Coverage Report

Created: 2024-11-22 11:49

/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
// NOLINTNEXTLINE(unused-includes)
52
#include "vec/exprs/vexpr_context.h" // IWYU pragma: keep
53
#include "vec/exprs/vliteral.h"
54
#include "vec/runtime/vdatetime_value.h"
55
56
namespace doris {
57
58
32
void OlapTableIndexSchema::to_protobuf(POlapTableIndexSchema* pindex) const {
59
32
    pindex->set_id(index_id);
60
32
    pindex->set_schema_hash(schema_hash);
61
32
    for (auto* slot : slots) {
62
0
        pindex->add_columns(slot->col_name());
63
0
    }
64
32
    for (auto* column : columns) {
65
0
        column->to_schema_pb(pindex->add_columns_desc());
66
0
    }
67
32
    for (auto* index : indexes) {
68
0
        index->to_schema_pb(pindex->add_indexes_desc());
69
0
    }
70
32
}
71
72
bool VOlapTablePartKeyComparator::operator()(const BlockRowWithIndicator& lhs,
73
0
                                             const BlockRowWithIndicator& rhs) const {
74
0
    vectorized::Block* l_block = std::get<0>(lhs);
75
0
    vectorized::Block* r_block = std::get<0>(rhs);
76
0
    int32_t l_row = std::get<1>(lhs);
77
0
    int32_t r_row = std::get<1>(rhs);
78
0
    bool l_use_new = std::get<2>(lhs);
79
0
    bool r_use_new = std::get<2>(rhs);
80
81
0
    VLOG_TRACE << '\n' << l_block->dump_data() << '\n' << r_block->dump_data();
82
83
0
    if (l_row == -1) {
84
0
        return false;
85
0
    } else if (r_row == -1) {
86
0
        return true;
87
0
    }
88
89
0
    if (_param_locs.empty()) { // no transform, use origin column
90
0
        for (auto slot_loc : _slot_locs) {
91
0
            auto res = l_block->get_by_position(slot_loc).column->compare_at(
92
0
                    l_row, r_row, *r_block->get_by_position(slot_loc).column, -1);
93
0
            if (res != 0) {
94
0
                return res < 0;
95
0
            }
96
0
        }
97
0
    } else { // use transformed column to compare
98
0
        DCHECK(_slot_locs.size() == _param_locs.size())
99
0
                << _slot_locs.size() << ' ' << _param_locs.size();
100
101
0
        const std::vector<uint16_t>* l_index = l_use_new ? &_param_locs : &_slot_locs;
102
0
        const std::vector<uint16_t>* r_index = r_use_new ? &_param_locs : &_slot_locs;
103
104
0
        for (int i = 0; i < _slot_locs.size(); i++) {
105
0
            vectorized::ColumnPtr l_col = l_block->get_by_position((*l_index)[i]).column;
106
0
            vectorized::ColumnPtr r_col = r_block->get_by_position((*r_index)[i]).column;
107
108
0
            auto res = l_col->compare_at(l_row, r_row, *r_col, -1);
109
0
            if (res != 0) {
110
0
                return res < 0;
111
0
            }
112
0
        }
113
0
    }
114
115
    // equal, return false
116
0
    return false;
117
0
}
118
119
14
Status OlapTableSchemaParam::init(const POlapTableSchemaParam& pschema) {
120
14
    _db_id = pschema.db_id();
121
14
    _table_id = pschema.table_id();
122
14
    _version = pschema.version();
123
14
    if (pschema.has_unique_key_update_mode()) {
124
14
        _unique_key_update_mode = pschema.unique_key_update_mode();
125
14
        if (pschema.has_sequence_map_col_unique_id()) {
126
14
            _sequence_map_col_uid = pschema.sequence_map_col_unique_id();
127
14
        }
128
14
    } else {
129
        // for backward compatibility
130
0
        if (pschema.has_partial_update() && pschema.partial_update()) {
131
0
            _unique_key_update_mode = UniqueKeyUpdateModePB::UPDATE_FIXED_COLUMNS;
132
0
        } else {
133
0
            _unique_key_update_mode = UniqueKeyUpdateModePB::UPSERT;
134
0
        }
135
0
    }
136
14
    _is_strict_mode = pschema.is_strict_mode();
137
14
    if (_unique_key_update_mode == UniqueKeyUpdateModePB::UPDATE_FIXED_COLUMNS) {
138
0
        _auto_increment_column = pschema.auto_increment_column();
139
0
        if (!_auto_increment_column.empty() && pschema.auto_increment_column_unique_id() == -1) {
140
0
            return Status::InternalError(
141
0
                    "Auto increment column id is not set in FE. Maybe FE is an older version "
142
0
                    "different from BE.");
143
0
        }
144
0
        _auto_increment_column_unique_id = pschema.auto_increment_column_unique_id();
145
0
    }
146
14
    _timestamp_ms = pschema.timestamp_ms();
147
14
    if (pschema.has_nano_seconds()) {
148
14
        _nano_seconds = pschema.nano_seconds();
149
14
    }
150
14
    _timezone = pschema.timezone();
151
152
14
    for (const auto& col : pschema.partial_update_input_columns()) {
153
0
        _partial_update_input_columns.insert(col);
154
0
    }
155
14
    std::unordered_map<std::string, SlotDescriptor*> slots_map;
156
157
14
    _tuple_desc = _obj_pool.add(new TupleDescriptor(pschema.tuple_desc()));
158
159
84
    for (const auto& p_slot_desc : pschema.slot_descs()) {
160
84
        auto* slot_desc = _obj_pool.add(new SlotDescriptor(p_slot_desc));
161
84
        _tuple_desc->add_slot(slot_desc);
162
84
        string data_type;
163
84
        EnumToString(TPrimitiveType, to_thrift(slot_desc->col_type()), data_type);
164
84
        std::string is_null_str = slot_desc->is_nullable() ? "true" : "false";
165
84
        std::string data_type_str =
166
84
                std::to_string(int64_t(TabletColumn::get_field_type_by_string(data_type)));
167
84
        slots_map.emplace(to_lower(slot_desc->col_name()) + "+" + data_type_str + is_null_str,
168
84
                          slot_desc);
169
84
    }
170
171
28
    for (const auto& p_index : pschema.indexes()) {
172
28
        auto* index = _obj_pool.add(new OlapTableIndexSchema());
173
28
        index->index_id = p_index.id();
174
28
        index->schema_hash = p_index.schema_hash();
175
28
        for (const auto& pcolumn_desc : p_index.columns_desc()) {
176
0
            if (_unique_key_update_mode != UniqueKeyUpdateModePB::UPDATE_FIXED_COLUMNS ||
177
0
                _partial_update_input_columns.contains(pcolumn_desc.name())) {
178
0
                std::string is_null_str = pcolumn_desc.is_nullable() ? "true" : "false";
179
0
                std::string data_type_str = std::to_string(
180
0
                        int64_t(TabletColumn::get_field_type_by_string(pcolumn_desc.type())));
181
0
                auto it = slots_map.find(to_lower(pcolumn_desc.name()) + "+" + data_type_str +
182
0
                                         is_null_str);
183
0
                if (it == std::end(slots_map)) {
184
0
                    std::string keys {};
185
0
                    for (const auto& [key, _] : slots_map) {
186
0
                        keys += fmt::format("{},", key);
187
0
                    }
188
0
                    LOG_EVERY_SECOND(WARNING) << fmt::format(
189
0
                            "[OlapTableSchemaParam::init(const POlapTableSchemaParam& pschema)]: "
190
0
                            "unknown index column, column={}, type={}, data_type_str={}, "
191
0
                            "is_null_str={}, slots_map.keys()=[{}], {}\npschema={}",
192
0
                            pcolumn_desc.name(), pcolumn_desc.type(), data_type_str, is_null_str,
193
0
                            keys, debug_string(), pschema.ShortDebugString());
194
195
0
                    return Status::InternalError("unknown index column, column={}, type={}",
196
0
                                                 pcolumn_desc.name(), pcolumn_desc.type());
197
0
                }
198
0
                index->slots.emplace_back(it->second);
199
0
            }
200
0
            TabletColumn* tc = _obj_pool.add(new TabletColumn());
201
0
            tc->init_from_pb(pcolumn_desc);
202
0
            index->columns.emplace_back(tc);
203
0
        }
204
28
        for (const auto& pindex_desc : p_index.indexes_desc()) {
205
0
            TabletIndex* ti = _obj_pool.add(new TabletIndex());
206
0
            ti->init_from_pb(pindex_desc);
207
0
            index->indexes.emplace_back(ti);
208
0
        }
209
28
        _indexes.emplace_back(index);
210
28
    }
211
212
14
    std::sort(_indexes.begin(), _indexes.end(),
213
28
              [](const OlapTableIndexSchema* lhs, const OlapTableIndexSchema* rhs) {
214
28
                  return lhs->index_id < rhs->index_id;
215
28
              });
216
14
    return Status::OK();
217
14
}
218
219
16
Status OlapTableSchemaParam::init_unique_key_update_mode(const TOlapTableSchemaParam& tschema) {
220
16
    if (tschema.__isset.unique_key_update_mode) {
221
0
        switch (tschema.unique_key_update_mode) {
222
0
        case doris::TUniqueKeyUpdateMode::UPSERT: {
223
0
            _unique_key_update_mode = UniqueKeyUpdateModePB::UPSERT;
224
0
            break;
225
0
        }
226
0
        case doris::TUniqueKeyUpdateMode::UPDATE_FIXED_COLUMNS: {
227
0
            _unique_key_update_mode = UniqueKeyUpdateModePB::UPDATE_FIXED_COLUMNS;
228
0
            break;
229
0
        }
230
0
        case doris::TUniqueKeyUpdateMode::UPDATE_FLEXIBLE_COLUMNS: {
231
0
            _unique_key_update_mode = UniqueKeyUpdateModePB::UPDATE_FLEXIBLE_COLUMNS;
232
0
            break;
233
0
        }
234
0
        default: {
235
0
            return Status::InternalError(
236
0
                    "Unknown unique_key_update_mode: {}, should be one of "
237
0
                    "UPSERT/UPDATE_FIXED_COLUMNS/UPDATE_FLEXIBLE_COLUMNS",
238
0
                    tschema.unique_key_update_mode);
239
0
        }
240
0
        }
241
0
        if (tschema.__isset.sequence_map_col_unique_id) {
242
0
            _sequence_map_col_uid = tschema.sequence_map_col_unique_id;
243
0
        }
244
16
    } else {
245
        // for backward compatibility
246
16
        if (tschema.__isset.is_partial_update && tschema.is_partial_update) {
247
0
            _unique_key_update_mode = UniqueKeyUpdateModePB::UPDATE_FIXED_COLUMNS;
248
16
        } else {
249
16
            _unique_key_update_mode = UniqueKeyUpdateModePB::UPSERT;
250
16
        }
251
16
    }
252
16
    return Status::OK();
253
16
}
254
255
16
Status OlapTableSchemaParam::init(const TOlapTableSchemaParam& tschema) {
256
16
    _db_id = tschema.db_id;
257
16
    _table_id = tschema.table_id;
258
16
    _version = tschema.version;
259
16
    RETURN_IF_ERROR(init_unique_key_update_mode(tschema));
260
16
    if (tschema.__isset.is_strict_mode) {
261
16
        _is_strict_mode = tschema.is_strict_mode;
262
16
    }
263
16
    if (_unique_key_update_mode == UniqueKeyUpdateModePB::UPDATE_FIXED_COLUMNS) {
264
0
        _auto_increment_column = tschema.auto_increment_column;
265
0
        if (!_auto_increment_column.empty() && tschema.auto_increment_column_unique_id == -1) {
266
0
            return Status::InternalError(
267
0
                    "Auto increment column id is not set in FE. Maybe FE is an older version "
268
0
                    "different from BE.");
269
0
        }
270
0
        _auto_increment_column_unique_id = tschema.auto_increment_column_unique_id;
271
0
    }
272
273
16
    for (const auto& tcolumn : tschema.partial_update_input_columns) {
274
0
        _partial_update_input_columns.insert(tcolumn);
275
0
    }
276
16
    std::unordered_map<std::string, SlotDescriptor*> slots_map;
277
16
    _tuple_desc = _obj_pool.add(new TupleDescriptor(tschema.tuple_desc));
278
96
    for (const auto& t_slot_desc : tschema.slot_descs) {
279
96
        auto* slot_desc = _obj_pool.add(new SlotDescriptor(t_slot_desc));
280
96
        _tuple_desc->add_slot(slot_desc);
281
96
        std::string is_null_str = slot_desc->is_nullable() ? "true" : "false";
282
96
        std::string data_type_str = std::to_string(int64_t(slot_desc->col_type()));
283
96
        slots_map.emplace(to_lower(slot_desc->col_name()) + "+" + data_type_str + is_null_str,
284
96
                          slot_desc);
285
96
    }
286
287
32
    for (const auto& t_index : tschema.indexes) {
288
32
        std::unordered_map<std::string, int32_t> index_slots_map;
289
32
        auto* index = _obj_pool.add(new OlapTableIndexSchema());
290
32
        index->index_id = t_index.id;
291
32
        index->schema_hash = t_index.schema_hash;
292
32
        for (const auto& tcolumn_desc : t_index.columns_desc) {
293
0
            if (_unique_key_update_mode != UniqueKeyUpdateModePB::UPDATE_FIXED_COLUMNS ||
294
0
                _partial_update_input_columns.contains(tcolumn_desc.column_name)) {
295
0
                std::string is_null_str = tcolumn_desc.is_allow_null ? "true" : "false";
296
0
                std::string data_type_str =
297
0
                        std::to_string(int64_t(thrift_to_type(tcolumn_desc.column_type.type)));
298
0
                auto it = slots_map.find(to_lower(tcolumn_desc.column_name) + "+" + data_type_str +
299
0
                                         is_null_str);
300
0
                if (it == slots_map.end()) {
301
0
                    std::stringstream ss;
302
0
                    ss << tschema;
303
0
                    std::string keys {};
304
0
                    for (const auto& [key, _] : slots_map) {
305
0
                        keys += fmt::format("{},", key);
306
0
                    }
307
0
                    LOG_EVERY_SECOND(WARNING) << fmt::format(
308
0
                            "[OlapTableSchemaParam::init(const TOlapTableSchemaParam& tschema)]: "
309
0
                            "unknown index column, column={}, type={}, data_type_str={}, "
310
0
                            "is_null_str={}, slots_map.keys()=[{}], {}\ntschema={}",
311
0
                            tcolumn_desc.column_name, tcolumn_desc.column_type.type, data_type_str,
312
0
                            is_null_str, keys, debug_string(), ss.str());
313
0
                    return Status::InternalError("unknown index column, column={}, type={}",
314
0
                                                 tcolumn_desc.column_name,
315
0
                                                 tcolumn_desc.column_type.type);
316
0
                }
317
0
                index->slots.emplace_back(it->second);
318
0
            }
319
0
            index_slots_map.emplace(to_lower(tcolumn_desc.column_name), tcolumn_desc.col_unique_id);
320
0
            TabletColumn* tc = _obj_pool.add(new TabletColumn());
321
0
            tc->init_from_thrift(tcolumn_desc);
322
0
            index->columns.emplace_back(tc);
323
0
        }
324
32
        if (t_index.__isset.indexes_desc) {
325
0
            for (const auto& tindex_desc : t_index.indexes_desc) {
326
0
                std::vector<int32_t> column_unique_ids(tindex_desc.columns.size());
327
0
                for (size_t i = 0; i < tindex_desc.columns.size(); i++) {
328
0
                    auto it = index_slots_map.find(to_lower(tindex_desc.columns[i]));
329
0
                    if (it != index_slots_map.end()) {
330
0
                        column_unique_ids[i] = it->second;
331
0
                    }
332
0
                }
333
0
                TabletIndex* ti = _obj_pool.add(new TabletIndex());
334
0
                ti->init_from_thrift(tindex_desc, column_unique_ids);
335
0
                index->indexes.emplace_back(ti);
336
0
            }
337
0
        }
338
32
        if (t_index.__isset.where_clause) {
339
0
            RETURN_IF_ERROR(
340
0
                    vectorized::VExpr::create_expr_tree(t_index.where_clause, index->where_clause));
341
0
        }
342
32
        _indexes.emplace_back(index);
343
32
    }
344
345
16
    std::sort(_indexes.begin(), _indexes.end(),
346
32
              [](const OlapTableIndexSchema* lhs, const OlapTableIndexSchema* rhs) {
347
32
                  return lhs->index_id < rhs->index_id;
348
32
              });
349
16
    return Status::OK();
350
16
}
351
352
16
void OlapTableSchemaParam::to_protobuf(POlapTableSchemaParam* pschema) const {
353
16
    pschema->set_db_id(_db_id);
354
16
    pschema->set_table_id(_table_id);
355
16
    pschema->set_version(_version);
356
16
    pschema->set_unique_key_update_mode(_unique_key_update_mode);
357
16
    if (_unique_key_update_mode == UniqueKeyUpdateModePB::UPDATE_FIXED_COLUMNS) {
358
        // for backward compatibility
359
0
        pschema->set_partial_update(true);
360
0
    }
361
16
    pschema->set_is_strict_mode(_is_strict_mode);
362
16
    pschema->set_auto_increment_column(_auto_increment_column);
363
16
    pschema->set_auto_increment_column_unique_id(_auto_increment_column_unique_id);
364
16
    pschema->set_timestamp_ms(_timestamp_ms);
365
16
    pschema->set_timezone(_timezone);
366
16
    pschema->set_nano_seconds(_nano_seconds);
367
16
    pschema->set_sequence_map_col_unique_id(_sequence_map_col_uid);
368
16
    for (auto col : _partial_update_input_columns) {
369
0
        *pschema->add_partial_update_input_columns() = col;
370
0
    }
371
16
    _tuple_desc->to_protobuf(pschema->mutable_tuple_desc());
372
96
    for (auto* slot : _tuple_desc->slots()) {
373
96
        slot->to_protobuf(pschema->add_slot_descs());
374
96
    }
375
32
    for (auto* index : _indexes) {
376
32
        index->to_protobuf(pschema->add_indexes());
377
32
    }
378
16
}
379
380
0
std::string OlapTableSchemaParam::debug_string() const {
381
0
    std::stringstream ss;
382
0
    ss << "tuple_desc=" << _tuple_desc->debug_string();
383
0
    return ss.str();
384
0
}
385
386
VOlapTablePartitionParam::VOlapTablePartitionParam(std::shared_ptr<OlapTableSchemaParam>& schema,
387
                                                   const TOlapTablePartitionParam& t_param)
388
        : _schema(schema),
389
          _t_param(t_param),
390
          _slots(_schema->tuple_desc()->slots()),
391
          _mem_tracker(std::make_unique<MemTracker>("OlapTablePartitionParam")),
392
0
          _part_type(t_param.partition_type) {
393
0
    if (t_param.__isset.enable_automatic_partition && t_param.enable_automatic_partition) {
394
0
        _is_auto_partition = true;
395
0
        auto size = t_param.partition_function_exprs.size();
396
0
        _part_func_ctx.resize(size);
397
0
        _partition_function.resize(size);
398
0
        DCHECK((t_param.partition_type == TPartitionType::RANGE_PARTITIONED && size == 1) ||
399
0
               (t_param.partition_type == TPartitionType::LIST_PARTITIONED && size >= 1))
400
0
                << "now support only 1 partition column for auto range partitions. "
401
0
                << t_param.partition_type << " " << size;
402
0
        for (int i = 0; i < size; ++i) {
403
0
            Status st = vectorized::VExpr::create_expr_tree(t_param.partition_function_exprs[i],
404
0
                                                            _part_func_ctx[i]);
405
0
            if (!st.ok()) {
406
0
                throw Exception(Status::InternalError("Partition function expr is not valid"),
407
0
                                "Partition function expr is not valid");
408
0
            }
409
0
            _partition_function[i] = _part_func_ctx[i]->root();
410
0
        }
411
0
    }
412
413
0
    if (t_param.__isset.enable_auto_detect_overwrite && t_param.enable_auto_detect_overwrite) {
414
0
        _is_auto_detect_overwrite = true;
415
0
        DCHECK(t_param.__isset.overwrite_group_id);
416
0
        _overwrite_group_id = t_param.overwrite_group_id;
417
0
    }
418
419
0
    if (_is_auto_partition) {
420
        // the nullable mode depends on partition_exprs. not column slots. so use them.
421
0
        DCHECK(_partition_function.size() <= _slots.size())
422
0
                << _partition_function.size() << ", " << _slots.size();
423
424
        // suppose (k0, [k1], [k2]), so get [k1, 0], [k2, 1]
425
0
        std::map<std::string, int> partition_slots_map; // name to idx in part_exprs
426
0
        for (size_t i = 0; i < t_param.partition_columns.size(); i++) {
427
0
            partition_slots_map.emplace(t_param.partition_columns[i], i);
428
0
        }
429
430
        // here we rely on the same order and number of the _part_funcs and _slots in the prefix
431
        // _part_block contains all slots of table.
432
0
        for (auto* slot : _slots) {
433
            // try to replace with partition expr.
434
0
            if (auto it = partition_slots_map.find(slot->col_name());
435
0
                it != partition_slots_map.end()) { // it's a partition column slot
436
0
                auto& expr_type = _partition_function[it->second]->data_type();
437
0
                _partition_block.insert({expr_type->create_column(), expr_type, slot->col_name()});
438
0
            } else {
439
0
                _partition_block.insert({slot->get_empty_mutable_column(),
440
0
                                         slot->get_data_type_ptr(), slot->col_name()});
441
0
            }
442
0
        }
443
0
        VLOG_TRACE << _partition_block.dump_structure();
444
0
    } else {
445
        // we insert all. but not all will be used. it will controlled by _partition_slot_locs
446
0
        for (auto* slot : _slots) {
447
0
            _partition_block.insert({slot->get_empty_mutable_column(), slot->get_data_type_ptr(),
448
0
                                     slot->col_name()});
449
0
        }
450
0
    }
451
0
}
452
453
0
VOlapTablePartitionParam::~VOlapTablePartitionParam() {
454
0
    _mem_tracker->release(_mem_usage);
455
0
}
456
457
0
Status VOlapTablePartitionParam::init() {
458
0
    std::vector<std::string> slot_column_names;
459
0
    for (auto* slot_desc : _schema->tuple_desc()->slots()) {
460
0
        slot_column_names.emplace_back(slot_desc->col_name());
461
0
    }
462
463
0
    auto find_slot_locs = [&slot_column_names](const std::string& slot_name,
464
0
                                               std::vector<uint16_t>& locs,
465
0
                                               const std::string& column_type) {
466
0
        auto it = std::find(slot_column_names.begin(), slot_column_names.end(), slot_name);
467
0
        if (it == slot_column_names.end()) {
468
0
            return Status::InternalError("{} column not found, column ={}", column_type, slot_name);
469
0
        }
470
0
        locs.emplace_back(it - slot_column_names.begin());
471
0
        return Status::OK();
472
0
    };
473
474
    // here we find the partition columns. others maybe non-partition columns/special columns.
475
0
    if (_t_param.__isset.partition_columns) {
476
0
        for (auto& part_col : _t_param.partition_columns) {
477
0
            RETURN_IF_ERROR(find_slot_locs(part_col, _partition_slot_locs, "partition"));
478
0
        }
479
0
    }
480
481
0
    _partitions_map = std::make_unique<
482
0
            std::map<BlockRowWithIndicator, VOlapTablePartition*, VOlapTablePartKeyComparator>>(
483
0
            VOlapTablePartKeyComparator(_partition_slot_locs, _transformed_slot_locs));
484
0
    if (_t_param.__isset.distributed_columns) {
485
0
        for (auto& col : _t_param.distributed_columns) {
486
0
            RETURN_IF_ERROR(find_slot_locs(col, _distributed_slot_locs, "distributed"));
487
0
        }
488
0
    }
489
490
    // for both auto/non-auto partition table.
491
0
    _is_in_partition = _part_type == TPartitionType::type::LIST_PARTITIONED;
492
493
    // initial partitions. if meet dummy partitions only for open BE nodes, not generate key of them for finding
494
0
    for (const auto& t_part : _t_param.partitions) {
495
0
        VOlapTablePartition* part = nullptr;
496
0
        RETURN_IF_ERROR(generate_partition_from(t_part, part));
497
0
        _partitions.emplace_back(part);
498
499
0
        if (!_t_param.partitions_is_fake) {
500
0
            if (_is_in_partition) {
501
0
                for (auto& in_key : part->in_keys) {
502
0
                    _partitions_map->emplace(std::tuple {in_key.first, in_key.second, false}, part);
503
0
                }
504
0
            } else {
505
0
                _partitions_map->emplace(
506
0
                        std::tuple {part->end_key.first, part->end_key.second, false}, part);
507
0
            }
508
0
        }
509
0
    }
510
511
0
    _mem_usage = _partition_block.allocated_bytes();
512
0
    _mem_tracker->consume(_mem_usage);
513
0
    return Status::OK();
514
0
}
515
516
bool VOlapTablePartitionParam::_part_contains(VOlapTablePartition* part,
517
0
                                              BlockRowWithIndicator key) const {
518
0
    VOlapTablePartKeyComparator comparator(_partition_slot_locs, _transformed_slot_locs);
519
    // we have used upper_bound to find to ensure key < part.right and this part is closest(right - key is min)
520
    // now we only have to check (key >= part.left). the comparator(a,b) means a < b, so we use anti
521
0
    return part->start_key.second == -1 /* spj: start_key.second == -1 means only single partition*/
522
0
           || !comparator(key, std::tuple {part->start_key.first, part->start_key.second, false});
523
0
}
524
525
// insert value into _partition_block's column
526
// NOLINTBEGIN(readability-function-size)
527
0
static Status _create_partition_key(const TExprNode& t_expr, BlockRow* part_key, uint16_t pos) {
528
0
    auto column = std::move(*part_key->first->get_by_position(pos).column).mutate();
529
    //TODO: use assert_cast before insert_data
530
0
    switch (t_expr.node_type) {
531
0
    case TExprNodeType::DATE_LITERAL: {
532
0
        if (TypeDescriptor::from_thrift(t_expr.type).is_date_v2_type()) {
533
0
            DateV2Value<DateV2ValueType> dt;
534
0
            if (!dt.from_date_str(t_expr.date_literal.value.c_str(),
535
0
                                  t_expr.date_literal.value.size())) {
536
0
                std::stringstream ss;
537
0
                ss << "invalid date literal in partition column, date=" << t_expr.date_literal;
538
0
                return Status::InternalError(ss.str());
539
0
            }
540
0
            column->insert_data(reinterpret_cast<const char*>(&dt), 0);
541
0
        } else if (TypeDescriptor::from_thrift(t_expr.type).is_datetime_v2_type()) {
542
0
            DateV2Value<DateTimeV2ValueType> dt;
543
0
            const int32_t scale =
544
0
                    t_expr.type.types.empty() ? -1 : t_expr.type.types.front().scalar_type.scale;
545
0
            if (!dt.from_date_str(t_expr.date_literal.value.c_str(),
546
0
                                  t_expr.date_literal.value.size(), scale)) {
547
0
                std::stringstream ss;
548
0
                ss << "invalid date literal in partition column, date=" << t_expr.date_literal;
549
0
                return Status::InternalError(ss.str());
550
0
            }
551
0
            column->insert_data(reinterpret_cast<const char*>(&dt), 0);
552
0
        } else {
553
0
            VecDateTimeValue dt;
554
0
            if (!dt.from_date_str(t_expr.date_literal.value.c_str(),
555
0
                                  t_expr.date_literal.value.size())) {
556
0
                std::stringstream ss;
557
0
                ss << "invalid date literal in partition column, date=" << t_expr.date_literal;
558
0
                return Status::InternalError(ss.str());
559
0
            }
560
0
            column->insert_data(reinterpret_cast<const char*>(&dt), 0);
561
0
        }
562
0
        break;
563
0
    }
564
0
    case TExprNodeType::INT_LITERAL: {
565
0
        switch (t_expr.type.types[0].scalar_type.type) {
566
0
        case TPrimitiveType::TINYINT: {
567
0
            int8_t value = t_expr.int_literal.value;
568
0
            column->insert_data(reinterpret_cast<const char*>(&value), 0);
569
0
            break;
570
0
        }
571
0
        case TPrimitiveType::SMALLINT: {
572
0
            int16_t value = t_expr.int_literal.value;
573
0
            column->insert_data(reinterpret_cast<const char*>(&value), 0);
574
0
            break;
575
0
        }
576
0
        case TPrimitiveType::INT: {
577
0
            int32_t value = t_expr.int_literal.value;
578
0
            column->insert_data(reinterpret_cast<const char*>(&value), 0);
579
0
            break;
580
0
        }
581
0
        default:
582
0
            int64_t value = t_expr.int_literal.value;
583
0
            column->insert_data(reinterpret_cast<const char*>(&value), 0);
584
0
        }
585
0
        break;
586
0
    }
587
0
    case TExprNodeType::LARGE_INT_LITERAL: {
588
0
        StringParser::ParseResult parse_result = StringParser::PARSE_SUCCESS;
589
0
        auto value = StringParser::string_to_int<__int128>(t_expr.large_int_literal.value.c_str(),
590
0
                                                           t_expr.large_int_literal.value.size(),
591
0
                                                           &parse_result);
592
0
        if (parse_result != StringParser::PARSE_SUCCESS) {
593
0
            value = MAX_INT128;
594
0
        }
595
0
        column->insert_data(reinterpret_cast<const char*>(&value), 0);
596
0
        break;
597
0
    }
598
0
    case TExprNodeType::STRING_LITERAL: {
599
0
        int len = t_expr.string_literal.value.size();
600
0
        const char* str_val = t_expr.string_literal.value.c_str();
601
0
        column->insert_data(str_val, len);
602
0
        break;
603
0
    }
604
0
    case TExprNodeType::BOOL_LITERAL: {
605
0
        column->insert_data(reinterpret_cast<const char*>(&t_expr.bool_literal.value), 0);
606
0
        break;
607
0
    }
608
0
    case TExprNodeType::NULL_LITERAL: {
609
        // insert a null literal
610
0
        if (!column->is_nullable()) {
611
            // https://github.com/apache/doris/pull/39449 have forbid this cause. always add this check as protective measures
612
0
            return Status::InternalError("The column {} is not null, can't insert into NULL value.",
613
0
                                         part_key->first->get_by_position(pos).name);
614
0
        }
615
0
        column->insert_data(nullptr, 0);
616
0
        break;
617
0
    }
618
0
    default: {
619
0
        return Status::InternalError("unsupported partition column node type, type={}",
620
0
                                     t_expr.node_type);
621
0
    }
622
0
    }
623
0
    part_key->second = column->size() - 1;
624
0
    return Status::OK();
625
0
}
626
// NOLINTEND(readability-function-size)
627
628
Status VOlapTablePartitionParam::_create_partition_keys(const std::vector<TExprNode>& t_exprs,
629
0
                                                        BlockRow* part_key) {
630
0
    for (int i = 0; i < t_exprs.size(); i++) {
631
0
        RETURN_IF_ERROR(_create_partition_key(t_exprs[i], part_key, _partition_slot_locs[i]));
632
0
    }
633
0
    return Status::OK();
634
0
}
635
636
Status VOlapTablePartitionParam::generate_partition_from(const TOlapTablePartition& t_part,
637
0
                                                         VOlapTablePartition*& part_result) {
638
0
    DCHECK(part_result == nullptr);
639
    // here we set the default value of partition bounds first! if it doesn't have some key, it will be -1.
640
0
    part_result = _obj_pool.add(new VOlapTablePartition(&_partition_block));
641
0
    part_result->id = t_part.id;
642
0
    part_result->is_mutable = t_part.is_mutable;
643
    // only load_to_single_tablet = true will set load_tablet_idx
644
0
    if (t_part.__isset.load_tablet_idx) {
645
0
        part_result->load_tablet_idx = t_part.load_tablet_idx;
646
0
    }
647
648
0
    if (_is_in_partition) {
649
0
        for (const auto& keys : t_part.in_keys) {
650
0
            RETURN_IF_ERROR(_create_partition_keys(
651
0
                    keys, &part_result->in_keys.emplace_back(&_partition_block, -1)));
652
0
        }
653
0
        if (t_part.__isset.is_default_partition && t_part.is_default_partition &&
654
0
            _default_partition == nullptr) {
655
0
            _default_partition = part_result;
656
0
        }
657
0
    } else { // range
658
0
        if (t_part.__isset.start_keys) {
659
0
            RETURN_IF_ERROR(_create_partition_keys(t_part.start_keys, &part_result->start_key));
660
0
        }
661
        // we generate the right bound but not insert into partition map
662
0
        if (t_part.__isset.end_keys) {
663
0
            RETURN_IF_ERROR(_create_partition_keys(t_part.end_keys, &part_result->end_key));
664
0
        }
665
0
    }
666
667
0
    part_result->num_buckets = t_part.num_buckets;
668
0
    auto num_indexes = _schema->indexes().size();
669
0
    if (t_part.indexes.size() != num_indexes) {
670
0
        return Status::InternalError(
671
0
                "number of partition's index is not equal with schema's"
672
0
                ", num_part_indexes={}, num_schema_indexes={}",
673
0
                t_part.indexes.size(), num_indexes);
674
0
    }
675
0
    part_result->indexes = t_part.indexes;
676
0
    std::sort(part_result->indexes.begin(), part_result->indexes.end(),
677
0
              [](const OlapTableIndexTablets& lhs, const OlapTableIndexTablets& rhs) {
678
0
                  return lhs.index_id < rhs.index_id;
679
0
              });
680
    // check index
681
0
    for (int j = 0; j < num_indexes; ++j) {
682
0
        if (part_result->indexes[j].index_id != _schema->indexes()[j]->index_id) {
683
0
            return Status::InternalError(
684
0
                    "partition's index is not equal with schema's"
685
0
                    ", part_index={}, schema_index={}",
686
0
                    part_result->indexes[j].index_id, _schema->indexes()[j]->index_id);
687
0
        }
688
0
    }
689
0
    return Status::OK();
690
0
}
691
692
Status VOlapTablePartitionParam::add_partitions(
693
0
        const std::vector<TOlapTablePartition>& partitions) {
694
0
    for (const auto& t_part : partitions) {
695
0
        auto* part = _obj_pool.add(new VOlapTablePartition(&_partition_block));
696
0
        part->id = t_part.id;
697
0
        part->is_mutable = t_part.is_mutable;
698
699
        // we dont pass right keys when it's MAX_VALUE. so there's possibility we only have start_key but not end_key
700
        // range partition
701
0
        if (t_part.__isset.start_keys) {
702
0
            RETURN_IF_ERROR(_create_partition_keys(t_part.start_keys, &part->start_key));
703
0
        }
704
0
        if (t_part.__isset.end_keys) {
705
0
            RETURN_IF_ERROR(_create_partition_keys(t_part.end_keys, &part->end_key));
706
0
        }
707
        // list partition - we only set 1 value in 1 partition for new created ones
708
0
        if (t_part.__isset.in_keys) {
709
0
            for (const auto& keys : t_part.in_keys) {
710
0
                RETURN_IF_ERROR(_create_partition_keys(
711
0
                        keys, &part->in_keys.emplace_back(&_partition_block, -1)));
712
0
            }
713
0
            if (t_part.__isset.is_default_partition && t_part.is_default_partition) {
714
0
                _default_partition = part;
715
0
            }
716
0
        }
717
718
0
        part->num_buckets = t_part.num_buckets;
719
0
        auto num_indexes = _schema->indexes().size();
720
0
        if (t_part.indexes.size() != num_indexes) {
721
0
            return Status::InternalError(
722
0
                    "number of partition's index is not equal with schema's"
723
0
                    ", num_part_indexes={}, num_schema_indexes={}",
724
0
                    t_part.indexes.size(), num_indexes);
725
0
        }
726
0
        part->indexes = t_part.indexes;
727
0
        std::sort(part->indexes.begin(), part->indexes.end(),
728
0
                  [](const OlapTableIndexTablets& lhs, const OlapTableIndexTablets& rhs) {
729
0
                      return lhs.index_id < rhs.index_id;
730
0
                  });
731
        // check index
732
0
        for (int j = 0; j < num_indexes; ++j) {
733
0
            if (part->indexes[j].index_id != _schema->indexes()[j]->index_id) {
734
0
                return Status::InternalError(
735
0
                        "partition's index is not equal with schema's"
736
0
                        ", part_index={}, schema_index={}",
737
0
                        part->indexes[j].index_id, _schema->indexes()[j]->index_id);
738
0
            }
739
0
        }
740
0
        _partitions.emplace_back(part);
741
        // after _creating_partiton_keys
742
0
        if (_is_in_partition) {
743
0
            for (auto& in_key : part->in_keys) {
744
0
                _partitions_map->emplace(std::tuple {in_key.first, in_key.second, false}, part);
745
0
            }
746
0
        } else {
747
0
            _partitions_map->emplace(std::tuple {part->end_key.first, part->end_key.second, false},
748
0
                                     part);
749
0
        }
750
0
    }
751
752
0
    return Status::OK();
753
0
}
754
755
Status VOlapTablePartitionParam::replace_partitions(
756
        std::vector<int64_t>& old_partition_ids,
757
0
        const std::vector<TOlapTablePartition>& new_partitions) {
758
    // remove old replaced partitions
759
0
    DCHECK(old_partition_ids.size() == new_partitions.size());
760
761
    // init and add new partitions. insert into _partitions
762
0
    for (int i = 0; i < new_partitions.size(); i++) {
763
0
        const auto& t_part = new_partitions[i];
764
        // pair old_partition_ids and new_partitions one by one. TODO: sort to opt performance
765
0
        VOlapTablePartition* old_part = nullptr;
766
0
        auto old_part_id = old_partition_ids[i];
767
0
        if (auto it = std::find_if(
768
0
                    _partitions.begin(), _partitions.end(),
769
0
                    [=](const VOlapTablePartition* lhs) { return lhs->id == old_part_id; });
770
0
            it != _partitions.end()) {
771
0
            old_part = *it;
772
0
        } else {
773
0
            return Status::InternalError("Cannot find old tablet {} in replacing", old_part_id);
774
0
        }
775
776
0
        auto* part = _obj_pool.add(new VOlapTablePartition(&_partition_block));
777
0
        part->id = t_part.id;
778
0
        part->is_mutable = t_part.is_mutable;
779
780
        /// just substitute directly. no need to remove and reinsert keys.
781
        // range partition
782
0
        part->start_key = std::move(old_part->start_key);
783
0
        part->end_key = std::move(old_part->end_key);
784
        // list partition
785
0
        part->in_keys = std::move(old_part->in_keys);
786
0
        if (t_part.__isset.is_default_partition && t_part.is_default_partition) {
787
0
            _default_partition = part;
788
0
        }
789
790
0
        part->num_buckets = t_part.num_buckets;
791
0
        auto num_indexes = _schema->indexes().size();
792
0
        if (t_part.indexes.size() != num_indexes) {
793
0
            return Status::InternalError(
794
0
                    "number of partition's index is not equal with schema's"
795
0
                    ", num_part_indexes={}, num_schema_indexes={}",
796
0
                    t_part.indexes.size(), num_indexes);
797
0
        }
798
0
        part->indexes = t_part.indexes;
799
0
        std::sort(part->indexes.begin(), part->indexes.end(),
800
0
                  [](const OlapTableIndexTablets& lhs, const OlapTableIndexTablets& rhs) {
801
0
                      return lhs.index_id < rhs.index_id;
802
0
                  });
803
        // check index
804
0
        for (int j = 0; j < num_indexes; ++j) {
805
0
            if (part->indexes[j].index_id != _schema->indexes()[j]->index_id) {
806
0
                return Status::InternalError(
807
0
                        "partition's index is not equal with schema's"
808
0
                        ", part_index={}, schema_index={}",
809
0
                        part->indexes[j].index_id, _schema->indexes()[j]->index_id);
810
0
            }
811
0
        }
812
813
        // add new partitions with new id.
814
0
        _partitions.emplace_back(part);
815
0
        VLOG_NOTICE << "params add new partition " << part->id;
816
817
        // replace items in _partition_maps
818
0
        if (_is_in_partition) {
819
0
            for (auto& in_key : part->in_keys) {
820
0
                (*_partitions_map)[std::tuple {in_key.first, in_key.second, false}] = part;
821
0
            }
822
0
        } else {
823
0
            (*_partitions_map)[std::tuple {part->end_key.first, part->end_key.second, false}] =
824
0
                    part;
825
0
        }
826
0
    }
827
    // remove old partitions by id
828
0
    std::ranges::sort(old_partition_ids);
829
0
    for (auto it = _partitions.begin(); it != _partitions.end();) {
830
0
        if (std::ranges::binary_search(old_partition_ids, (*it)->id)) {
831
0
            it = _partitions.erase(it);
832
0
        } else {
833
0
            it++;
834
0
        }
835
0
    }
836
837
0
    return Status::OK();
838
0
}
839
840
} // namespace doris