Coverage Report

Created: 2025-07-30 19:44

next uncovered line (L), next uncovered region (R), next uncovered branch (B)
/root/doris/be/src/exec/tablet_info.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 "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
#include "common/compile_check_begin.h"
60
61
32
void OlapTableIndexSchema::to_protobuf(POlapTableIndexSchema* pindex) const {
62
32
    pindex->set_id(index_id);
63
32
    pindex->set_schema_hash(schema_hash);
64
32
    for (auto* slot : slots) {
65
0
        pindex->add_columns(slot->col_name());
66
0
    }
67
32
    for (auto* column : columns) {
68
0
        column->to_schema_pb(pindex->add_columns_desc());
69
0
    }
70
32
    for (auto* index : indexes) {
71
0
        index->to_schema_pb(pindex->add_indexes_desc());
72
0
    }
73
32
}
74
75
bool VOlapTablePartKeyComparator::operator()(const BlockRowWithIndicator& lhs,
76
0
                                             const BlockRowWithIndicator& rhs) const {
77
0
    vectorized::Block* l_block = std::get<0>(lhs);
78
0
    vectorized::Block* r_block = std::get<0>(rhs);
79
0
    int32_t l_row = std::get<1>(lhs);
80
0
    int32_t r_row = std::get<1>(rhs);
81
0
    bool l_use_new = std::get<2>(lhs);
82
0
    bool r_use_new = std::get<2>(rhs);
83
84
0
    VLOG_TRACE << '\n' << l_block->dump_data() << '\n' << r_block->dump_data();
85
86
0
    if (l_row == -1) {
87
0
        return false;
88
0
    } else if (r_row == -1) {
89
0
        return true;
90
0
    }
91
92
0
    if (_param_locs.empty()) { // no transform, use origin column
93
0
        for (auto slot_loc : _slot_locs) {
94
0
            auto res = l_block->get_by_position(slot_loc).column->compare_at(
95
0
                    l_row, r_row, *r_block->get_by_position(slot_loc).column, -1);
96
0
            if (res != 0) {
97
0
                return res < 0;
98
0
            }
99
0
        }
100
0
    } else { // use transformed column to compare
101
0
        DCHECK(_slot_locs.size() == _param_locs.size())
102
0
                << _slot_locs.size() << ' ' << _param_locs.size();
103
104
0
        const std::vector<uint16_t>* l_index = l_use_new ? &_param_locs : &_slot_locs;
105
0
        const std::vector<uint16_t>* r_index = r_use_new ? &_param_locs : &_slot_locs;
106
107
0
        for (int i = 0; i < _slot_locs.size(); i++) {
108
0
            vectorized::ColumnPtr l_col = l_block->get_by_position((*l_index)[i]).column;
109
0
            vectorized::ColumnPtr r_col = r_block->get_by_position((*r_index)[i]).column;
110
111
0
            auto res = l_col->compare_at(l_row, r_row, *r_col, -1);
112
0
            if (res != 0) {
113
0
                return res < 0;
114
0
            }
115
0
        }
116
0
    }
117
118
    // equal, return false
119
0
    return false;
120
0
}
121
122
14
Status OlapTableSchemaParam::init(const POlapTableSchemaParam& pschema) {
123
14
    _db_id = pschema.db_id();
124
14
    _table_id = pschema.table_id();
125
14
    _version = pschema.version();
126
14
    if (pschema.has_unique_key_update_mode()) {
127
14
        _unique_key_update_mode = pschema.unique_key_update_mode();
128
14
        if (pschema.has_sequence_map_col_unique_id()) {
129
14
            _sequence_map_col_uid = pschema.sequence_map_col_unique_id();
130
14
        }
131
14
    } else {
132
        // for backward compatibility
133
0
        if (pschema.has_partial_update() && pschema.partial_update()) {
134
0
            _unique_key_update_mode = UniqueKeyUpdateModePB::UPDATE_FIXED_COLUMNS;
135
0
        } else {
136
0
            _unique_key_update_mode = UniqueKeyUpdateModePB::UPSERT;
137
0
        }
138
0
    }
139
14
    _is_strict_mode = pschema.is_strict_mode();
140
14
    if (_unique_key_update_mode == UniqueKeyUpdateModePB::UPDATE_FIXED_COLUMNS) {
141
0
        _auto_increment_column = pschema.auto_increment_column();
142
0
        if (!_auto_increment_column.empty() && pschema.auto_increment_column_unique_id() == -1) {
143
0
            return Status::InternalError(
144
0
                    "Auto increment column id is not set in FE. Maybe FE is an older version "
145
0
                    "different from BE.");
146
0
        }
147
0
        _auto_increment_column_unique_id = pschema.auto_increment_column_unique_id();
148
0
    }
149
14
    if (_unique_key_update_mode != UniqueKeyUpdateModePB::UPSERT) {
150
0
        if (pschema.has_partial_update_new_key_policy()) {
151
0
            _partial_update_new_row_policy = pschema.partial_update_new_key_policy();
152
0
        }
153
0
    }
154
14
    _timestamp_ms = pschema.timestamp_ms();
155
14
    if (pschema.has_nano_seconds()) {
156
14
        _nano_seconds = pschema.nano_seconds();
157
14
    }
158
14
    _timezone = pschema.timezone();
159
160
14
    for (const auto& col : pschema.partial_update_input_columns()) {
161
0
        _partial_update_input_columns.insert(col);
162
0
    }
163
14
    std::unordered_map<std::string, SlotDescriptor*> slots_map;
164
165
14
    _tuple_desc = _obj_pool.add(new TupleDescriptor(pschema.tuple_desc()));
166
167
84
    for (const auto& p_slot_desc : pschema.slot_descs()) {
168
84
        auto* slot_desc = _obj_pool.add(new SlotDescriptor(p_slot_desc));
169
84
        _tuple_desc->add_slot(slot_desc);
170
84
        std::string data_type;
171
84
        EnumToString(TPrimitiveType, to_thrift(slot_desc->col_type()), data_type);
172
84
        std::string is_null_str = slot_desc->is_nullable() ? "true" : "false";
173
84
        std::string data_type_str =
174
84
                std::to_string(int64_t(TabletColumn::get_field_type_by_string(data_type)));
175
84
        slots_map.emplace(to_lower(slot_desc->col_name()) + "+" + data_type_str + is_null_str,
176
84
                          slot_desc);
177
84
    }
178
179
28
    for (const auto& p_index : pschema.indexes()) {
180
28
        auto* index = _obj_pool.add(new OlapTableIndexSchema());
181
28
        index->index_id = p_index.id();
182
28
        index->schema_hash = p_index.schema_hash();
183
28
        for (const auto& pcolumn_desc : p_index.columns_desc()) {
184
0
            if (_unique_key_update_mode != UniqueKeyUpdateModePB::UPDATE_FIXED_COLUMNS ||
185
0
                _partial_update_input_columns.contains(pcolumn_desc.name())) {
186
0
                std::string is_null_str = pcolumn_desc.is_nullable() ? "true" : "false";
187
0
                std::string data_type_str = std::to_string(
188
0
                        int64_t(TabletColumn::get_field_type_by_string(pcolumn_desc.type())));
189
0
                auto it = slots_map.find(to_lower(pcolumn_desc.name()) + "+" + data_type_str +
190
0
                                         is_null_str);
191
0
                if (it == std::end(slots_map)) {
192
0
                    std::string keys {};
193
0
                    for (const auto& [key, _] : slots_map) {
194
0
                        keys += fmt::format("{},", key);
195
0
                    }
196
0
                    LOG_EVERY_SECOND(WARNING) << fmt::format(
197
0
                            "[OlapTableSchemaParam::init(const POlapTableSchemaParam& pschema)]: "
198
0
                            "unknown index column, column={}, type={}, data_type_str={}, "
199
0
                            "is_null_str={}, slots_map.keys()=[{}], {}\npschema={}",
200
0
                            pcolumn_desc.name(), pcolumn_desc.type(), data_type_str, is_null_str,
201
0
                            keys, debug_string(), pschema.ShortDebugString());
202
203
0
                    return Status::InternalError("unknown index column, column={}, type={}",
204
0
                                                 pcolumn_desc.name(), pcolumn_desc.type());
205
0
                }
206
0
                index->slots.emplace_back(it->second);
207
0
            }
208
0
            TabletColumn* tc = _obj_pool.add(new TabletColumn());
209
0
            tc->init_from_pb(pcolumn_desc);
210
0
            index->columns.emplace_back(tc);
211
0
        }
212
28
        for (const auto& pindex_desc : p_index.indexes_desc()) {
213
0
            TabletIndex* ti = _obj_pool.add(new TabletIndex());
214
0
            ti->init_from_pb(pindex_desc);
215
0
            index->indexes.emplace_back(ti);
216
0
        }
217
28
        _indexes.emplace_back(index);
218
28
    }
219
220
14
    std::sort(_indexes.begin(), _indexes.end(),
221
28
              [](const OlapTableIndexSchema* lhs, const OlapTableIndexSchema* rhs) {
222
28
                  return lhs->index_id < rhs->index_id;
223
28
              });
224
14
    return Status::OK();
225
14
}
226
227
16
Status OlapTableSchemaParam::init_unique_key_update_mode(const TOlapTableSchemaParam& tschema) {
228
16
    if (tschema.__isset.unique_key_update_mode) {
229
0
        switch (tschema.unique_key_update_mode) {
230
0
        case doris::TUniqueKeyUpdateMode::UPSERT: {
231
0
            _unique_key_update_mode = UniqueKeyUpdateModePB::UPSERT;
232
0
            break;
233
0
        }
234
0
        case doris::TUniqueKeyUpdateMode::UPDATE_FIXED_COLUMNS: {
235
0
            _unique_key_update_mode = UniqueKeyUpdateModePB::UPDATE_FIXED_COLUMNS;
236
0
            break;
237
0
        }
238
0
        case doris::TUniqueKeyUpdateMode::UPDATE_FLEXIBLE_COLUMNS: {
239
0
            _unique_key_update_mode = UniqueKeyUpdateModePB::UPDATE_FLEXIBLE_COLUMNS;
240
0
            break;
241
0
        }
242
0
        default: {
243
0
            return Status::InternalError(
244
0
                    "Unknown unique_key_update_mode: {}, should be one of "
245
0
                    "UPSERT/UPDATE_FIXED_COLUMNS/UPDATE_FLEXIBLE_COLUMNS",
246
0
                    tschema.unique_key_update_mode);
247
0
        }
248
0
        }
249
0
        if (tschema.__isset.sequence_map_col_unique_id) {
250
0
            _sequence_map_col_uid = tschema.sequence_map_col_unique_id;
251
0
        }
252
16
    } else {
253
        // for backward compatibility
254
16
        if (tschema.__isset.is_partial_update && tschema.is_partial_update) {
255
0
            _unique_key_update_mode = UniqueKeyUpdateModePB::UPDATE_FIXED_COLUMNS;
256
16
        } else {
257
16
            _unique_key_update_mode = UniqueKeyUpdateModePB::UPSERT;
258
16
        }
259
16
    }
260
16
    return Status::OK();
261
16
}
262
263
16
Status OlapTableSchemaParam::init(const TOlapTableSchemaParam& tschema) {
264
16
    _db_id = tschema.db_id;
265
16
    _table_id = tschema.table_id;
266
16
    _version = tschema.version;
267
16
    RETURN_IF_ERROR(init_unique_key_update_mode(tschema));
268
16
    if (tschema.__isset.is_strict_mode) {
269
16
        _is_strict_mode = tschema.is_strict_mode;
270
16
    }
271
16
    if (_unique_key_update_mode == UniqueKeyUpdateModePB::UPDATE_FIXED_COLUMNS) {
272
0
        _auto_increment_column = tschema.auto_increment_column;
273
0
        if (!_auto_increment_column.empty() && tschema.auto_increment_column_unique_id == -1) {
274
0
            return Status::InternalError(
275
0
                    "Auto increment column id is not set in FE. Maybe FE is an older version "
276
0
                    "different from BE.");
277
0
        }
278
0
        _auto_increment_column_unique_id = tschema.auto_increment_column_unique_id;
279
0
    }
280
281
16
    if (_unique_key_update_mode != UniqueKeyUpdateModePB::UPSERT) {
282
0
        if (tschema.__isset.partial_update_new_key_policy) {
283
0
            switch (tschema.partial_update_new_key_policy) {
284
0
            case doris::TPartialUpdateNewRowPolicy::APPEND: {
285
0
                _partial_update_new_row_policy = PartialUpdateNewRowPolicyPB::APPEND;
286
0
                break;
287
0
            }
288
0
            case doris::TPartialUpdateNewRowPolicy::ERROR: {
289
0
                _partial_update_new_row_policy = PartialUpdateNewRowPolicyPB::ERROR;
290
0
                break;
291
0
            }
292
0
            default: {
293
0
                return Status::InvalidArgument(
294
0
                        "Unknown partial_update_new_key_behavior: {}, should be one of "
295
0
                        "'APPEND' or 'ERROR'",
296
0
                        tschema.partial_update_new_key_policy);
297
0
            }
298
0
            }
299
0
        }
300
0
    }
301
302
16
    for (const auto& tcolumn : tschema.partial_update_input_columns) {
303
0
        _partial_update_input_columns.insert(tcolumn);
304
0
    }
305
16
    std::unordered_map<std::string, SlotDescriptor*> slots_map;
306
16
    _tuple_desc = _obj_pool.add(new TupleDescriptor(tschema.tuple_desc));
307
96
    for (const auto& t_slot_desc : tschema.slot_descs) {
308
96
        auto* slot_desc = _obj_pool.add(new SlotDescriptor(t_slot_desc));
309
96
        _tuple_desc->add_slot(slot_desc);
310
96
        std::string is_null_str = slot_desc->is_nullable() ? "true" : "false";
311
96
        std::string data_type_str = std::to_string(int64_t(slot_desc->col_type()));
312
96
        slots_map.emplace(to_lower(slot_desc->col_name()) + "+" + data_type_str + is_null_str,
313
96
                          slot_desc);
314
96
    }
315
316
32
    for (const auto& t_index : tschema.indexes) {
317
32
        std::unordered_map<std::string, int32_t> index_slots_map;
318
32
        auto* index = _obj_pool.add(new OlapTableIndexSchema());
319
32
        index->index_id = t_index.id;
320
32
        index->schema_hash = t_index.schema_hash;
321
32
        for (const auto& tcolumn_desc : t_index.columns_desc) {
322
0
            if (_unique_key_update_mode != UniqueKeyUpdateModePB::UPDATE_FIXED_COLUMNS ||
323
0
                _partial_update_input_columns.contains(tcolumn_desc.column_name)) {
324
0
                std::string is_null_str = tcolumn_desc.is_allow_null ? "true" : "false";
325
0
                std::string data_type_str =
326
0
                        std::to_string(int64_t(thrift_to_type(tcolumn_desc.column_type.type)));
327
0
                auto it = slots_map.find(to_lower(tcolumn_desc.column_name) + "+" + data_type_str +
328
0
                                         is_null_str);
329
0
                if (it == slots_map.end()) {
330
0
                    std::stringstream ss;
331
0
                    ss << tschema;
332
0
                    std::string keys {};
333
0
                    for (const auto& [key, _] : slots_map) {
334
0
                        keys += fmt::format("{},", key);
335
0
                    }
336
0
                    LOG_EVERY_SECOND(WARNING) << fmt::format(
337
0
                            "[OlapTableSchemaParam::init(const TOlapTableSchemaParam& tschema)]: "
338
0
                            "unknown index column, column={}, type={}, data_type_str={}, "
339
0
                            "is_null_str={}, slots_map.keys()=[{}], {}\ntschema={}",
340
0
                            tcolumn_desc.column_name, tcolumn_desc.column_type.type, data_type_str,
341
0
                            is_null_str, keys, debug_string(), ss.str());
342
0
                    return Status::InternalError("unknown index column, column={}, type={}",
343
0
                                                 tcolumn_desc.column_name,
344
0
                                                 tcolumn_desc.column_type.type);
345
0
                }
346
0
                index->slots.emplace_back(it->second);
347
0
            }
348
0
            index_slots_map.emplace(to_lower(tcolumn_desc.column_name), tcolumn_desc.col_unique_id);
349
0
            TabletColumn* tc = _obj_pool.add(new TabletColumn());
350
0
            tc->init_from_thrift(tcolumn_desc);
351
0
            index->columns.emplace_back(tc);
352
0
        }
353
32
        if (t_index.__isset.indexes_desc) {
354
0
            for (const auto& tindex_desc : t_index.indexes_desc) {
355
0
                std::vector<int32_t> column_unique_ids(tindex_desc.columns.size());
356
0
                for (size_t i = 0; i < tindex_desc.columns.size(); i++) {
357
0
                    auto it = index_slots_map.find(to_lower(tindex_desc.columns[i]));
358
0
                    if (it != index_slots_map.end()) {
359
0
                        column_unique_ids[i] = it->second;
360
0
                    }
361
0
                }
362
0
                TabletIndex* ti = _obj_pool.add(new TabletIndex());
363
0
                ti->init_from_thrift(tindex_desc, column_unique_ids);
364
0
                index->indexes.emplace_back(ti);
365
0
            }
366
0
        }
367
32
        if (t_index.__isset.where_clause) {
368
0
            RETURN_IF_ERROR(
369
0
                    vectorized::VExpr::create_expr_tree(t_index.where_clause, index->where_clause));
370
0
        }
371
32
        _indexes.emplace_back(index);
372
32
    }
373
374
16
    std::sort(_indexes.begin(), _indexes.end(),
375
32
              [](const OlapTableIndexSchema* lhs, const OlapTableIndexSchema* rhs) {
376
32
                  return lhs->index_id < rhs->index_id;
377
32
              });
378
16
    return Status::OK();
379
16
}
380
381
16
void OlapTableSchemaParam::to_protobuf(POlapTableSchemaParam* pschema) const {
382
16
    pschema->set_db_id(_db_id);
383
16
    pschema->set_table_id(_table_id);
384
16
    pschema->set_version(_version);
385
16
    pschema->set_unique_key_update_mode(_unique_key_update_mode);
386
16
    if (_unique_key_update_mode == UniqueKeyUpdateModePB::UPDATE_FIXED_COLUMNS) {
387
        // for backward compatibility
388
0
        pschema->set_partial_update(true);
389
0
    }
390
16
    pschema->set_partial_update_new_key_policy(_partial_update_new_row_policy);
391
16
    pschema->set_is_strict_mode(_is_strict_mode);
392
16
    pschema->set_auto_increment_column(_auto_increment_column);
393
16
    pschema->set_auto_increment_column_unique_id(_auto_increment_column_unique_id);
394
16
    pschema->set_timestamp_ms(_timestamp_ms);
395
16
    pschema->set_timezone(_timezone);
396
16
    pschema->set_nano_seconds(_nano_seconds);
397
16
    pschema->set_sequence_map_col_unique_id(_sequence_map_col_uid);
398
16
    for (auto col : _partial_update_input_columns) {
399
0
        *pschema->add_partial_update_input_columns() = col;
400
0
    }
401
16
    _tuple_desc->to_protobuf(pschema->mutable_tuple_desc());
402
96
    for (auto* slot : _tuple_desc->slots()) {
403
96
        slot->to_protobuf(pschema->add_slot_descs());
404
96
    }
405
32
    for (auto* index : _indexes) {
406
32
        index->to_protobuf(pschema->add_indexes());
407
32
    }
408
16
}
409
410
0
std::string OlapTableSchemaParam::debug_string() const {
411
0
    std::stringstream ss;
412
0
    ss << "tuple_desc=" << _tuple_desc->debug_string();
413
0
    return ss.str();
414
0
}
415
416
VOlapTablePartitionParam::VOlapTablePartitionParam(std::shared_ptr<OlapTableSchemaParam>& schema,
417
                                                   const TOlapTablePartitionParam& t_param)
418
0
        : _schema(schema),
419
0
          _t_param(t_param),
420
0
          _slots(_schema->tuple_desc()->slots()),
421
0
          _mem_tracker(std::make_unique<MemTracker>("OlapTablePartitionParam")),
422
0
          _part_type(t_param.partition_type) {
423
0
    if (t_param.__isset.enable_automatic_partition && t_param.enable_automatic_partition) {
424
0
        _is_auto_partition = true;
425
0
        auto size = t_param.partition_function_exprs.size();
426
0
        _part_func_ctx.resize(size);
427
0
        _partition_function.resize(size);
428
0
        DCHECK((t_param.partition_type == TPartitionType::RANGE_PARTITIONED && size == 1) ||
429
0
               (t_param.partition_type == TPartitionType::LIST_PARTITIONED && size >= 1))
430
0
                << "now support only 1 partition column for auto range partitions. "
431
0
                << t_param.partition_type << " " << size;
432
0
        for (int i = 0; i < size; ++i) {
433
0
            Status st = vectorized::VExpr::create_expr_tree(t_param.partition_function_exprs[i],
434
0
                                                            _part_func_ctx[i]);
435
0
            if (!st.ok()) {
436
0
                throw Exception(Status::InternalError("Partition function expr is not valid"),
437
0
                                "Partition function expr is not valid");
438
0
            }
439
0
            _partition_function[i] = _part_func_ctx[i]->root();
440
0
        }
441
0
    }
442
443
0
    if (t_param.__isset.enable_auto_detect_overwrite && t_param.enable_auto_detect_overwrite) {
444
0
        _is_auto_detect_overwrite = true;
445
0
        DCHECK(t_param.__isset.overwrite_group_id);
446
0
        _overwrite_group_id = t_param.overwrite_group_id;
447
0
    }
448
449
0
    if (_is_auto_partition) {
450
        // the nullable mode depends on partition_exprs. not column slots. so use them.
451
0
        DCHECK(_partition_function.size() <= _slots.size())
452
0
                << _partition_function.size() << ", " << _slots.size();
453
454
        // suppose (k0, [k1], [k2]), so get [k1, 0], [k2, 1]
455
0
        std::map<std::string, int> partition_slots_map; // name to idx in part_exprs
456
0
        for (size_t i = 0; i < t_param.partition_columns.size(); i++) {
457
0
            partition_slots_map.emplace(t_param.partition_columns[i], i);
458
0
        }
459
460
        // here we rely on the same order and number of the _part_funcs and _slots in the prefix
461
        // _part_block contains all slots of table.
462
0
        for (auto* slot : _slots) {
463
            // try to replace with partition expr.
464
0
            if (auto it = partition_slots_map.find(slot->col_name());
465
0
                it != partition_slots_map.end()) { // it's a partition column slot
466
0
                auto& expr_type = _partition_function[it->second]->data_type();
467
0
                _partition_block.insert({expr_type->create_column(), expr_type, slot->col_name()});
468
0
            } else {
469
0
                _partition_block.insert({slot->get_empty_mutable_column(),
470
0
                                         slot->get_data_type_ptr(), slot->col_name()});
471
0
            }
472
0
        }
473
0
        VLOG_TRACE << _partition_block.dump_structure();
474
0
    } else {
475
        // we insert all. but not all will be used. it will controlled by _partition_slot_locs
476
0
        for (auto* slot : _slots) {
477
0
            _partition_block.insert({slot->get_empty_mutable_column(), slot->get_data_type_ptr(),
478
0
                                     slot->col_name()});
479
0
        }
480
0
    }
481
0
}
482
483
0
VOlapTablePartitionParam::~VOlapTablePartitionParam() {
484
0
    _mem_tracker->release(_mem_usage);
485
0
}
486
487
0
Status VOlapTablePartitionParam::init() {
488
0
    std::vector<std::string> slot_column_names;
489
0
    for (auto* slot_desc : _schema->tuple_desc()->slots()) {
490
0
        slot_column_names.emplace_back(slot_desc->col_name());
491
0
    }
492
493
0
    auto find_slot_locs = [&slot_column_names](const std::string& slot_name,
494
0
                                               std::vector<uint16_t>& locs,
495
0
                                               const std::string& column_type) {
496
0
        auto it = std::find(slot_column_names.begin(), slot_column_names.end(), slot_name);
497
0
        if (it == slot_column_names.end()) {
498
0
            return Status::InternalError("{} column not found, column ={}", column_type, slot_name);
499
0
        }
500
0
        locs.emplace_back(it - slot_column_names.begin());
501
0
        return Status::OK();
502
0
    };
503
504
    // here we find the partition columns. others maybe non-partition columns/special columns.
505
0
    if (_t_param.__isset.partition_columns) {
506
0
        for (auto& part_col : _t_param.partition_columns) {
507
0
            RETURN_IF_ERROR(find_slot_locs(part_col, _partition_slot_locs, "partition"));
508
0
        }
509
0
    }
510
511
0
    _partitions_map = std::make_unique<
512
0
            std::map<BlockRowWithIndicator, VOlapTablePartition*, VOlapTablePartKeyComparator>>(
513
0
            VOlapTablePartKeyComparator(_partition_slot_locs, _transformed_slot_locs));
514
0
    if (_t_param.__isset.distributed_columns) {
515
0
        for (auto& col : _t_param.distributed_columns) {
516
0
            RETURN_IF_ERROR(find_slot_locs(col, _distributed_slot_locs, "distributed"));
517
0
        }
518
0
    }
519
520
    // for both auto/non-auto partition table.
521
0
    _is_in_partition = _part_type == TPartitionType::type::LIST_PARTITIONED;
522
523
    // initial partitions. if meet dummy partitions only for open BE nodes, not generate key of them for finding
524
0
    for (const auto& t_part : _t_param.partitions) {
525
0
        VOlapTablePartition* part = nullptr;
526
0
        RETURN_IF_ERROR(generate_partition_from(t_part, part));
527
0
        _partitions.emplace_back(part);
528
529
0
        if (!_t_param.partitions_is_fake) {
530
0
            if (_is_in_partition) {
531
0
                for (auto& in_key : part->in_keys) {
532
0
                    _partitions_map->emplace(std::tuple {in_key.first, in_key.second, false}, part);
533
0
                }
534
0
            } else {
535
0
                _partitions_map->emplace(
536
0
                        std::tuple {part->end_key.first, part->end_key.second, false}, part);
537
0
            }
538
0
        }
539
0
    }
540
541
0
    _mem_usage = _partition_block.allocated_bytes();
542
0
    _mem_tracker->consume(_mem_usage);
543
0
    return Status::OK();
544
0
}
545
546
bool VOlapTablePartitionParam::_part_contains(VOlapTablePartition* part,
547
0
                                              BlockRowWithIndicator key) const {
548
0
    VOlapTablePartKeyComparator comparator(_partition_slot_locs, _transformed_slot_locs);
549
    // we have used upper_bound to find to ensure key < part.right and this part is closest(right - key is min)
550
    // now we only have to check (key >= part.left). the comparator(a,b) means a < b, so we use anti
551
0
    return part->start_key.second == -1 /* spj: start_key.second == -1 means only single partition*/
552
0
           || !comparator(key, std::tuple {part->start_key.first, part->start_key.second, false});
553
0
}
554
555
// insert value into _partition_block's column
556
// NOLINTBEGIN(readability-function-size)
557
0
static Status _create_partition_key(const TExprNode& t_expr, BlockRow* part_key, uint16_t pos) {
558
0
    auto column = std::move(*part_key->first->get_by_position(pos).column).mutate();
559
    //TODO: use assert_cast before insert_data
560
0
    switch (t_expr.node_type) {
561
0
    case TExprNodeType::DATE_LITERAL: {
562
0
        if (vectorized::DataTypeFactory::instance()
563
0
                    .create_data_type(t_expr.type)
564
0
                    ->get_primitive_type() == TYPE_DATEV2) {
565
0
            DateV2Value<DateV2ValueType> dt;
566
0
            if (!dt.from_date_str(t_expr.date_literal.value.c_str(),
567
0
                                  t_expr.date_literal.value.size())) {
568
0
                std::stringstream ss;
569
0
                ss << "invalid date literal in partition column, date=" << t_expr.date_literal;
570
0
                return Status::InternalError(ss.str());
571
0
            }
572
0
            column->insert_data(reinterpret_cast<const char*>(&dt), 0);
573
0
        } else if (vectorized::DataTypeFactory::instance()
574
0
                           .create_data_type(t_expr.type)
575
0
                           ->get_primitive_type() == TYPE_DATETIMEV2) {
576
0
            DateV2Value<DateTimeV2ValueType> dt;
577
0
            const int32_t scale =
578
0
                    t_expr.type.types.empty() ? -1 : t_expr.type.types.front().scalar_type.scale;
579
0
            if (!dt.from_date_str(t_expr.date_literal.value.c_str(),
580
0
                                  t_expr.date_literal.value.size(), scale)) {
581
0
                std::stringstream ss;
582
0
                ss << "invalid date literal in partition column, date=" << t_expr.date_literal;
583
0
                return Status::InternalError(ss.str());
584
0
            }
585
0
            column->insert_data(reinterpret_cast<const char*>(&dt), 0);
586
0
        } else {
587
0
            VecDateTimeValue dt;
588
0
            if (!dt.from_date_str(t_expr.date_literal.value.c_str(),
589
0
                                  t_expr.date_literal.value.size())) {
590
0
                std::stringstream ss;
591
0
                ss << "invalid date literal in partition column, date=" << t_expr.date_literal;
592
0
                return Status::InternalError(ss.str());
593
0
            }
594
0
            column->insert_data(reinterpret_cast<const char*>(&dt), 0);
595
0
        }
596
0
        break;
597
0
    }
598
0
    case TExprNodeType::INT_LITERAL: {
599
0
        switch (t_expr.type.types[0].scalar_type.type) {
600
0
        case TPrimitiveType::TINYINT: {
601
0
            auto value = cast_set<int8_t>(t_expr.int_literal.value);
602
0
            column->insert_data(reinterpret_cast<const char*>(&value), 0);
603
0
            break;
604
0
        }
605
0
        case TPrimitiveType::SMALLINT: {
606
0
            auto value = cast_set<int16_t>(t_expr.int_literal.value);
607
0
            column->insert_data(reinterpret_cast<const char*>(&value), 0);
608
0
            break;
609
0
        }
610
0
        case TPrimitiveType::INT: {
611
0
            auto value = cast_set<int32_t>(t_expr.int_literal.value);
612
0
            column->insert_data(reinterpret_cast<const char*>(&value), 0);
613
0
            break;
614
0
        }
615
0
        default:
616
0
            int64_t value = t_expr.int_literal.value;
617
0
            column->insert_data(reinterpret_cast<const char*>(&value), 0);
618
0
        }
619
0
        break;
620
0
    }
621
0
    case TExprNodeType::LARGE_INT_LITERAL: {
622
0
        StringParser::ParseResult parse_result = StringParser::PARSE_SUCCESS;
623
0
        auto value = StringParser::string_to_int<__int128>(t_expr.large_int_literal.value.c_str(),
624
0
                                                           t_expr.large_int_literal.value.size(),
625
0
                                                           &parse_result);
626
0
        if (parse_result != StringParser::PARSE_SUCCESS) {
627
0
            value = MAX_INT128;
628
0
        }
629
0
        column->insert_data(reinterpret_cast<const char*>(&value), 0);
630
0
        break;
631
0
    }
632
0
    case TExprNodeType::STRING_LITERAL: {
633
0
        size_t len = t_expr.string_literal.value.size();
634
0
        const char* str_val = t_expr.string_literal.value.c_str();
635
0
        column->insert_data(str_val, len);
636
0
        break;
637
0
    }
638
0
    case TExprNodeType::BOOL_LITERAL: {
639
0
        column->insert_data(reinterpret_cast<const char*>(&t_expr.bool_literal.value), 0);
640
0
        break;
641
0
    }
642
0
    case TExprNodeType::NULL_LITERAL: {
643
        // insert a null literal
644
0
        if (!column->is_nullable()) {
645
            // https://github.com/apache/doris/pull/39449 have forbid this cause. always add this check as protective measures
646
0
            return Status::InternalError("The column {} is not null, can't insert into NULL value.",
647
0
                                         part_key->first->get_by_position(pos).name);
648
0
        }
649
0
        column->insert_data(nullptr, 0);
650
0
        break;
651
0
    }
652
0
    default: {
653
0
        return Status::InternalError("unsupported partition column node type, type={}",
654
0
                                     t_expr.node_type);
655
0
    }
656
0
    }
657
0
    part_key->second = cast_set<int32_t>(column->size() - 1);
658
0
    return Status::OK();
659
0
}
660
// NOLINTEND(readability-function-size)
661
662
Status VOlapTablePartitionParam::_create_partition_keys(const std::vector<TExprNode>& t_exprs,
663
0
                                                        BlockRow* part_key) {
664
0
    for (int i = 0; i < t_exprs.size(); i++) {
665
0
        RETURN_IF_ERROR(_create_partition_key(t_exprs[i], part_key, _partition_slot_locs[i]));
666
0
    }
667
0
    return Status::OK();
668
0
}
669
670
Status VOlapTablePartitionParam::generate_partition_from(const TOlapTablePartition& t_part,
671
0
                                                         VOlapTablePartition*& part_result) {
672
0
    DCHECK(part_result == nullptr);
673
    // here we set the default value of partition bounds first! if it doesn't have some key, it will be -1.
674
0
    part_result = _obj_pool.add(new VOlapTablePartition(&_partition_block));
675
0
    part_result->id = t_part.id;
676
0
    part_result->is_mutable = t_part.is_mutable;
677
    // only load_to_single_tablet = true will set load_tablet_idx
678
0
    if (t_part.__isset.load_tablet_idx) {
679
0
        part_result->load_tablet_idx = t_part.load_tablet_idx;
680
0
    }
681
682
0
    if (_is_in_partition) {
683
0
        for (const auto& keys : t_part.in_keys) {
684
0
            RETURN_IF_ERROR(_create_partition_keys(
685
0
                    keys, &part_result->in_keys.emplace_back(&_partition_block, -1)));
686
0
        }
687
0
        if (t_part.__isset.is_default_partition && t_part.is_default_partition &&
688
0
            _default_partition == nullptr) {
689
0
            _default_partition = part_result;
690
0
        }
691
0
    } else { // range
692
0
        if (t_part.__isset.start_keys) {
693
0
            RETURN_IF_ERROR(_create_partition_keys(t_part.start_keys, &part_result->start_key));
694
0
        }
695
        // we generate the right bound but not insert into partition map
696
0
        if (t_part.__isset.end_keys) {
697
0
            RETURN_IF_ERROR(_create_partition_keys(t_part.end_keys, &part_result->end_key));
698
0
        }
699
0
    }
700
701
0
    part_result->num_buckets = t_part.num_buckets;
702
0
    auto num_indexes = _schema->indexes().size();
703
0
    if (t_part.indexes.size() != num_indexes) {
704
0
        return Status::InternalError(
705
0
                "number of partition's index is not equal with schema's"
706
0
                ", num_part_indexes={}, num_schema_indexes={}",
707
0
                t_part.indexes.size(), num_indexes);
708
0
    }
709
0
    part_result->indexes = t_part.indexes;
710
0
    std::sort(part_result->indexes.begin(), part_result->indexes.end(),
711
0
              [](const OlapTableIndexTablets& lhs, const OlapTableIndexTablets& rhs) {
712
0
                  return lhs.index_id < rhs.index_id;
713
0
              });
714
    // check index
715
0
    for (int j = 0; j < num_indexes; ++j) {
716
0
        if (part_result->indexes[j].index_id != _schema->indexes()[j]->index_id) {
717
0
            return Status::InternalError(
718
0
                    "partition's index is not equal with schema's"
719
0
                    ", part_index={}, schema_index={}",
720
0
                    part_result->indexes[j].index_id, _schema->indexes()[j]->index_id);
721
0
        }
722
0
    }
723
0
    if (t_part.__isset.total_replica_num) {
724
0
        part_result->total_replica_num = t_part.total_replica_num;
725
0
    }
726
0
    if (t_part.__isset.load_required_replica_num) {
727
0
        part_result->load_required_replica_num = t_part.load_required_replica_num;
728
0
    }
729
0
    return Status::OK();
730
0
}
731
732
Status VOlapTablePartitionParam::add_partitions(
733
0
        const std::vector<TOlapTablePartition>& partitions) {
734
0
    for (const auto& t_part : partitions) {
735
0
        auto* part = _obj_pool.add(new VOlapTablePartition(&_partition_block));
736
0
        part->id = t_part.id;
737
0
        part->is_mutable = t_part.is_mutable;
738
739
        // we dont pass right keys when it's MAX_VALUE. so there's possibility we only have start_key but not end_key
740
        // range partition
741
0
        if (t_part.__isset.start_keys) {
742
0
            RETURN_IF_ERROR(_create_partition_keys(t_part.start_keys, &part->start_key));
743
0
        }
744
0
        if (t_part.__isset.end_keys) {
745
0
            RETURN_IF_ERROR(_create_partition_keys(t_part.end_keys, &part->end_key));
746
0
        }
747
        // list partition - we only set 1 value in 1 partition for new created ones
748
0
        if (t_part.__isset.in_keys) {
749
0
            for (const auto& keys : t_part.in_keys) {
750
0
                RETURN_IF_ERROR(_create_partition_keys(
751
0
                        keys, &part->in_keys.emplace_back(&_partition_block, -1)));
752
0
            }
753
0
            if (t_part.__isset.is_default_partition && t_part.is_default_partition) {
754
0
                _default_partition = part;
755
0
            }
756
0
        }
757
758
0
        part->num_buckets = t_part.num_buckets;
759
0
        auto num_indexes = _schema->indexes().size();
760
0
        if (t_part.indexes.size() != num_indexes) {
761
0
            return Status::InternalError(
762
0
                    "number of partition's index is not equal with schema's"
763
0
                    ", num_part_indexes={}, num_schema_indexes={}",
764
0
                    t_part.indexes.size(), num_indexes);
765
0
        }
766
0
        part->indexes = t_part.indexes;
767
0
        std::sort(part->indexes.begin(), part->indexes.end(),
768
0
                  [](const OlapTableIndexTablets& lhs, const OlapTableIndexTablets& rhs) {
769
0
                      return lhs.index_id < rhs.index_id;
770
0
                  });
771
        // check index
772
0
        for (int j = 0; j < num_indexes; ++j) {
773
0
            if (part->indexes[j].index_id != _schema->indexes()[j]->index_id) {
774
0
                return Status::InternalError(
775
0
                        "partition's index is not equal with schema's"
776
0
                        ", part_index={}, schema_index={}",
777
0
                        part->indexes[j].index_id, _schema->indexes()[j]->index_id);
778
0
            }
779
0
        }
780
0
        _partitions.emplace_back(part);
781
        // after _creating_partiton_keys
782
0
        if (_is_in_partition) {
783
0
            for (auto& in_key : part->in_keys) {
784
0
                _partitions_map->emplace(std::tuple {in_key.first, in_key.second, false}, part);
785
0
            }
786
0
        } else {
787
0
            _partitions_map->emplace(std::tuple {part->end_key.first, part->end_key.second, false},
788
0
                                     part);
789
0
        }
790
0
    }
791
792
0
    return Status::OK();
793
0
}
794
795
Status VOlapTablePartitionParam::replace_partitions(
796
        std::vector<int64_t>& old_partition_ids,
797
0
        const std::vector<TOlapTablePartition>& new_partitions) {
798
    // remove old replaced partitions
799
0
    DCHECK(old_partition_ids.size() == new_partitions.size());
800
801
    // init and add new partitions. insert into _partitions
802
0
    for (int i = 0; i < new_partitions.size(); i++) {
803
0
        const auto& t_part = new_partitions[i];
804
        // pair old_partition_ids and new_partitions one by one. TODO: sort to opt performance
805
0
        VOlapTablePartition* old_part = nullptr;
806
0
        auto old_part_id = old_partition_ids[i];
807
0
        if (auto it = std::find_if(
808
0
                    _partitions.begin(), _partitions.end(),
809
0
                    [=](const VOlapTablePartition* lhs) { return lhs->id == old_part_id; });
810
0
            it != _partitions.end()) {
811
0
            old_part = *it;
812
0
        } else {
813
0
            return Status::InternalError("Cannot find old tablet {} in replacing", old_part_id);
814
0
        }
815
816
0
        auto* part = _obj_pool.add(new VOlapTablePartition(&_partition_block));
817
0
        part->id = t_part.id;
818
0
        part->is_mutable = t_part.is_mutable;
819
820
        /// just substitute directly. no need to remove and reinsert keys.
821
        // range partition
822
0
        part->start_key = std::move(old_part->start_key);
823
0
        part->end_key = std::move(old_part->end_key);
824
        // list partition
825
0
        part->in_keys = std::move(old_part->in_keys);
826
0
        if (t_part.__isset.is_default_partition && t_part.is_default_partition) {
827
0
            _default_partition = part;
828
0
        }
829
830
0
        part->num_buckets = t_part.num_buckets;
831
0
        auto num_indexes = _schema->indexes().size();
832
0
        if (t_part.indexes.size() != num_indexes) {
833
0
            return Status::InternalError(
834
0
                    "number of partition's index is not equal with schema's"
835
0
                    ", num_part_indexes={}, num_schema_indexes={}",
836
0
                    t_part.indexes.size(), num_indexes);
837
0
        }
838
0
        part->indexes = t_part.indexes;
839
0
        std::sort(part->indexes.begin(), part->indexes.end(),
840
0
                  [](const OlapTableIndexTablets& lhs, const OlapTableIndexTablets& rhs) {
841
0
                      return lhs.index_id < rhs.index_id;
842
0
                  });
843
        // check index
844
0
        for (int j = 0; j < num_indexes; ++j) {
845
0
            if (part->indexes[j].index_id != _schema->indexes()[j]->index_id) {
846
0
                return Status::InternalError(
847
0
                        "partition's index is not equal with schema's"
848
0
                        ", part_index={}, schema_index={}",
849
0
                        part->indexes[j].index_id, _schema->indexes()[j]->index_id);
850
0
            }
851
0
        }
852
853
        // add new partitions with new id.
854
0
        _partitions.emplace_back(part);
855
0
        VLOG_NOTICE << "params add new partition " << part->id;
856
857
        // replace items in _partition_maps
858
0
        if (_is_in_partition) {
859
0
            for (auto& in_key : part->in_keys) {
860
0
                (*_partitions_map)[std::tuple {in_key.first, in_key.second, false}] = part;
861
0
            }
862
0
        } else {
863
0
            (*_partitions_map)[std::tuple {part->end_key.first, part->end_key.second, false}] =
864
0
                    part;
865
0
        }
866
0
    }
867
    // remove old partitions by id
868
0
    std::ranges::sort(old_partition_ids);
869
0
    for (auto it = _partitions.begin(); it != _partitions.end();) {
870
0
        if (std::ranges::binary_search(old_partition_ids, (*it)->id)) {
871
0
            it = _partitions.erase(it);
872
0
        } else {
873
0
            it++;
874
0
        }
875
0
    }
876
877
0
    return Status::OK();
878
0
}
879
#include "common/compile_check_end.h"
880
881
} // namespace doris