[Enhancement] Allow disable type auto-detection on CSV Files

be/src/exec/file_scanner/csv_scanner.cpp

@@ -590,7 +590,10 @@ void CSVScanner::_report_rejected_record(const CSVReader::Record& record, const
     _state->append_rejected_record_to_file(record.to_string(), err_msg, _curr_reader->filename());
 }
 
-static TypeDescriptor get_type_desc(const Slice& field) {
+static TypeDescriptor get_type_desc(const Slice& field, const bool& sampleTypes) {
+    if (!sampleTypes) {
+        return TypeDescriptor::create_varchar_type(TypeDescriptor::MAX_VARCHAR_LENGTH);
+    }
     StringParser::ParseResult result;
 
     StringParser::string_to_int<int64_t>(field.get_data(), field.get_size(), &result);
@@ -642,7 +645,8 @@ Status CSVScanner::_get_schema(std::vector<SlotDescriptor>* merged_schema) {
         _curr_reader->split_record(record, &fields);
         for (size_t i = 0; i < fields.size(); i++) {
             // column name: $1, $2, $3...
-            schema.emplace_back(i, fmt::format("${}", i + 1), get_type_desc(fields[i]));
+            schema.emplace_back(i, fmt::format("${}", i + 1),
+                                get_type_desc(fields[i], _scan_range.params.schema_sample_types));
         }
         schemas.emplace_back(schema);
         i++;
@@ -679,7 +683,8 @@ Status CSVScanner::_get_schema_v2(std::vector<SlotDescriptor>* merged_schema) {
             const Slice field(basePtr + column.start_pos, column.length);
 
             // column name: $1, $2, $3...
-            schema.emplace_back(i, fmt::format("${}", i + 1), get_type_desc(field));
+            schema.emplace_back(i, fmt::format("${}", i + 1),
+                                get_type_desc(field, _scan_range.params.schema_sample_types));
         }
         schemas.emplace_back(schema);
         i++;

be/test/exec/file_scanner/csv_scanner_test.cpp

@@ -1308,6 +1308,65 @@ TEST_P(CSVScannerTest, test_get_schema) {
             EXPECT_EQ(expected_schema[i].second, schema[i].type().type) << schema[i].col_name();
         }
     }
+
+    {
+        // sample 1 row, enclose ", escape "\"
+        std::vector<std::pair<std::string, LogicalType>> expected_schema = {
+                {"$1", TYPE_BIGINT}, {"$2", TYPE_VARCHAR}, {"$3", TYPE_DOUBLE}, {"$4", TYPE_BOOLEAN}};
+
+        std::vector<TBrokerRangeDesc> ranges;
+        TBrokerRangeDesc range;
+        range.__set_path("./be/test/exec/test_data/csv_scanner/type_sniff.csv");
+        range.__set_num_of_columns_from_file(0);
+        ranges.push_back(range);
+
+        TBrokerScanRangeParams* params = _obj_pool.add(new TBrokerScanRangeParams());
+        params->__set_row_delimiter('\n');
+        params->__set_column_separator(',');
+        params->__set_enclose('"');
+        params->__set_escape('\\');
+        params->__set_schema_sample_file_row_count(1);
+        auto scanner = create_csv_scanner({}, ranges, params);
+        EXPECT_OK(scanner->open());
+        std::vector<SlotDescriptor> schema;
+        EXPECT_OK(scanner->get_schema(&schema));
+        EXPECT_EQ(expected_schema.size(), schema.size());
+
+        for (size_t i = 0; i < schema.size(); i++) {
+            EXPECT_EQ(expected_schema[i].first, schema[i].col_name());
+            EXPECT_EQ(expected_schema[i].second, schema[i].type().type) << schema[i].col_name();
+        }
+    }
+
+    {
+        // sample 1 row, enclose ", escape "\", no type detection
+        std::vector<std::pair<std::string, LogicalType>> expected_schema = {
+                {"$1", TYPE_VARCHAR}, {"$2", TYPE_VARCHAR}, {"$3", TYPE_VARCHAR}, {"$4", TYPE_VARCHAR}};
+
+        std::vector<TBrokerRangeDesc> ranges;
+        TBrokerRangeDesc range;
+        range.__set_path("./be/test/exec/test_data/csv_scanner/type_sniff.csv");
+        range.__set_num_of_columns_from_file(0);
+        ranges.push_back(range);
+
+        TBrokerScanRangeParams* params = _obj_pool.add(new TBrokerScanRangeParams());
+        params->__set_row_delimiter('\n');
+        params->__set_column_separator(',');
+        params->__set_enclose('"');
+        params->__set_escape('\\');
+        params->__set_schema_sample_file_row_count(1);
+        params->__set_schema_sample_types(false);
+        auto scanner = create_csv_scanner({}, ranges, params);
+        EXPECT_OK(scanner->open());
+        std::vector<SlotDescriptor> schema;
+        EXPECT_OK(scanner->get_schema(&schema));
+        EXPECT_EQ(expected_schema.size(), schema.size());
+
+        for (size_t i = 0; i < schema.size(); i++) {
+            EXPECT_EQ(expected_schema[i].first, schema[i].col_name());
+            EXPECT_EQ(expected_schema[i].second, schema[i].type().type) << schema[i].col_name();
+        }
+    }
 }
 
 TEST_P(CSVScannerTest, test_flexible_column_mapping) {

be/test/exec/test_data/csv_scanner/type_sniff.csv

@@ -0,0 +1 @@
+001,Jok\,e,"21.3",false

docs/en/sql-reference/sql-functions/table-functions/files.md

@@ -218,6 +218,7 @@ You can configure the sampling rule using the following parameters:
 
 - `auto_detect_sample_files`: the number of random data files to sample in each batch. By default, the first and last files are selected. Range: `[0, + ∞]`. Default: `2`.
 - `auto_detect_sample_rows`: the number of data rows to scan in each sampled data file. Range: `[0, + ∞]`. Default: `500`.
+- `auto_detect_types`: (valid for CSV only) - whether to guess the data types of sampled columns, or just assume String. `{true | false}`. Default: `true`.
 
 After the sampling, StarRocks unionizes the columns from all the data files according to these rules:
 
@@ -227,6 +228,7 @@ After the sampling, StarRocks unionizes the columns from all the data files acco
   - Integer columns together with `FLOAT` type columns will be unionized as the DECIMAL type.
   - String types are used for unionizing other types.
 - Generally, the `STRING` type can be used to unionize all data types.
+- If type auto-detection is turned off, all columns will return as `STRING`
 
 You can refer to Example 5.
 

docs/ja/sql-reference/sql-functions/table-functions/files.md

@@ -218,6 +218,7 @@ v3.2 以降、`FILES()` は同じバッチのデータファイルの自動ス
 
 - `auto_detect_sample_files`: 各バッチでサンプリングするランダムなデータファイルの数。デフォルトでは、最初と最後のファイルが選択されます。範囲: `[0, + ∞]`。デフォルト: `2`。
 - `auto_detect_sample_rows`: 各サンプリングされたデータファイルでスキャンするデータ行の数。範囲: `[0, + ∞]`。デフォルト: `500`。
+- `auto_detect_types`: (CSV のみ有効) - サンプリングされた列のデータ型を推測するか、文字列であると想定するかを指定します。`{true | false}`。デフォルト: `true`。
 
 サンプリング後、StarRocks は次のルールに従ってすべてのデータファイルから列を統合します。
 
@@ -227,6 +228,7 @@ v3.2 以降、`FILES()` は同じバッチのデータファイルの自動ス
   - 整数列と `FLOAT` 型列は DECIMAL 型として統合されます。
   - 文字列型は他の型を統合するために使用されます。
 - 一般的に、`STRING` 型はすべてのデータ型を統合するために使用できます。
+- 型の自動検出がオフになっている場合、すべての列は `STRING` として返されます
 
 例 5 を参照してください。
 

docs/zh/sql-reference/sql-functions/table-functions/files.md

@@ -218,6 +218,7 @@ CSV 格式示例：
 
 - `auto_detect_sample_files`：每批中要采样的随机数据文件数量。默认情况下，选择第一个和最后一个文件。范围：`[0, + ∞]`。默认值：`2`。
 - `auto_detect_sample_rows`：每个采样数据文件中要扫描的数据行数。范围：`[0, + ∞]`。默认值：`500`。
+- `auto_detect_types`：（仅适用于 CSV 文件）- 是否猜测采样列的数据类型，或者直接假定为字符串。`{true | false}`。默认值：`true`。
 
 采样后，StarRocks 根据以下规则联合化所有数据文件的列：
 
@@ -227,6 +228,7 @@ CSV 格式示例：
   - 整数列与 `FLOAT` 类型列一起将被联合化为 DECIMAL 类型。
   - 字符串类型用于联合化其他类型。
 - 通常，`STRING` 类型可以用于联合化所有数据类型。
+- 如果类型自动检测已关闭，则所有列都将返回 `STRING` 类型。
 
 您可以参考示例 5。
 

fe/fe-core/src/main/java/com/starrocks/catalog/TableFunctionTable.java

@@ -127,6 +127,7 @@
 
     public static final String PROPERTY_AUTO_DETECT_SAMPLE_FILES = "auto_detect_sample_files";
     public static final String PROPERTY_AUTO_DETECT_SAMPLE_ROWS = "auto_detect_sample_rows";
+    public static final String PROPERTY_AUTO_DETECT_TYPES = "auto_detect_types";
 
     private static final String PROPERTY_FILL_MISMATCH_COLUMN_WITH = "fill_mismatch_column_with";
 
@@ -177,6 +178,7 @@
     // for load/query data
     private int autoDetectSampleFiles = DEFAULT_AUTO_DETECT_SAMPLE_FILES;
     private int autoDetectSampleRows = DEFAULT_AUTO_DETECT_SAMPLE_ROWS;
+    private boolean autoDetectTypes = true;
 
     private List<String> columnsFromPath = new ArrayList<>();
     private boolean strictMode = false;
@@ -504,6 +506,21 @@
             }
         }
 
+        if (properties.containsKey(PROPERTY_AUTO_DETECT_TYPES)) {
+            String property = properties.get(PROPERTY_AUTO_DETECT_TYPES);
+            if (property.equalsIgnoreCase("true")) {
+                autoDetectTypes = true;
+            } else if (property.equalsIgnoreCase("false")) {
+                autoDetectTypes = false;
+            } else {
+                throw new DdlException(
+                    String.format(
+                        "Illegal value of %s: %s, only true/false allowed", PROPERTY_AUTO_DETECT_TYPES, property
+                    )
+                );
+            }
+        }
+
         if (properties.containsKey(PROPERTY_CSV_COLUMN_SEPARATOR)) {
             csvColumnSeparator = Delimiter.convertDelimiter(properties.get(PROPERTY_CSV_COLUMN_SEPARATOR));
             int len = csvColumnSeparator.getBytes(StandardCharsets.UTF_8).length;
@@ -604,6 +621,7 @@
         params.setProperties(properties);
         params.setSchema_sample_file_count(autoDetectSampleFiles);
         params.setSchema_sample_file_row_count(autoDetectSampleRows);
+        params.setSchema_sample_types(autoDetectTypes);
         params.setEnclose(csvEnclose);
         params.setEscape(csvEscape);
         params.setSkip_header(csvSkipHeader);
@@ -783,7 +801,7 @@
         return csvTrimSpace;
     }
 
     public void parsePropertiesForUnload(List<Column> columns, SessionVariable sessionVariable) {
         List<String> columnNames = columns.stream()
                 .map(Column::getName)
                 .collect(Collectors.toList());

fe/fe-core/src/test/java/com/starrocks/catalog/TableFunctionTableTest.java

@@ -42,6 +42,7 @@
 import org.junit.jupiter.api.Assertions;
 import org.junit.jupiter.api.Test;
 
+import java.lang.reflect.Field;
 import java.lang.reflect.Method;
 import java.util.ArrayList;
 import java.util.Arrays;
@@ -362,4 +363,49 @@
                 "Invalid parquet.version: '2.0'. Expected values should be 2.4, 2.6, 1.0",
                 () -> new TableFunctionTable(new ArrayList<>(), properties, new SessionVariable()));
     }
+
+    @Test
+    public void testAutoDetectTypes() throws NoSuchFieldException {
+        Map<String, String> properties = new HashMap<>();
+        properties.put(TableFunctionTable.PROPERTY_PATH, "fake://test/path");
+        properties.put(TableFunctionTable.PROPERTY_FORMAT, "csv");
+
+        Field field = TableFunctionTable.class.getDeclaredField("autoDetectTypes");
+        field.setAccessible(true);
+
+        {
+            // Case 1: default
+            Assertions.assertDoesNotThrow(() -> {
+                TableFunctionTable table = new TableFunctionTable(properties);
+                Assertions.assertTrue((Boolean) field.get(table));
+            });
+        }
+
+        {
+            // Case 2: auto_detect_types = false
+            properties.put(TableFunctionTable.PROPERTY_AUTO_DETECT_TYPES, "false");
+            Assertions.assertDoesNotThrow(() -> {
+                TableFunctionTable table = new TableFunctionTable(properties);
+                Assertions.assertFalse((Boolean) field.get(table));
+            });
+        }
+
+        {
+            // Case 3: auto_detect_types = true
+            properties.put(TableFunctionTable.PROPERTY_AUTO_DETECT_TYPES, "true");
+            Assertions.assertDoesNotThrow(() -> {
+                TableFunctionTable table = new TableFunctionTable(properties);
+                Assertions.assertTrue((Boolean) field.get(table));
+            });
+        }
+
+        {
+            // abnormal
+            properties.put(TableFunctionTable.PROPERTY_AUTO_DETECT_TYPES, "notaboolean");
+            ExceptionChecker.expectThrowsWithMsg(DdlException.class,
+                    "Illegal value of auto_detect_types: notaboolean, only true/false allowed",
+                    () -> new TableFunctionTable(properties)
+            );
+        }
+    }
 }

gensrc/thrift/PlanNodes.thrift

@@ -284,6 +284,7 @@ struct TBrokerScanRangeParams {
     31: optional i64 schema_sample_file_row_count
     32: optional bool flexible_column_mapping
     33: optional TFileScanType file_scan_type
+    34: optional bool schema_sample_types = true
 }
 
 // Broker scan range