
+{
+    dupHandleVersionAndDeltaIndexAdvancedThanSnapshot();
+}
+CATCH
+
 } // namespace tests
 } // namespace DM
 } // namespace DB

dbms/src/Storages/DeltaMerge/tests/gtest_dm_delta_merge_store_test_basic.h

@@ -125,6 +125,7 @@ class DeltaMergeStoreRWTest
     {
         TiFlashStorageTestBasic::SetUp();
         store = reload();
+        dm_context = store->newDMContext(*db_context, db_context->getSettingsRef());
     }
 
     DeltaMergeStorePtr
@@ -180,8 +181,11 @@ class DeltaMergeStoreRWTest
 protected:
     TestMode mode;
     DeltaMergeStorePtr store;
+    DMContextPtr dm_context;
 
     constexpr static const char * TRACING_NAME = "DeltaMergeStoreRWTest";
+
+    void dupHandleVersionAndDeltaIndexAdvancedThanSnapshot();
 };
 } // namespace tests
 } // namespace DM