fix: build text index when loading field data (#39070) (#39113)

fix: #39053 may fix #38644 which could be caused by #39053 --------- Signed-off-by: SpadeA-Tang <[email protected]>
milvus-io · Jan 9, 2025 · f70262c · f70262c
1 parent f896b0e
commit f70262c
Show file tree

Hide file tree

Showing 4 changed files with 110 additions and 0 deletions.
diff --git a/internal/core/src/index/TextMatchIndex.cpp b/internal/core/src/index/TextMatchIndex.cpp
@@ -198,6 +198,39 @@ TextMatchIndex::AddTexts(size_t n,
     }
 }
 
+// schema_ may not be initialized so we need this `nullable` parameter
+void
+TextMatchIndex::BuildIndexFromFieldData(
+    const std::vector<FieldDataPtr>& field_datas, bool nullable) {
+    int64_t offset = 0;
+    if (nullable) {
+        int64_t total = 0;
+        for (const auto& data : field_datas) {
+            total += data->get_null_count();
+        }
+        null_offset.reserve(total);
+        for (const auto& data : field_datas) {
+            auto n = data->get_num_rows();
+            for (int i = 0; i < n; i++) {
+                if (!data->is_valid(i)) {
+                    null_offset.push_back(i);
+                }
+                wrapper_->add_data(
+                    static_cast<const std::string*>(data->RawValue(i)),
+                    data->is_valid(i) ? 1 : 0,
+                    offset++);
+            }
+        }
+    } else {
+        for (const auto& data : field_datas) {
+            auto n = data->get_num_rows();
+            wrapper_->add_data(
+                static_cast<const std::string*>(data->Data()), n, offset);
+            offset += n;
+        }
+    }
+}
+
 void
 TextMatchIndex::Finish() {
     finish();

diff --git a/internal/core/src/index/TextMatchIndex.h b/internal/core/src/index/TextMatchIndex.h
@@ -58,6 +58,10 @@ class TextMatchIndex : public InvertedIndexTantivy<std::string> {
              const bool* valids,
              int64_t offset_begin);
 
+    void
+    BuildIndexFromFieldData(const std::vector<FieldDataPtr>& field_datas,
+                            bool nullable);
+
     void
     Finish();
 

diff --git a/internal/core/src/segcore/SegmentGrowingImpl.cpp b/internal/core/src/segcore/SegmentGrowingImpl.cpp
@@ -274,6 +274,16 @@ SegmentGrowingImpl::LoadFieldData(const LoadFieldDataInfo& infos) {
                 storage::GetByteSizeOfFieldDatas(field_data));
         }
 
+        // build text match index
+        if (field_meta.enable_match()) {
+            auto index = GetTextIndex(field_id);
+            index->BuildIndexFromFieldData(field_data,
+                                           field_meta.is_nullable());
+            index->Commit();
+            // Reload reader so that the index can be read immediately
+            index->Reload();
+        }
+
         // update the mem size
         stats_.mem_size += storage::GetByteSizeOfFieldDatas(field_data);
 

diff --git a/internal/core/unittest/test_text_match.cpp b/internal/core/unittest/test_text_match.cpp
@@ -20,6 +20,8 @@
 #include "query/PlanProto.h"
 #include "query/ExecPlanNodeVisitor.h"
 #include "expr/ITypeExpr.h"
+#include "segcore/segment_c.h"
+#include "test_utils/storage_test_utils.h"
 
 using namespace milvus;
 using namespace milvus::query;
@@ -751,3 +753,64 @@ TEST(TextMatch, SealedJieBaNullable) {
         ASSERT_FALSE(final[2]);
     }
 }
+
+// Test that growing segment loading flushed binlogs will build text match index.
+TEST(TextMatch, GrowingLoadData) {
+    int64_t N = 7;
+    auto schema = GenTestSchema({}, true);
+    schema->AddField(FieldName("RowID"), FieldId(0), DataType::INT64, false);
+    schema->AddField(
+        FieldName("Timestamp"), FieldId(1), DataType::INT64, false);
+    std::vector<std::string> raw_str = {"football, basketball, pingpang",
+                                        "swimming, football",
+                                        "golf",
+                                        "",
+                                        "baseball",
+                                        "kungfu, football",
+                                        ""};
+    auto raw_data = DataGen(schema, N);
+    auto str_col = raw_data.raw_->mutable_fields_data()
+                       ->at(1)
+                       .mutable_scalars()
+                       ->mutable_string_data()
+                       ->mutable_data();
+    for (int64_t i = 0; i < N; i++) {
+        str_col->at(i) = raw_str[i];
+    }
+    auto str_col_valid =
+        raw_data.raw_->mutable_fields_data()->at(1).mutable_valid_data();
+    for (int64_t i = 0; i < N; i++) {
+        str_col_valid->at(i) = true;
+    }
+    // so we cannot match the second row
+    str_col_valid->at(1) = false;
+
+    auto storage_config = get_default_local_storage_config();
+    auto cm = storage::CreateChunkManager(storage_config);
+    auto load_info = PrepareInsertBinlog(
+        1,
+        2,
+        3,
+        storage_config.root_path + "/" + "test_growing_segment_load_data",
+        raw_data,
+        cm);
+
+    auto segment = CreateGrowingSegment(schema, empty_index_meta);
+    auto status = LoadFieldData(segment.get(), &load_info);
+    ASSERT_EQ(status.error_code, Success);
+    ASSERT_EQ(segment->get_real_count(), N);
+    ASSERT_NE(segment->get_field_avg_size(FieldId(101)), 0);
+
+    // Check whether the text index has been built.
+    auto expr = GetTextMatchExpr(schema, "football");
+    BitsetType final;
+    final = ExecuteQueryExpr(expr, segment.get(), N, MAX_TIMESTAMP);
+    ASSERT_EQ(final.size(), N);
+    ASSERT_TRUE(final[0]);
+    ASSERT_FALSE(final[1]);
+    ASSERT_FALSE(final[2]);
+    ASSERT_FALSE(final[3]);
+    ASSERT_FALSE(final[4]);
+    ASSERT_TRUE(final[5]);
+    ASSERT_FALSE(final[6]);
+}