/root/doris/be/src/exec/tablet_info.cpp
Line | Count | Source (jump to first uncovered line) |
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 "exec/tablet_info.h" |
19 | | |
20 | | #include <butil/logging.h> |
21 | | #include <gen_cpp/Descriptors_types.h> |
22 | | #include <gen_cpp/Exprs_types.h> |
23 | | #include <gen_cpp/Partitions_types.h> |
24 | | #include <gen_cpp/Types_types.h> |
25 | | #include <gen_cpp/descriptors.pb.h> |
26 | | #include <gen_cpp/olap_file.pb.h> |
27 | | #include <glog/logging.h> |
28 | | |
29 | | #include <algorithm> |
30 | | #include <cstddef> |
31 | | #include <cstdint> |
32 | | #include <memory> |
33 | | #include <ostream> |
34 | | #include <string> |
35 | | #include <tuple> |
36 | | |
37 | | #include "common/exception.h" |
38 | | #include "common/logging.h" |
39 | | #include "common/status.h" |
40 | | #include "olap/tablet_schema.h" |
41 | | #include "runtime/define_primitive_type.h" |
42 | | #include "runtime/descriptors.h" |
43 | | #include "runtime/large_int_value.h" |
44 | | #include "runtime/memory/mem_tracker.h" |
45 | | #include "runtime/primitive_type.h" |
46 | | #include "runtime/raw_value.h" |
47 | | #include "runtime/types.h" |
48 | | #include "util/string_parser.hpp" |
49 | | #include "util/string_util.h" |
50 | | #include "vec/columns/column.h" |
51 | | #include "vec/data_types/data_type.h" |
52 | | #include "vec/data_types/data_type_factory.hpp" |
53 | | // NOLINTNEXTLINE(unused-includes) |
54 | | #include "vec/exprs/vexpr_context.h" // IWYU pragma: keep |
55 | | #include "vec/exprs/vliteral.h" |
56 | | #include "vec/runtime/vdatetime_value.h" |
57 | | |
58 | | namespace doris { |
59 | | |
60 | 32 | void OlapTableIndexSchema::to_protobuf(POlapTableIndexSchema* pindex) const { |
61 | 32 | pindex->set_id(index_id); |
62 | 32 | pindex->set_schema_hash(schema_hash); |
63 | 32 | for (auto* slot : slots) { |
64 | 0 | pindex->add_columns(slot->col_name()); |
65 | 0 | } |
66 | 32 | for (auto* column : columns) { |
67 | 0 | column->to_schema_pb(pindex->add_columns_desc()); |
68 | 0 | } |
69 | 32 | for (auto* index : indexes) { |
70 | 0 | index->to_schema_pb(pindex->add_indexes_desc()); |
71 | 0 | } |
72 | 32 | } |
73 | | |
74 | | bool VOlapTablePartKeyComparator::operator()(const BlockRowWithIndicator& lhs, |
75 | 0 | const BlockRowWithIndicator& rhs) const { |
76 | 0 | vectorized::Block* l_block = std::get<0>(lhs); |
77 | 0 | vectorized::Block* r_block = std::get<0>(rhs); |
78 | 0 | int32_t l_row = std::get<1>(lhs); |
79 | 0 | int32_t r_row = std::get<1>(rhs); |
80 | 0 | bool l_use_new = std::get<2>(lhs); |
81 | 0 | bool r_use_new = std::get<2>(rhs); |
82 | |
|
83 | 0 | VLOG_TRACE << '\n' << l_block->dump_data() << '\n' << r_block->dump_data(); |
84 | |
|
85 | 0 | if (l_row == -1) { |
86 | 0 | return false; |
87 | 0 | } else if (r_row == -1) { |
88 | 0 | return true; |
89 | 0 | } |
90 | | |
91 | 0 | if (_param_locs.empty()) { // no transform, use origin column |
92 | 0 | for (auto slot_loc : _slot_locs) { |
93 | 0 | auto res = l_block->get_by_position(slot_loc).column->compare_at( |
94 | 0 | l_row, r_row, *r_block->get_by_position(slot_loc).column, -1); |
95 | 0 | if (res != 0) { |
96 | 0 | return res < 0; |
97 | 0 | } |
98 | 0 | } |
99 | 0 | } else { // use transformed column to compare |
100 | 0 | DCHECK(_slot_locs.size() == _param_locs.size()) |
101 | 0 | << _slot_locs.size() << ' ' << _param_locs.size(); |
102 | |
|
103 | 0 | const std::vector<uint16_t>* l_index = l_use_new ? &_param_locs : &_slot_locs; |
104 | 0 | const std::vector<uint16_t>* r_index = r_use_new ? &_param_locs : &_slot_locs; |
105 | |
|
106 | 0 | for (int i = 0; i < _slot_locs.size(); i++) { |
107 | 0 | vectorized::ColumnPtr l_col = l_block->get_by_position((*l_index)[i]).column; |
108 | 0 | vectorized::ColumnPtr r_col = r_block->get_by_position((*r_index)[i]).column; |
109 | |
|
110 | 0 | auto res = l_col->compare_at(l_row, r_row, *r_col, -1); |
111 | 0 | if (res != 0) { |
112 | 0 | return res < 0; |
113 | 0 | } |
114 | 0 | } |
115 | 0 | } |
116 | | |
117 | | // equal, return false |
118 | 0 | return false; |
119 | 0 | } |
120 | | |
121 | 14 | Status OlapTableSchemaParam::init(const POlapTableSchemaParam& pschema) { |
122 | 14 | _db_id = pschema.db_id(); |
123 | 14 | _table_id = pschema.table_id(); |
124 | 14 | _version = pschema.version(); |
125 | 14 | if (pschema.has_unique_key_update_mode()) { |
126 | 14 | _unique_key_update_mode = pschema.unique_key_update_mode(); |
127 | 14 | if (pschema.has_sequence_map_col_unique_id()) { |
128 | 14 | _sequence_map_col_uid = pschema.sequence_map_col_unique_id(); |
129 | 14 | } |
130 | 14 | } else { |
131 | | // for backward compatibility |
132 | 0 | if (pschema.has_partial_update() && pschema.partial_update()) { |
133 | 0 | _unique_key_update_mode = UniqueKeyUpdateModePB::UPDATE_FIXED_COLUMNS; |
134 | 0 | } else { |
135 | 0 | _unique_key_update_mode = UniqueKeyUpdateModePB::UPSERT; |
136 | 0 | } |
137 | 0 | } |
138 | 14 | _is_strict_mode = pschema.is_strict_mode(); |
139 | 14 | if (_unique_key_update_mode == UniqueKeyUpdateModePB::UPDATE_FIXED_COLUMNS) { |
140 | 0 | _auto_increment_column = pschema.auto_increment_column(); |
141 | 0 | if (!_auto_increment_column.empty() && pschema.auto_increment_column_unique_id() == -1) { |
142 | 0 | return Status::InternalError( |
143 | 0 | "Auto increment column id is not set in FE. Maybe FE is an older version " |
144 | 0 | "different from BE."); |
145 | 0 | } |
146 | 0 | _auto_increment_column_unique_id = pschema.auto_increment_column_unique_id(); |
147 | 0 | } |
148 | 14 | _timestamp_ms = pschema.timestamp_ms(); |
149 | 14 | if (pschema.has_nano_seconds()) { |
150 | 14 | _nano_seconds = pschema.nano_seconds(); |
151 | 14 | } |
152 | 14 | _timezone = pschema.timezone(); |
153 | | |
154 | 14 | for (const auto& col : pschema.partial_update_input_columns()) { |
155 | 0 | _partial_update_input_columns.insert(col); |
156 | 0 | } |
157 | 14 | std::unordered_map<std::string, SlotDescriptor*> slots_map; |
158 | | |
159 | 14 | _tuple_desc = _obj_pool.add(new TupleDescriptor(pschema.tuple_desc())); |
160 | | |
161 | 84 | for (const auto& p_slot_desc : pschema.slot_descs()) { |
162 | 84 | auto* slot_desc = _obj_pool.add(new SlotDescriptor(p_slot_desc)); |
163 | 84 | _tuple_desc->add_slot(slot_desc); |
164 | 84 | string data_type; |
165 | 84 | EnumToString(TPrimitiveType, to_thrift(slot_desc->col_type()), data_type); |
166 | 84 | std::string is_null_str = slot_desc->is_nullable() ? "true" : "false"; |
167 | 84 | std::string data_type_str = |
168 | 84 | std::to_string(int64_t(TabletColumn::get_field_type_by_string(data_type))); |
169 | 84 | slots_map.emplace(to_lower(slot_desc->col_name()) + "+" + data_type_str + is_null_str, |
170 | 84 | slot_desc); |
171 | 84 | } |
172 | | |
173 | 28 | for (const auto& p_index : pschema.indexes()) { |
174 | 28 | auto* index = _obj_pool.add(new OlapTableIndexSchema()); |
175 | 28 | index->index_id = p_index.id(); |
176 | 28 | index->schema_hash = p_index.schema_hash(); |
177 | 28 | for (const auto& pcolumn_desc : p_index.columns_desc()) { |
178 | 0 | if (_unique_key_update_mode != UniqueKeyUpdateModePB::UPDATE_FIXED_COLUMNS || |
179 | 0 | _partial_update_input_columns.contains(pcolumn_desc.name())) { |
180 | 0 | std::string is_null_str = pcolumn_desc.is_nullable() ? "true" : "false"; |
181 | 0 | std::string data_type_str = std::to_string( |
182 | 0 | int64_t(TabletColumn::get_field_type_by_string(pcolumn_desc.type()))); |
183 | 0 | auto it = slots_map.find(to_lower(pcolumn_desc.name()) + "+" + data_type_str + |
184 | 0 | is_null_str); |
185 | 0 | if (it == std::end(slots_map)) { |
186 | 0 | std::string keys {}; |
187 | 0 | for (const auto& [key, _] : slots_map) { |
188 | 0 | keys += fmt::format("{},", key); |
189 | 0 | } |
190 | 0 | LOG_EVERY_SECOND(WARNING) << fmt::format( |
191 | 0 | "[OlapTableSchemaParam::init(const POlapTableSchemaParam& pschema)]: " |
192 | 0 | "unknown index column, column={}, type={}, data_type_str={}, " |
193 | 0 | "is_null_str={}, slots_map.keys()=[{}], {}\npschema={}", |
194 | 0 | pcolumn_desc.name(), pcolumn_desc.type(), data_type_str, is_null_str, |
195 | 0 | keys, debug_string(), pschema.ShortDebugString()); |
196 | |
|
197 | 0 | return Status::InternalError("unknown index column, column={}, type={}", |
198 | 0 | pcolumn_desc.name(), pcolumn_desc.type()); |
199 | 0 | } |
200 | 0 | index->slots.emplace_back(it->second); |
201 | 0 | } |
202 | 0 | TabletColumn* tc = _obj_pool.add(new TabletColumn()); |
203 | 0 | tc->init_from_pb(pcolumn_desc); |
204 | 0 | index->columns.emplace_back(tc); |
205 | 0 | } |
206 | 28 | for (const auto& pindex_desc : p_index.indexes_desc()) { |
207 | 0 | TabletIndex* ti = _obj_pool.add(new TabletIndex()); |
208 | 0 | ti->init_from_pb(pindex_desc); |
209 | 0 | index->indexes.emplace_back(ti); |
210 | 0 | } |
211 | 28 | _indexes.emplace_back(index); |
212 | 28 | } |
213 | | |
214 | 14 | std::sort(_indexes.begin(), _indexes.end(), |
215 | 28 | [](const OlapTableIndexSchema* lhs, const OlapTableIndexSchema* rhs) { |
216 | 28 | return lhs->index_id < rhs->index_id; |
217 | 28 | }); |
218 | 14 | return Status::OK(); |
219 | 14 | } |
220 | | |
221 | 16 | Status OlapTableSchemaParam::init_unique_key_update_mode(const TOlapTableSchemaParam& tschema) { |
222 | 16 | if (tschema.__isset.unique_key_update_mode) { |
223 | 0 | switch (tschema.unique_key_update_mode) { |
224 | 0 | case doris::TUniqueKeyUpdateMode::UPSERT: { |
225 | 0 | _unique_key_update_mode = UniqueKeyUpdateModePB::UPSERT; |
226 | 0 | break; |
227 | 0 | } |
228 | 0 | case doris::TUniqueKeyUpdateMode::UPDATE_FIXED_COLUMNS: { |
229 | 0 | _unique_key_update_mode = UniqueKeyUpdateModePB::UPDATE_FIXED_COLUMNS; |
230 | 0 | break; |
231 | 0 | } |
232 | 0 | case doris::TUniqueKeyUpdateMode::UPDATE_FLEXIBLE_COLUMNS: { |
233 | 0 | _unique_key_update_mode = UniqueKeyUpdateModePB::UPDATE_FLEXIBLE_COLUMNS; |
234 | 0 | break; |
235 | 0 | } |
236 | 0 | default: { |
237 | 0 | return Status::InternalError( |
238 | 0 | "Unknown unique_key_update_mode: {}, should be one of " |
239 | 0 | "UPSERT/UPDATE_FIXED_COLUMNS/UPDATE_FLEXIBLE_COLUMNS", |
240 | 0 | tschema.unique_key_update_mode); |
241 | 0 | } |
242 | 0 | } |
243 | 0 | if (tschema.__isset.sequence_map_col_unique_id) { |
244 | 0 | _sequence_map_col_uid = tschema.sequence_map_col_unique_id; |
245 | 0 | } |
246 | 16 | } else { |
247 | | // for backward compatibility |
248 | 16 | if (tschema.__isset.is_partial_update && tschema.is_partial_update) { |
249 | 0 | _unique_key_update_mode = UniqueKeyUpdateModePB::UPDATE_FIXED_COLUMNS; |
250 | 16 | } else { |
251 | 16 | _unique_key_update_mode = UniqueKeyUpdateModePB::UPSERT; |
252 | 16 | } |
253 | 16 | } |
254 | 16 | return Status::OK(); |
255 | 16 | } |
256 | | |
257 | 16 | Status OlapTableSchemaParam::init(const TOlapTableSchemaParam& tschema) { |
258 | 16 | _db_id = tschema.db_id; |
259 | 16 | _table_id = tschema.table_id; |
260 | 16 | _version = tschema.version; |
261 | 16 | RETURN_IF_ERROR(init_unique_key_update_mode(tschema)); |
262 | 16 | if (tschema.__isset.is_strict_mode) { |
263 | 16 | _is_strict_mode = tschema.is_strict_mode; |
264 | 16 | } |
265 | 16 | if (_unique_key_update_mode == UniqueKeyUpdateModePB::UPDATE_FIXED_COLUMNS) { |
266 | 0 | _auto_increment_column = tschema.auto_increment_column; |
267 | 0 | if (!_auto_increment_column.empty() && tschema.auto_increment_column_unique_id == -1) { |
268 | 0 | return Status::InternalError( |
269 | 0 | "Auto increment column id is not set in FE. Maybe FE is an older version " |
270 | 0 | "different from BE."); |
271 | 0 | } |
272 | 0 | _auto_increment_column_unique_id = tschema.auto_increment_column_unique_id; |
273 | 0 | } |
274 | | |
275 | 16 | for (const auto& tcolumn : tschema.partial_update_input_columns) { |
276 | 0 | _partial_update_input_columns.insert(tcolumn); |
277 | 0 | } |
278 | 16 | std::unordered_map<std::string, SlotDescriptor*> slots_map; |
279 | 16 | _tuple_desc = _obj_pool.add(new TupleDescriptor(tschema.tuple_desc)); |
280 | 96 | for (const auto& t_slot_desc : tschema.slot_descs) { |
281 | 96 | auto* slot_desc = _obj_pool.add(new SlotDescriptor(t_slot_desc)); |
282 | 96 | _tuple_desc->add_slot(slot_desc); |
283 | 96 | std::string is_null_str = slot_desc->is_nullable() ? "true" : "false"; |
284 | 96 | std::string data_type_str = std::to_string(int64_t(slot_desc->col_type())); |
285 | 96 | slots_map.emplace(to_lower(slot_desc->col_name()) + "+" + data_type_str + is_null_str, |
286 | 96 | slot_desc); |
287 | 96 | } |
288 | | |
289 | 32 | for (const auto& t_index : tschema.indexes) { |
290 | 32 | std::unordered_map<std::string, int32_t> index_slots_map; |
291 | 32 | auto* index = _obj_pool.add(new OlapTableIndexSchema()); |
292 | 32 | index->index_id = t_index.id; |
293 | 32 | index->schema_hash = t_index.schema_hash; |
294 | 32 | for (const auto& tcolumn_desc : t_index.columns_desc) { |
295 | 0 | if (_unique_key_update_mode != UniqueKeyUpdateModePB::UPDATE_FIXED_COLUMNS || |
296 | 0 | _partial_update_input_columns.contains(tcolumn_desc.column_name)) { |
297 | 0 | std::string is_null_str = tcolumn_desc.is_allow_null ? "true" : "false"; |
298 | 0 | std::string data_type_str = |
299 | 0 | std::to_string(int64_t(thrift_to_type(tcolumn_desc.column_type.type))); |
300 | 0 | auto it = slots_map.find(to_lower(tcolumn_desc.column_name) + "+" + data_type_str + |
301 | 0 | is_null_str); |
302 | 0 | if (it == slots_map.end()) { |
303 | 0 | std::stringstream ss; |
304 | 0 | ss << tschema; |
305 | 0 | std::string keys {}; |
306 | 0 | for (const auto& [key, _] : slots_map) { |
307 | 0 | keys += fmt::format("{},", key); |
308 | 0 | } |
309 | 0 | LOG_EVERY_SECOND(WARNING) << fmt::format( |
310 | 0 | "[OlapTableSchemaParam::init(const TOlapTableSchemaParam& tschema)]: " |
311 | 0 | "unknown index column, column={}, type={}, data_type_str={}, " |
312 | 0 | "is_null_str={}, slots_map.keys()=[{}], {}\ntschema={}", |
313 | 0 | tcolumn_desc.column_name, tcolumn_desc.column_type.type, data_type_str, |
314 | 0 | is_null_str, keys, debug_string(), ss.str()); |
315 | 0 | return Status::InternalError("unknown index column, column={}, type={}", |
316 | 0 | tcolumn_desc.column_name, |
317 | 0 | tcolumn_desc.column_type.type); |
318 | 0 | } |
319 | 0 | index->slots.emplace_back(it->second); |
320 | 0 | } |
321 | 0 | index_slots_map.emplace(to_lower(tcolumn_desc.column_name), tcolumn_desc.col_unique_id); |
322 | 0 | TabletColumn* tc = _obj_pool.add(new TabletColumn()); |
323 | 0 | tc->init_from_thrift(tcolumn_desc); |
324 | 0 | index->columns.emplace_back(tc); |
325 | 0 | } |
326 | 32 | if (t_index.__isset.indexes_desc) { |
327 | 0 | for (const auto& tindex_desc : t_index.indexes_desc) { |
328 | 0 | std::vector<int32_t> column_unique_ids(tindex_desc.columns.size()); |
329 | 0 | for (size_t i = 0; i < tindex_desc.columns.size(); i++) { |
330 | 0 | auto it = index_slots_map.find(to_lower(tindex_desc.columns[i])); |
331 | 0 | if (it != index_slots_map.end()) { |
332 | 0 | column_unique_ids[i] = it->second; |
333 | 0 | } |
334 | 0 | } |
335 | 0 | TabletIndex* ti = _obj_pool.add(new TabletIndex()); |
336 | 0 | ti->init_from_thrift(tindex_desc, column_unique_ids); |
337 | 0 | index->indexes.emplace_back(ti); |
338 | 0 | } |
339 | 0 | } |
340 | 32 | if (t_index.__isset.where_clause) { |
341 | 0 | RETURN_IF_ERROR( |
342 | 0 | vectorized::VExpr::create_expr_tree(t_index.where_clause, index->where_clause)); |
343 | 0 | } |
344 | 32 | _indexes.emplace_back(index); |
345 | 32 | } |
346 | | |
347 | 16 | std::sort(_indexes.begin(), _indexes.end(), |
348 | 32 | [](const OlapTableIndexSchema* lhs, const OlapTableIndexSchema* rhs) { |
349 | 32 | return lhs->index_id < rhs->index_id; |
350 | 32 | }); |
351 | 16 | return Status::OK(); |
352 | 16 | } |
353 | | |
354 | 16 | void OlapTableSchemaParam::to_protobuf(POlapTableSchemaParam* pschema) const { |
355 | 16 | pschema->set_db_id(_db_id); |
356 | 16 | pschema->set_table_id(_table_id); |
357 | 16 | pschema->set_version(_version); |
358 | 16 | pschema->set_unique_key_update_mode(_unique_key_update_mode); |
359 | 16 | if (_unique_key_update_mode == UniqueKeyUpdateModePB::UPDATE_FIXED_COLUMNS) { |
360 | | // for backward compatibility |
361 | 0 | pschema->set_partial_update(true); |
362 | 0 | } |
363 | 16 | pschema->set_is_strict_mode(_is_strict_mode); |
364 | 16 | pschema->set_auto_increment_column(_auto_increment_column); |
365 | 16 | pschema->set_auto_increment_column_unique_id(_auto_increment_column_unique_id); |
366 | 16 | pschema->set_timestamp_ms(_timestamp_ms); |
367 | 16 | pschema->set_timezone(_timezone); |
368 | 16 | pschema->set_nano_seconds(_nano_seconds); |
369 | 16 | pschema->set_sequence_map_col_unique_id(_sequence_map_col_uid); |
370 | 16 | for (auto col : _partial_update_input_columns) { |
371 | 0 | *pschema->add_partial_update_input_columns() = col; |
372 | 0 | } |
373 | 16 | _tuple_desc->to_protobuf(pschema->mutable_tuple_desc()); |
374 | 96 | for (auto* slot : _tuple_desc->slots()) { |
375 | 96 | slot->to_protobuf(pschema->add_slot_descs()); |
376 | 96 | } |
377 | 32 | for (auto* index : _indexes) { |
378 | 32 | index->to_protobuf(pschema->add_indexes()); |
379 | 32 | } |
380 | 16 | } |
381 | | |
382 | 0 | std::string OlapTableSchemaParam::debug_string() const { |
383 | 0 | std::stringstream ss; |
384 | 0 | ss << "tuple_desc=" << _tuple_desc->debug_string(); |
385 | 0 | return ss.str(); |
386 | 0 | } |
387 | | |
388 | | VOlapTablePartitionParam::VOlapTablePartitionParam(std::shared_ptr<OlapTableSchemaParam>& schema, |
389 | | const TOlapTablePartitionParam& t_param) |
390 | | : _schema(schema), |
391 | | _t_param(t_param), |
392 | | _slots(_schema->tuple_desc()->slots()), |
393 | | _mem_tracker(std::make_unique<MemTracker>("OlapTablePartitionParam")), |
394 | 0 | _part_type(t_param.partition_type) { |
395 | 0 | if (t_param.__isset.enable_automatic_partition && t_param.enable_automatic_partition) { |
396 | 0 | _is_auto_partition = true; |
397 | 0 | auto size = t_param.partition_function_exprs.size(); |
398 | 0 | _part_func_ctx.resize(size); |
399 | 0 | _partition_function.resize(size); |
400 | 0 | DCHECK((t_param.partition_type == TPartitionType::RANGE_PARTITIONED && size == 1) || |
401 | 0 | (t_param.partition_type == TPartitionType::LIST_PARTITIONED && size >= 1)) |
402 | 0 | << "now support only 1 partition column for auto range partitions. " |
403 | 0 | << t_param.partition_type << " " << size; |
404 | 0 | for (int i = 0; i < size; ++i) { |
405 | 0 | Status st = vectorized::VExpr::create_expr_tree(t_param.partition_function_exprs[i], |
406 | 0 | _part_func_ctx[i]); |
407 | 0 | if (!st.ok()) { |
408 | 0 | throw Exception(Status::InternalError("Partition function expr is not valid"), |
409 | 0 | "Partition function expr is not valid"); |
410 | 0 | } |
411 | 0 | _partition_function[i] = _part_func_ctx[i]->root(); |
412 | 0 | } |
413 | 0 | } |
414 | | |
415 | 0 | if (t_param.__isset.enable_auto_detect_overwrite && t_param.enable_auto_detect_overwrite) { |
416 | 0 | _is_auto_detect_overwrite = true; |
417 | 0 | DCHECK(t_param.__isset.overwrite_group_id); |
418 | 0 | _overwrite_group_id = t_param.overwrite_group_id; |
419 | 0 | } |
420 | |
|
421 | 0 | if (_is_auto_partition) { |
422 | | // the nullable mode depends on partition_exprs. not column slots. so use them. |
423 | 0 | DCHECK(_partition_function.size() <= _slots.size()) |
424 | 0 | << _partition_function.size() << ", " << _slots.size(); |
425 | | |
426 | | // suppose (k0, [k1], [k2]), so get [k1, 0], [k2, 1] |
427 | 0 | std::map<std::string, int> partition_slots_map; // name to idx in part_exprs |
428 | 0 | for (size_t i = 0; i < t_param.partition_columns.size(); i++) { |
429 | 0 | partition_slots_map.emplace(t_param.partition_columns[i], i); |
430 | 0 | } |
431 | | |
432 | | // here we rely on the same order and number of the _part_funcs and _slots in the prefix |
433 | | // _part_block contains all slots of table. |
434 | 0 | for (auto* slot : _slots) { |
435 | | // try to replace with partition expr. |
436 | 0 | if (auto it = partition_slots_map.find(slot->col_name()); |
437 | 0 | it != partition_slots_map.end()) { // it's a partition column slot |
438 | 0 | auto& expr_type = _partition_function[it->second]->data_type(); |
439 | 0 | _partition_block.insert({expr_type->create_column(), expr_type, slot->col_name()}); |
440 | 0 | } else { |
441 | 0 | _partition_block.insert({slot->get_empty_mutable_column(), |
442 | 0 | slot->get_data_type_ptr(), slot->col_name()}); |
443 | 0 | } |
444 | 0 | } |
445 | 0 | VLOG_TRACE << _partition_block.dump_structure(); |
446 | 0 | } else { |
447 | | // we insert all. but not all will be used. it will controlled by _partition_slot_locs |
448 | 0 | for (auto* slot : _slots) { |
449 | 0 | _partition_block.insert({slot->get_empty_mutable_column(), slot->get_data_type_ptr(), |
450 | 0 | slot->col_name()}); |
451 | 0 | } |
452 | 0 | } |
453 | 0 | } |
454 | | |
455 | 0 | VOlapTablePartitionParam::~VOlapTablePartitionParam() { |
456 | 0 | _mem_tracker->release(_mem_usage); |
457 | 0 | } |
458 | | |
459 | 0 | Status VOlapTablePartitionParam::init() { |
460 | 0 | std::vector<std::string> slot_column_names; |
461 | 0 | for (auto* slot_desc : _schema->tuple_desc()->slots()) { |
462 | 0 | slot_column_names.emplace_back(slot_desc->col_name()); |
463 | 0 | } |
464 | |
|
465 | 0 | auto find_slot_locs = [&slot_column_names](const std::string& slot_name, |
466 | 0 | std::vector<uint16_t>& locs, |
467 | 0 | const std::string& column_type) { |
468 | 0 | auto it = std::find(slot_column_names.begin(), slot_column_names.end(), slot_name); |
469 | 0 | if (it == slot_column_names.end()) { |
470 | 0 | return Status::InternalError("{} column not found, column ={}", column_type, slot_name); |
471 | 0 | } |
472 | 0 | locs.emplace_back(it - slot_column_names.begin()); |
473 | 0 | return Status::OK(); |
474 | 0 | }; |
475 | | |
476 | | // here we find the partition columns. others maybe non-partition columns/special columns. |
477 | 0 | if (_t_param.__isset.partition_columns) { |
478 | 0 | for (auto& part_col : _t_param.partition_columns) { |
479 | 0 | RETURN_IF_ERROR(find_slot_locs(part_col, _partition_slot_locs, "partition")); |
480 | 0 | } |
481 | 0 | } |
482 | | |
483 | 0 | _partitions_map = std::make_unique< |
484 | 0 | std::map<BlockRowWithIndicator, VOlapTablePartition*, VOlapTablePartKeyComparator>>( |
485 | 0 | VOlapTablePartKeyComparator(_partition_slot_locs, _transformed_slot_locs)); |
486 | 0 | if (_t_param.__isset.distributed_columns) { |
487 | 0 | for (auto& col : _t_param.distributed_columns) { |
488 | 0 | RETURN_IF_ERROR(find_slot_locs(col, _distributed_slot_locs, "distributed")); |
489 | 0 | } |
490 | 0 | } |
491 | | |
492 | | // for both auto/non-auto partition table. |
493 | 0 | _is_in_partition = _part_type == TPartitionType::type::LIST_PARTITIONED; |
494 | | |
495 | | // initial partitions. if meet dummy partitions only for open BE nodes, not generate key of them for finding |
496 | 0 | for (const auto& t_part : _t_param.partitions) { |
497 | 0 | VOlapTablePartition* part = nullptr; |
498 | 0 | RETURN_IF_ERROR(generate_partition_from(t_part, part)); |
499 | 0 | _partitions.emplace_back(part); |
500 | |
|
501 | 0 | if (!_t_param.partitions_is_fake) { |
502 | 0 | if (_is_in_partition) { |
503 | 0 | for (auto& in_key : part->in_keys) { |
504 | 0 | _partitions_map->emplace(std::tuple {in_key.first, in_key.second, false}, part); |
505 | 0 | } |
506 | 0 | } else { |
507 | 0 | _partitions_map->emplace( |
508 | 0 | std::tuple {part->end_key.first, part->end_key.second, false}, part); |
509 | 0 | } |
510 | 0 | } |
511 | 0 | } |
512 | | |
513 | 0 | _mem_usage = _partition_block.allocated_bytes(); |
514 | 0 | _mem_tracker->consume(_mem_usage); |
515 | 0 | return Status::OK(); |
516 | 0 | } |
517 | | |
518 | | bool VOlapTablePartitionParam::_part_contains(VOlapTablePartition* part, |
519 | 0 | BlockRowWithIndicator key) const { |
520 | 0 | VOlapTablePartKeyComparator comparator(_partition_slot_locs, _transformed_slot_locs); |
521 | | // we have used upper_bound to find to ensure key < part.right and this part is closest(right - key is min) |
522 | | // now we only have to check (key >= part.left). the comparator(a,b) means a < b, so we use anti |
523 | 0 | return part->start_key.second == -1 /* spj: start_key.second == -1 means only single partition*/ |
524 | 0 | || !comparator(key, std::tuple {part->start_key.first, part->start_key.second, false}); |
525 | 0 | } |
526 | | |
527 | | // insert value into _partition_block's column |
528 | | // NOLINTBEGIN(readability-function-size) |
529 | 0 | static Status _create_partition_key(const TExprNode& t_expr, BlockRow* part_key, uint16_t pos) { |
530 | 0 | auto column = std::move(*part_key->first->get_by_position(pos).column).mutate(); |
531 | | //TODO: use assert_cast before insert_data |
532 | 0 | switch (t_expr.node_type) { |
533 | 0 | case TExprNodeType::DATE_LITERAL: { |
534 | 0 | if (vectorized::DataTypeFactory::instance() |
535 | 0 | .create_data_type(t_expr.type) |
536 | 0 | ->get_primitive_type() == TYPE_DATEV2) { |
537 | 0 | DateV2Value<DateV2ValueType> dt; |
538 | 0 | if (!dt.from_date_str(t_expr.date_literal.value.c_str(), |
539 | 0 | t_expr.date_literal.value.size())) { |
540 | 0 | std::stringstream ss; |
541 | 0 | ss << "invalid date literal in partition column, date=" << t_expr.date_literal; |
542 | 0 | return Status::InternalError(ss.str()); |
543 | 0 | } |
544 | 0 | column->insert_data(reinterpret_cast<const char*>(&dt), 0); |
545 | 0 | } else if (vectorized::DataTypeFactory::instance() |
546 | 0 | .create_data_type(t_expr.type) |
547 | 0 | ->get_primitive_type() == TYPE_DATETIMEV2) { |
548 | 0 | DateV2Value<DateTimeV2ValueType> dt; |
549 | 0 | const int32_t scale = |
550 | 0 | t_expr.type.types.empty() ? -1 : t_expr.type.types.front().scalar_type.scale; |
551 | 0 | if (!dt.from_date_str(t_expr.date_literal.value.c_str(), |
552 | 0 | t_expr.date_literal.value.size(), scale)) { |
553 | 0 | std::stringstream ss; |
554 | 0 | ss << "invalid date literal in partition column, date=" << t_expr.date_literal; |
555 | 0 | return Status::InternalError(ss.str()); |
556 | 0 | } |
557 | 0 | column->insert_data(reinterpret_cast<const char*>(&dt), 0); |
558 | 0 | } else { |
559 | 0 | VecDateTimeValue dt; |
560 | 0 | if (!dt.from_date_str(t_expr.date_literal.value.c_str(), |
561 | 0 | t_expr.date_literal.value.size())) { |
562 | 0 | std::stringstream ss; |
563 | 0 | ss << "invalid date literal in partition column, date=" << t_expr.date_literal; |
564 | 0 | return Status::InternalError(ss.str()); |
565 | 0 | } |
566 | 0 | column->insert_data(reinterpret_cast<const char*>(&dt), 0); |
567 | 0 | } |
568 | 0 | break; |
569 | 0 | } |
570 | 0 | case TExprNodeType::INT_LITERAL: { |
571 | 0 | switch (t_expr.type.types[0].scalar_type.type) { |
572 | 0 | case TPrimitiveType::TINYINT: { |
573 | 0 | int8_t value = t_expr.int_literal.value; |
574 | 0 | column->insert_data(reinterpret_cast<const char*>(&value), 0); |
575 | 0 | break; |
576 | 0 | } |
577 | 0 | case TPrimitiveType::SMALLINT: { |
578 | 0 | int16_t value = t_expr.int_literal.value; |
579 | 0 | column->insert_data(reinterpret_cast<const char*>(&value), 0); |
580 | 0 | break; |
581 | 0 | } |
582 | 0 | case TPrimitiveType::INT: { |
583 | 0 | int32_t value = t_expr.int_literal.value; |
584 | 0 | column->insert_data(reinterpret_cast<const char*>(&value), 0); |
585 | 0 | break; |
586 | 0 | } |
587 | 0 | default: |
588 | 0 | int64_t value = t_expr.int_literal.value; |
589 | 0 | column->insert_data(reinterpret_cast<const char*>(&value), 0); |
590 | 0 | } |
591 | 0 | break; |
592 | 0 | } |
593 | 0 | case TExprNodeType::LARGE_INT_LITERAL: { |
594 | 0 | StringParser::ParseResult parse_result = StringParser::PARSE_SUCCESS; |
595 | 0 | auto value = StringParser::string_to_int<__int128>(t_expr.large_int_literal.value.c_str(), |
596 | 0 | t_expr.large_int_literal.value.size(), |
597 | 0 | &parse_result); |
598 | 0 | if (parse_result != StringParser::PARSE_SUCCESS) { |
599 | 0 | value = MAX_INT128; |
600 | 0 | } |
601 | 0 | column->insert_data(reinterpret_cast<const char*>(&value), 0); |
602 | 0 | break; |
603 | 0 | } |
604 | 0 | case TExprNodeType::STRING_LITERAL: { |
605 | 0 | int len = t_expr.string_literal.value.size(); |
606 | 0 | const char* str_val = t_expr.string_literal.value.c_str(); |
607 | 0 | column->insert_data(str_val, len); |
608 | 0 | break; |
609 | 0 | } |
610 | 0 | case TExprNodeType::BOOL_LITERAL: { |
611 | 0 | column->insert_data(reinterpret_cast<const char*>(&t_expr.bool_literal.value), 0); |
612 | 0 | break; |
613 | 0 | } |
614 | 0 | case TExprNodeType::NULL_LITERAL: { |
615 | | // insert a null literal |
616 | 0 | if (!column->is_nullable()) { |
617 | | // https://github.com/apache/doris/pull/39449 have forbid this cause. always add this check as protective measures |
618 | 0 | return Status::InternalError("The column {} is not null, can't insert into NULL value.", |
619 | 0 | part_key->first->get_by_position(pos).name); |
620 | 0 | } |
621 | 0 | column->insert_data(nullptr, 0); |
622 | 0 | break; |
623 | 0 | } |
624 | 0 | default: { |
625 | 0 | return Status::InternalError("unsupported partition column node type, type={}", |
626 | 0 | t_expr.node_type); |
627 | 0 | } |
628 | 0 | } |
629 | 0 | part_key->second = column->size() - 1; |
630 | 0 | return Status::OK(); |
631 | 0 | } |
632 | | // NOLINTEND(readability-function-size) |
633 | | |
634 | | Status VOlapTablePartitionParam::_create_partition_keys(const std::vector<TExprNode>& t_exprs, |
635 | 0 | BlockRow* part_key) { |
636 | 0 | for (int i = 0; i < t_exprs.size(); i++) { |
637 | 0 | RETURN_IF_ERROR(_create_partition_key(t_exprs[i], part_key, _partition_slot_locs[i])); |
638 | 0 | } |
639 | 0 | return Status::OK(); |
640 | 0 | } |
641 | | |
642 | | Status VOlapTablePartitionParam::generate_partition_from(const TOlapTablePartition& t_part, |
643 | 0 | VOlapTablePartition*& part_result) { |
644 | 0 | DCHECK(part_result == nullptr); |
645 | | // here we set the default value of partition bounds first! if it doesn't have some key, it will be -1. |
646 | 0 | part_result = _obj_pool.add(new VOlapTablePartition(&_partition_block)); |
647 | 0 | part_result->id = t_part.id; |
648 | 0 | part_result->is_mutable = t_part.is_mutable; |
649 | | // only load_to_single_tablet = true will set load_tablet_idx |
650 | 0 | if (t_part.__isset.load_tablet_idx) { |
651 | 0 | part_result->load_tablet_idx = t_part.load_tablet_idx; |
652 | 0 | } |
653 | |
|
654 | 0 | if (_is_in_partition) { |
655 | 0 | for (const auto& keys : t_part.in_keys) { |
656 | 0 | RETURN_IF_ERROR(_create_partition_keys( |
657 | 0 | keys, &part_result->in_keys.emplace_back(&_partition_block, -1))); |
658 | 0 | } |
659 | 0 | if (t_part.__isset.is_default_partition && t_part.is_default_partition && |
660 | 0 | _default_partition == nullptr) { |
661 | 0 | _default_partition = part_result; |
662 | 0 | } |
663 | 0 | } else { // range |
664 | 0 | if (t_part.__isset.start_keys) { |
665 | 0 | RETURN_IF_ERROR(_create_partition_keys(t_part.start_keys, &part_result->start_key)); |
666 | 0 | } |
667 | | // we generate the right bound but not insert into partition map |
668 | 0 | if (t_part.__isset.end_keys) { |
669 | 0 | RETURN_IF_ERROR(_create_partition_keys(t_part.end_keys, &part_result->end_key)); |
670 | 0 | } |
671 | 0 | } |
672 | | |
673 | 0 | part_result->num_buckets = t_part.num_buckets; |
674 | 0 | auto num_indexes = _schema->indexes().size(); |
675 | 0 | if (t_part.indexes.size() != num_indexes) { |
676 | 0 | return Status::InternalError( |
677 | 0 | "number of partition's index is not equal with schema's" |
678 | 0 | ", num_part_indexes={}, num_schema_indexes={}", |
679 | 0 | t_part.indexes.size(), num_indexes); |
680 | 0 | } |
681 | 0 | part_result->indexes = t_part.indexes; |
682 | 0 | std::sort(part_result->indexes.begin(), part_result->indexes.end(), |
683 | 0 | [](const OlapTableIndexTablets& lhs, const OlapTableIndexTablets& rhs) { |
684 | 0 | return lhs.index_id < rhs.index_id; |
685 | 0 | }); |
686 | | // check index |
687 | 0 | for (int j = 0; j < num_indexes; ++j) { |
688 | 0 | if (part_result->indexes[j].index_id != _schema->indexes()[j]->index_id) { |
689 | 0 | return Status::InternalError( |
690 | 0 | "partition's index is not equal with schema's" |
691 | 0 | ", part_index={}, schema_index={}", |
692 | 0 | part_result->indexes[j].index_id, _schema->indexes()[j]->index_id); |
693 | 0 | } |
694 | 0 | } |
695 | 0 | return Status::OK(); |
696 | 0 | } |
697 | | |
698 | | Status VOlapTablePartitionParam::add_partitions( |
699 | 0 | const std::vector<TOlapTablePartition>& partitions) { |
700 | 0 | for (const auto& t_part : partitions) { |
701 | 0 | auto* part = _obj_pool.add(new VOlapTablePartition(&_partition_block)); |
702 | 0 | part->id = t_part.id; |
703 | 0 | part->is_mutable = t_part.is_mutable; |
704 | | |
705 | | // we dont pass right keys when it's MAX_VALUE. so there's possibility we only have start_key but not end_key |
706 | | // range partition |
707 | 0 | if (t_part.__isset.start_keys) { |
708 | 0 | RETURN_IF_ERROR(_create_partition_keys(t_part.start_keys, &part->start_key)); |
709 | 0 | } |
710 | 0 | if (t_part.__isset.end_keys) { |
711 | 0 | RETURN_IF_ERROR(_create_partition_keys(t_part.end_keys, &part->end_key)); |
712 | 0 | } |
713 | | // list partition - we only set 1 value in 1 partition for new created ones |
714 | 0 | if (t_part.__isset.in_keys) { |
715 | 0 | for (const auto& keys : t_part.in_keys) { |
716 | 0 | RETURN_IF_ERROR(_create_partition_keys( |
717 | 0 | keys, &part->in_keys.emplace_back(&_partition_block, -1))); |
718 | 0 | } |
719 | 0 | if (t_part.__isset.is_default_partition && t_part.is_default_partition) { |
720 | 0 | _default_partition = part; |
721 | 0 | } |
722 | 0 | } |
723 | | |
724 | 0 | part->num_buckets = t_part.num_buckets; |
725 | 0 | auto num_indexes = _schema->indexes().size(); |
726 | 0 | if (t_part.indexes.size() != num_indexes) { |
727 | 0 | return Status::InternalError( |
728 | 0 | "number of partition's index is not equal with schema's" |
729 | 0 | ", num_part_indexes={}, num_schema_indexes={}", |
730 | 0 | t_part.indexes.size(), num_indexes); |
731 | 0 | } |
732 | 0 | part->indexes = t_part.indexes; |
733 | 0 | std::sort(part->indexes.begin(), part->indexes.end(), |
734 | 0 | [](const OlapTableIndexTablets& lhs, const OlapTableIndexTablets& rhs) { |
735 | 0 | return lhs.index_id < rhs.index_id; |
736 | 0 | }); |
737 | | // check index |
738 | 0 | for (int j = 0; j < num_indexes; ++j) { |
739 | 0 | if (part->indexes[j].index_id != _schema->indexes()[j]->index_id) { |
740 | 0 | return Status::InternalError( |
741 | 0 | "partition's index is not equal with schema's" |
742 | 0 | ", part_index={}, schema_index={}", |
743 | 0 | part->indexes[j].index_id, _schema->indexes()[j]->index_id); |
744 | 0 | } |
745 | 0 | } |
746 | 0 | _partitions.emplace_back(part); |
747 | | // after _creating_partiton_keys |
748 | 0 | if (_is_in_partition) { |
749 | 0 | for (auto& in_key : part->in_keys) { |
750 | 0 | _partitions_map->emplace(std::tuple {in_key.first, in_key.second, false}, part); |
751 | 0 | } |
752 | 0 | } else { |
753 | 0 | _partitions_map->emplace(std::tuple {part->end_key.first, part->end_key.second, false}, |
754 | 0 | part); |
755 | 0 | } |
756 | 0 | } |
757 | | |
758 | 0 | return Status::OK(); |
759 | 0 | } |
760 | | |
761 | | Status VOlapTablePartitionParam::replace_partitions( |
762 | | std::vector<int64_t>& old_partition_ids, |
763 | 0 | const std::vector<TOlapTablePartition>& new_partitions) { |
764 | | // remove old replaced partitions |
765 | 0 | DCHECK(old_partition_ids.size() == new_partitions.size()); |
766 | | |
767 | | // init and add new partitions. insert into _partitions |
768 | 0 | for (int i = 0; i < new_partitions.size(); i++) { |
769 | 0 | const auto& t_part = new_partitions[i]; |
770 | | // pair old_partition_ids and new_partitions one by one. TODO: sort to opt performance |
771 | 0 | VOlapTablePartition* old_part = nullptr; |
772 | 0 | auto old_part_id = old_partition_ids[i]; |
773 | 0 | if (auto it = std::find_if( |
774 | 0 | _partitions.begin(), _partitions.end(), |
775 | 0 | [=](const VOlapTablePartition* lhs) { return lhs->id == old_part_id; }); |
776 | 0 | it != _partitions.end()) { |
777 | 0 | old_part = *it; |
778 | 0 | } else { |
779 | 0 | return Status::InternalError("Cannot find old tablet {} in replacing", old_part_id); |
780 | 0 | } |
781 | | |
782 | 0 | auto* part = _obj_pool.add(new VOlapTablePartition(&_partition_block)); |
783 | 0 | part->id = t_part.id; |
784 | 0 | part->is_mutable = t_part.is_mutable; |
785 | | |
786 | | /// just substitute directly. no need to remove and reinsert keys. |
787 | | // range partition |
788 | 0 | part->start_key = std::move(old_part->start_key); |
789 | 0 | part->end_key = std::move(old_part->end_key); |
790 | | // list partition |
791 | 0 | part->in_keys = std::move(old_part->in_keys); |
792 | 0 | if (t_part.__isset.is_default_partition && t_part.is_default_partition) { |
793 | 0 | _default_partition = part; |
794 | 0 | } |
795 | |
|
796 | 0 | part->num_buckets = t_part.num_buckets; |
797 | 0 | auto num_indexes = _schema->indexes().size(); |
798 | 0 | if (t_part.indexes.size() != num_indexes) { |
799 | 0 | return Status::InternalError( |
800 | 0 | "number of partition's index is not equal with schema's" |
801 | 0 | ", num_part_indexes={}, num_schema_indexes={}", |
802 | 0 | t_part.indexes.size(), num_indexes); |
803 | 0 | } |
804 | 0 | part->indexes = t_part.indexes; |
805 | 0 | std::sort(part->indexes.begin(), part->indexes.end(), |
806 | 0 | [](const OlapTableIndexTablets& lhs, const OlapTableIndexTablets& rhs) { |
807 | 0 | return lhs.index_id < rhs.index_id; |
808 | 0 | }); |
809 | | // check index |
810 | 0 | for (int j = 0; j < num_indexes; ++j) { |
811 | 0 | if (part->indexes[j].index_id != _schema->indexes()[j]->index_id) { |
812 | 0 | return Status::InternalError( |
813 | 0 | "partition's index is not equal with schema's" |
814 | 0 | ", part_index={}, schema_index={}", |
815 | 0 | part->indexes[j].index_id, _schema->indexes()[j]->index_id); |
816 | 0 | } |
817 | 0 | } |
818 | | |
819 | | // add new partitions with new id. |
820 | 0 | _partitions.emplace_back(part); |
821 | 0 | VLOG_NOTICE << "params add new partition " << part->id; |
822 | | |
823 | | // replace items in _partition_maps |
824 | 0 | if (_is_in_partition) { |
825 | 0 | for (auto& in_key : part->in_keys) { |
826 | 0 | (*_partitions_map)[std::tuple {in_key.first, in_key.second, false}] = part; |
827 | 0 | } |
828 | 0 | } else { |
829 | 0 | (*_partitions_map)[std::tuple {part->end_key.first, part->end_key.second, false}] = |
830 | 0 | part; |
831 | 0 | } |
832 | 0 | } |
833 | | // remove old partitions by id |
834 | 0 | std::ranges::sort(old_partition_ids); |
835 | 0 | for (auto it = _partitions.begin(); it != _partitions.end();) { |
836 | 0 | if (std::ranges::binary_search(old_partition_ids, (*it)->id)) { |
837 | 0 | it = _partitions.erase(it); |
838 | 0 | } else { |
839 | 0 | it++; |
840 | 0 | } |
841 | 0 | } |
842 | |
|
843 | 0 | return Status::OK(); |
844 | 0 | } |
845 | | |
846 | | } // namespace doris |