be/src/format/table/hive_reader.cpp
Line | Count | Source |
1 | | // Licensed to the Apache Software Foundation (ASF) under one |
2 | | // or more contributor license agreements. See the NOTICE file |
3 | | // distributed with this work for additional information |
4 | | // regarding copyright ownership. The ASF licenses this file |
5 | | // to you under the Apache License, Version 2.0 (the |
6 | | // "License"); you may not use this file except in compliance |
7 | | // with the License. You may obtain a copy of the License at |
8 | | // |
9 | | // http://www.apache.org/licenses/LICENSE-2.0 |
10 | | // |
11 | | // Unless required by applicable law or agreed to in writing, |
12 | | // software distributed under the License is distributed on an |
13 | | // "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY |
14 | | // KIND, either express or implied. See the License for the |
15 | | // specific language governing permissions and limitations |
16 | | // under the License. |
17 | | |
18 | | #include "format/table/hive_reader.h" |
19 | | |
20 | | #include <vector> |
21 | | |
22 | | #include "common/status.h" |
23 | | #include "format/table/hive/hive_orc_nested_column_utils.h" |
24 | | #include "format/table/hive/hive_parquet_nested_column_utils.h" |
25 | | #include "format/table/nested_column_access_helper.h" |
26 | | #include "runtime/runtime_state.h" |
27 | | |
28 | | namespace doris { |
29 | | #include "common/compile_check_begin.h" |
30 | | |
31 | 11 | Status HiveOrcReader::on_before_init_reader(ReaderInitContext* ctx) { |
32 | 11 | _column_descs = ctx->column_descs; |
33 | 11 | _fill_col_name_to_block_idx = ctx->col_name_to_block_idx; |
34 | 11 | RETURN_IF_ERROR( |
35 | 11 | _extract_partition_values(*ctx->range, ctx->tuple_descriptor, _fill_partition_values)); |
36 | 99 | for (auto& desc : *ctx->column_descs) { |
37 | 99 | if (desc.category == ColumnCategory::REGULAR || |
38 | 99 | desc.category == ColumnCategory::GENERATED) { |
39 | 99 | ctx->column_names.push_back(desc.name); |
40 | 99 | } |
41 | 99 | } |
42 | | |
43 | | // Get file type (available because _create_file_reader() runs before this hook) |
44 | 11 | const orc::Type* orc_type_ptr = nullptr; |
45 | 11 | RETURN_IF_ERROR(get_file_type(&orc_type_ptr)); |
46 | 11 | bool is_hive_col_name = OrcReader::is_hive1_col_name(orc_type_ptr); |
47 | | |
48 | | // Build table_info_node based on config |
49 | 11 | if (get_state()->query_options().hive_orc_use_column_names && !is_hive_col_name) { |
50 | 11 | RETURN_IF_ERROR(BuildTableInfoUtil::by_orc_name(ctx->tuple_descriptor, orc_type_ptr, |
51 | 11 | ctx->table_info_node, _is_file_slot)); |
52 | 11 | } else { |
53 | 0 | ctx->table_info_node = std::make_shared<StructNode>(); |
54 | 0 | std::map<std::string, const SlotDescriptor*> slot_map; |
55 | 0 | for (const auto& slot : ctx->tuple_descriptor->slots()) { |
56 | 0 | slot_map.emplace(slot->col_name_lower_case(), slot); |
57 | 0 | } |
58 | |
|
59 | 0 | for (size_t idx = 0; idx < get_scan_params().column_idxs.size(); idx++) { |
60 | 0 | auto table_column_name = ctx->column_names[idx]; |
61 | 0 | auto file_index = get_scan_params().column_idxs[idx]; |
62 | |
|
63 | 0 | if (file_index >= orc_type_ptr->getSubtypeCount()) { |
64 | 0 | ctx->table_info_node->add_not_exist_children(table_column_name); |
65 | 0 | } else { |
66 | 0 | auto field_node = std::make_shared<Node>(); |
67 | 0 | RETURN_IF_ERROR(BuildTableInfoUtil::by_orc_name( |
68 | 0 | slot_map[table_column_name]->type(), orc_type_ptr->getSubtype(file_index), |
69 | 0 | field_node)); |
70 | 0 | ctx->table_info_node->add_children( |
71 | 0 | table_column_name, orc_type_ptr->getFieldName(file_index), field_node); |
72 | 0 | } |
73 | 0 | slot_map.erase(table_column_name); |
74 | 0 | } |
75 | 0 | for (const auto& [partition_col_name, _] : slot_map) { |
76 | 0 | ctx->table_info_node->add_not_exist_children(partition_col_name); |
77 | 0 | } |
78 | 0 | } |
79 | | |
80 | | // Compute column_ids |
81 | 11 | auto column_id_result = ColumnIdResult(); |
82 | 11 | if (get_state()->query_options().hive_orc_use_column_names && !is_hive_col_name) { |
83 | 11 | column_id_result = _create_column_ids(orc_type_ptr, ctx->tuple_descriptor); |
84 | 11 | } else { |
85 | 0 | column_id_result = |
86 | 0 | _create_column_ids_by_top_level_col_index(orc_type_ptr, ctx->tuple_descriptor); |
87 | 0 | } |
88 | 11 | ctx->column_ids = std::move(column_id_result.column_ids); |
89 | 11 | ctx->filter_column_ids = std::move(column_id_result.filter_column_ids); |
90 | | |
91 | | // _is_acid is false by default, no need to set explicitly |
92 | 11 | return Status::OK(); |
93 | 11 | } |
94 | | |
95 | | ColumnIdResult HiveOrcReader::_create_column_ids(const orc::Type* orc_type, |
96 | 17 | const TupleDescriptor* tuple_descriptor) { |
97 | | // map top-level table column name (lower-cased) -> orc::Type* |
98 | 17 | std::unordered_map<std::string, const orc::Type*> table_col_name_to_orc_type_map; |
99 | 164 | for (uint64_t i = 0; i < orc_type->getSubtypeCount(); ++i) { |
100 | 147 | auto orc_sub_type = orc_type->getSubtype(i); |
101 | 147 | if (!orc_sub_type) continue; |
102 | | |
103 | 147 | std::string table_col_name = to_lower(orc_type->getFieldName(i)); |
104 | 147 | table_col_name_to_orc_type_map[table_col_name] = orc_sub_type; |
105 | 147 | } |
106 | | |
107 | 17 | std::set<uint64_t> column_ids; |
108 | 17 | std::set<uint64_t> filter_column_ids; |
109 | | |
110 | | // helper to process access paths for a given top-level orc field |
111 | 17 | auto process_access_paths = [](const orc::Type* orc_field, |
112 | 17 | const std::vector<TColumnAccessPath>& access_paths, |
113 | 17 | std::set<uint64_t>& out_ids) { |
114 | 13 | process_nested_access_paths( |
115 | 13 | orc_field, access_paths, out_ids, |
116 | 13 | [](const orc::Type* type) { return type->getColumnId(); }, |
117 | 13 | [](const orc::Type* type) { return type->getMaximumColumnId(); }, |
118 | 13 | HiveOrcNestedColumnUtils::extract_nested_column_ids); |
119 | 13 | }; |
120 | | |
121 | 112 | for (const auto* slot : tuple_descriptor->slots()) { |
122 | 112 | auto it = table_col_name_to_orc_type_map.find(slot->col_name_lower_case()); |
123 | 112 | if (it == table_col_name_to_orc_type_map.end()) { |
124 | | // Column not found in file |
125 | 0 | continue; |
126 | 0 | } |
127 | 112 | const orc::Type* orc_field = it->second; |
128 | | |
129 | | // primitive (non-nested) types |
130 | 112 | if ((slot->col_type() != TYPE_STRUCT && slot->col_type() != TYPE_ARRAY && |
131 | 112 | slot->col_type() != TYPE_MAP)) { |
132 | 105 | column_ids.insert(orc_field->getColumnId()); |
133 | 105 | if (slot->is_predicate()) { |
134 | 0 | filter_column_ids.insert(orc_field->getColumnId()); |
135 | 0 | } |
136 | 105 | continue; |
137 | 105 | } |
138 | | |
139 | | // complex types |
140 | 7 | const auto& all_access_paths = slot->all_access_paths(); |
141 | 7 | process_access_paths(orc_field, all_access_paths, column_ids); |
142 | | |
143 | 7 | const auto& predicate_access_paths = slot->predicate_access_paths(); |
144 | 7 | if (!predicate_access_paths.empty()) { |
145 | 6 | process_access_paths(orc_field, predicate_access_paths, filter_column_ids); |
146 | 6 | } |
147 | 7 | } |
148 | | |
149 | 17 | return ColumnIdResult(std::move(column_ids), std::move(filter_column_ids)); |
150 | 17 | } |
151 | | |
152 | | ColumnIdResult HiveOrcReader::_create_column_ids_by_top_level_col_index( |
153 | 6 | const orc::Type* orc_type, const TupleDescriptor* tuple_descriptor) { |
154 | | // map top-level table column position -> orc::Type* |
155 | 6 | std::unordered_map<uint64_t, const orc::Type*> table_col_pos_to_orc_type_map; |
156 | 54 | for (uint64_t i = 0; i < orc_type->getSubtypeCount(); ++i) { |
157 | 48 | auto orc_sub_type = orc_type->getSubtype(i); |
158 | 48 | if (!orc_sub_type) continue; |
159 | | |
160 | 48 | table_col_pos_to_orc_type_map[i] = orc_sub_type; |
161 | 48 | } |
162 | | |
163 | 6 | std::set<uint64_t> column_ids; |
164 | 6 | std::set<uint64_t> filter_column_ids; |
165 | | |
166 | | // helper to process access paths for a given top-level orc field |
167 | 6 | auto process_access_paths = [](const orc::Type* orc_field, |
168 | 6 | const std::vector<TColumnAccessPath>& access_paths, |
169 | 13 | std::set<uint64_t>& out_ids) { |
170 | 13 | process_nested_access_paths( |
171 | 13 | orc_field, access_paths, out_ids, |
172 | 13 | [](const orc::Type* type) { return type->getColumnId(); }, |
173 | 13 | [](const orc::Type* type) { return type->getMaximumColumnId(); }, |
174 | 13 | HiveOrcNestedColumnUtils::extract_nested_column_ids); |
175 | 13 | }; |
176 | | |
177 | 13 | for (const auto* slot : tuple_descriptor->slots()) { |
178 | 13 | auto it = table_col_pos_to_orc_type_map.find(slot->col_pos()); |
179 | 13 | if (it == table_col_pos_to_orc_type_map.end()) { |
180 | | // Column not found in file |
181 | 0 | continue; |
182 | 0 | } |
183 | 13 | const orc::Type* orc_field = it->second; |
184 | | |
185 | | // primitive (non-nested) types |
186 | 13 | if ((slot->col_type() != TYPE_STRUCT && slot->col_type() != TYPE_ARRAY && |
187 | 13 | slot->col_type() != TYPE_MAP)) { |
188 | 6 | column_ids.insert(orc_field->getColumnId()); |
189 | 6 | if (slot->is_predicate()) { |
190 | 0 | filter_column_ids.insert(orc_field->getColumnId()); |
191 | 0 | } |
192 | 6 | continue; |
193 | 6 | } |
194 | | |
195 | 7 | const auto& all_access_paths = slot->all_access_paths(); |
196 | | // complex types |
197 | 7 | process_access_paths(orc_field, all_access_paths, column_ids); |
198 | | |
199 | 7 | const auto& predicate_access_paths = slot->predicate_access_paths(); |
200 | 7 | if (!predicate_access_paths.empty()) { |
201 | 6 | process_access_paths(orc_field, predicate_access_paths, filter_column_ids); |
202 | 6 | } |
203 | 7 | } |
204 | | |
205 | 6 | return ColumnIdResult(std::move(column_ids), std::move(filter_column_ids)); |
206 | 6 | } |
207 | | |
208 | 14 | Status HiveParquetReader::on_before_init_reader(ReaderInitContext* ctx) { |
209 | 14 | _column_descs = ctx->column_descs; |
210 | 14 | _fill_col_name_to_block_idx = ctx->col_name_to_block_idx; |
211 | 14 | RETURN_IF_ERROR( |
212 | 14 | _extract_partition_values(*ctx->range, ctx->tuple_descriptor, _fill_partition_values)); |
213 | 51 | for (auto& desc : *ctx->column_descs) { |
214 | 51 | if (desc.category == ColumnCategory::REGULAR || |
215 | 51 | desc.category == ColumnCategory::GENERATED) { |
216 | 51 | ctx->column_names.push_back(desc.name); |
217 | 51 | } |
218 | 51 | } |
219 | | |
220 | | // Get file metadata schema (available because _open_file() runs before this hook) |
221 | 14 | const FieldDescriptor* field_desc = nullptr; |
222 | 14 | RETURN_IF_ERROR(get_file_metadata_schema(&field_desc)); |
223 | 14 | DCHECK(field_desc != nullptr); |
224 | | |
225 | | // Build table_info_node based on config |
226 | 14 | if (get_state()->query_options().hive_parquet_use_column_names) { |
227 | 14 | RETURN_IF_ERROR(BuildTableInfoUtil::by_parquet_name(ctx->tuple_descriptor, *field_desc, |
228 | 14 | ctx->table_info_node, _is_file_slot)); |
229 | 14 | } else { |
230 | 0 | ctx->table_info_node = std::make_shared<StructNode>(); |
231 | 0 | std::map<std::string, const SlotDescriptor*> slot_map; |
232 | 0 | for (const auto& slot : ctx->tuple_descriptor->slots()) { |
233 | 0 | slot_map.emplace(slot->col_name_lower_case(), slot); |
234 | 0 | } |
235 | |
|
236 | 0 | auto parquet_fields_schema = field_desc->get_fields_schema(); |
237 | 0 | for (size_t idx = 0; idx < get_scan_params().column_idxs.size(); idx++) { |
238 | 0 | auto table_column_name = ctx->column_names[idx]; |
239 | 0 | auto file_index = get_scan_params().column_idxs[idx]; |
240 | |
|
241 | 0 | if (file_index >= parquet_fields_schema.size()) { |
242 | 0 | ctx->table_info_node->add_not_exist_children(table_column_name); |
243 | 0 | } else { |
244 | 0 | auto field_node = std::make_shared<Node>(); |
245 | 0 | RETURN_IF_ERROR(BuildTableInfoUtil::by_parquet_name( |
246 | 0 | slot_map[table_column_name]->type(), parquet_fields_schema[file_index], |
247 | 0 | field_node)); |
248 | 0 | ctx->table_info_node->add_children( |
249 | 0 | table_column_name, parquet_fields_schema[file_index].name, field_node); |
250 | 0 | } |
251 | 0 | slot_map.erase(table_column_name); |
252 | 0 | } |
253 | 0 | for (const auto& [partition_col_name, _] : slot_map) { |
254 | 0 | ctx->table_info_node->add_not_exist_children(partition_col_name); |
255 | 0 | } |
256 | 0 | } |
257 | | |
258 | | // Compute column_ids for lazy materialization |
259 | 14 | auto column_id_result = ColumnIdResult(); |
260 | 14 | if (get_state()->query_options().hive_parquet_use_column_names) { |
261 | 14 | column_id_result = _create_column_ids(field_desc, ctx->tuple_descriptor); |
262 | 14 | } else { |
263 | 0 | column_id_result = |
264 | 0 | _create_column_ids_by_top_level_col_index(field_desc, ctx->tuple_descriptor); |
265 | 0 | } |
266 | 14 | ctx->column_ids = std::move(column_id_result.column_ids); |
267 | 14 | ctx->filter_column_ids = std::move(column_id_result.filter_column_ids); |
268 | | |
269 | 14 | _filter_groups = true; |
270 | 14 | return Status::OK(); |
271 | 14 | } |
272 | | |
273 | | ColumnIdResult HiveParquetReader::_create_column_ids(const FieldDescriptor* field_desc, |
274 | 20 | const TupleDescriptor* tuple_descriptor) { |
275 | | // First, assign column IDs to the field descriptor |
276 | 20 | auto* mutable_field_desc = const_cast<FieldDescriptor*>(field_desc); |
277 | 20 | mutable_field_desc->assign_ids(); |
278 | | |
279 | | // map top-level table column name (lower-cased) -> FieldSchema* |
280 | 20 | std::unordered_map<std::string, const FieldSchema*> table_col_name_to_field_schema_map; |
281 | 278 | for (int i = 0; i < field_desc->size(); ++i) { |
282 | 258 | auto field_schema = field_desc->get_column(i); |
283 | 258 | if (!field_schema) continue; |
284 | | |
285 | 258 | table_col_name_to_field_schema_map[field_schema->lower_case_name] = field_schema; |
286 | 258 | } |
287 | | |
288 | 20 | std::set<uint64_t> column_ids; |
289 | 20 | std::set<uint64_t> filter_column_ids; |
290 | | |
291 | | // helper to process access paths for a given top-level parquet field |
292 | 20 | auto process_access_paths = [](const FieldSchema* parquet_field, |
293 | 20 | const std::vector<TColumnAccessPath>& access_paths, |
294 | 20 | std::set<uint64_t>& out_ids) { |
295 | 13 | process_nested_access_paths( |
296 | 13 | parquet_field, access_paths, out_ids, |
297 | 13 | [](const FieldSchema* field) { return field->get_column_id(); }, |
298 | 13 | [](const FieldSchema* field) { return field->get_max_column_id(); }, |
299 | 13 | HiveParquetNestedColumnUtils::extract_nested_column_ids); |
300 | 13 | }; |
301 | | |
302 | 64 | for (const auto* slot : tuple_descriptor->slots()) { |
303 | 64 | auto it = table_col_name_to_field_schema_map.find(slot->col_name_lower_case()); |
304 | 64 | if (it == table_col_name_to_field_schema_map.end()) { |
305 | | // Column not found in file |
306 | 0 | continue; |
307 | 0 | } |
308 | 64 | auto field_schema = it->second; |
309 | | |
310 | | // primitive (non-nested) types |
311 | 64 | if ((slot->col_type() != TYPE_STRUCT && slot->col_type() != TYPE_ARRAY && |
312 | 64 | slot->col_type() != TYPE_MAP)) { |
313 | 57 | column_ids.insert(field_schema->column_id); |
314 | | |
315 | 57 | if (slot->is_predicate()) { |
316 | 0 | filter_column_ids.insert(field_schema->column_id); |
317 | 0 | } |
318 | 57 | continue; |
319 | 57 | } |
320 | | |
321 | | // complex types |
322 | 7 | const auto& all_access_paths = slot->all_access_paths(); |
323 | 7 | process_access_paths(field_schema, all_access_paths, column_ids); |
324 | | |
325 | 7 | const auto& predicate_access_paths = slot->predicate_access_paths(); |
326 | 7 | if (!predicate_access_paths.empty()) { |
327 | 6 | process_access_paths(field_schema, predicate_access_paths, filter_column_ids); |
328 | 6 | } |
329 | 7 | } |
330 | | |
331 | 20 | return ColumnIdResult(std::move(column_ids), std::move(filter_column_ids)); |
332 | 20 | } |
333 | | |
334 | | ColumnIdResult HiveParquetReader::_create_column_ids_by_top_level_col_index( |
335 | 6 | const FieldDescriptor* field_desc, const TupleDescriptor* tuple_descriptor) { |
336 | | // First, assign column IDs to the field descriptor |
337 | 6 | auto* mutable_field_desc = const_cast<FieldDescriptor*>(field_desc); |
338 | 6 | mutable_field_desc->assign_ids(); |
339 | | |
340 | | // map top-level table column position -> FieldSchema* |
341 | 6 | std::unordered_map<uint64_t, const FieldSchema*> table_col_pos_to_field_schema_map; |
342 | 54 | for (int i = 0; i < field_desc->size(); ++i) { |
343 | 48 | auto field_schema = field_desc->get_column(i); |
344 | 48 | if (!field_schema) continue; |
345 | | |
346 | 48 | table_col_pos_to_field_schema_map[i] = field_schema; |
347 | 48 | } |
348 | | |
349 | 6 | std::set<uint64_t> column_ids; |
350 | 6 | std::set<uint64_t> filter_column_ids; |
351 | | |
352 | | // helper to process access paths for a given top-level parquet field |
353 | 6 | auto process_access_paths = [](const FieldSchema* parquet_field, |
354 | 6 | const std::vector<TColumnAccessPath>& access_paths, |
355 | 13 | std::set<uint64_t>& out_ids) { |
356 | 13 | process_nested_access_paths( |
357 | 13 | parquet_field, access_paths, out_ids, |
358 | 13 | [](const FieldSchema* field) { return field->get_column_id(); }, |
359 | 13 | [](const FieldSchema* field) { return field->get_max_column_id(); }, |
360 | 13 | HiveParquetNestedColumnUtils::extract_nested_column_ids); |
361 | 13 | }; |
362 | | |
363 | 13 | for (const auto* slot : tuple_descriptor->slots()) { |
364 | 13 | auto it = table_col_pos_to_field_schema_map.find(slot->col_pos()); |
365 | 13 | if (it == table_col_pos_to_field_schema_map.end()) { |
366 | | // Column not found in file |
367 | 0 | continue; |
368 | 0 | } |
369 | 13 | auto field_schema = it->second; |
370 | | |
371 | | // primitive (non-nested) types |
372 | 13 | if ((slot->col_type() != TYPE_STRUCT && slot->col_type() != TYPE_ARRAY && |
373 | 13 | slot->col_type() != TYPE_MAP)) { |
374 | 6 | column_ids.insert(field_schema->column_id); |
375 | | |
376 | 6 | if (slot->is_predicate()) { |
377 | 0 | filter_column_ids.insert(field_schema->column_id); |
378 | 0 | } |
379 | 6 | continue; |
380 | 6 | } |
381 | | |
382 | | // complex types |
383 | 7 | const auto& all_access_paths = slot->all_access_paths(); |
384 | 7 | process_access_paths(field_schema, all_access_paths, column_ids); |
385 | | |
386 | 7 | const auto& predicate_access_paths = slot->predicate_access_paths(); |
387 | 7 | if (!predicate_access_paths.empty()) { |
388 | 6 | process_access_paths(field_schema, predicate_access_paths, filter_column_ids); |
389 | 6 | } |
390 | 7 | } |
391 | | |
392 | 6 | return ColumnIdResult(std::move(column_ids), std::move(filter_column_ids)); |
393 | 6 | } |
394 | | |
395 | | #include "common/compile_check_end.h" |
396 | | } // namespace doris |