Coverage Report

Created: 2026-04-10 16:11

next uncovered line (L), next uncovered region (R), next uncovered branch (B)
be/src/service/point_query_executor.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 "service/point_query_executor.h"
19
20
#include <fmt/format.h>
21
#include <gen_cpp/Descriptors_types.h>
22
#include <gen_cpp/Exprs_types.h>
23
#include <gen_cpp/PaloInternalService_types.h>
24
#include <gen_cpp/internal_service.pb.h>
25
#include <glog/logging.h>
26
#include <google/protobuf/extension_set.h>
27
#include <stdlib.h>
28
29
#include <memory>
30
#include <unordered_map>
31
#include <vector>
32
33
#include "cloud/cloud_tablet.h"
34
#include "cloud/config.h"
35
#include "common/cast_set.h"
36
#include "common/consts.h"
37
#include "common/status.h"
38
#include "core/data_type/data_type_factory.hpp"
39
#include "core/data_type_serde/data_type_serde.h"
40
#include "exec/sink/writer/vmysql_result_writer.h"
41
#include "exprs/vexpr.h"
42
#include "exprs/vexpr_context.h"
43
#include "exprs/vexpr_fwd.h"
44
#include "exprs/vslot_ref.h"
45
#include "runtime/descriptors.h"
46
#include "runtime/exec_env.h"
47
#include "runtime/result_block_buffer.h"
48
#include "runtime/runtime_profile.h"
49
#include "runtime/runtime_state.h"
50
#include "runtime/thread_context.h"
51
#include "storage/row_cursor.h"
52
#include "storage/rowset/beta_rowset.h"
53
#include "storage/rowset/rowset_fwd.h"
54
#include "storage/segment/column_reader.h"
55
#include "storage/tablet/tablet_schema.h"
56
#include "storage/utils.h"
57
#include "util/jsonb/serialize.h"
58
#include "util/lru_cache.h"
59
#include "util/simd/bits.h"
60
#include "util/thrift_util.h"
61
62
namespace doris {
63
64
class PointQueryResultBlockBuffer final : public MySQLResultBlockBuffer {
65
public:
66
249
    PointQueryResultBlockBuffer(RuntimeState* state) : MySQLResultBlockBuffer(state) {}
67
    ~PointQueryResultBlockBuffer() override = default;
68
249
    std::shared_ptr<TFetchDataResult> get_block() {
69
249
        std::lock_guard<std::mutex> l(_lock);
70
249
        DCHECK_EQ(_result_batch_queue.size(), 1);
71
249
        auto result = std::move(_result_batch_queue.front());
72
249
        _result_batch_queue.pop_front();
73
249
        return result;
74
249
    }
75
};
76
77
4.16k
Reusable::~Reusable() = default;
78
79
// get missing and include column ids
80
// input include_cids : the output expr slots columns unique ids
81
// missing_cids : the output expr columns that not in row columns cids
82
static void get_missing_and_include_cids(const TabletSchema& schema,
83
                                         const std::vector<SlotDescriptor*>& slots,
84
                                         int target_rs_column_id,
85
                                         std::unordered_set<int>& missing_cids,
86
4.20k
                                         std::unordered_set<int>& include_cids) {
87
4.20k
    missing_cids.clear();
88
4.20k
    include_cids.clear();
89
5.20k
    for (auto* slot : slots) {
90
5.20k
        missing_cids.insert(slot->col_unique_id());
91
5.20k
    }
92
    // insert delete sign column id
93
4.20k
    missing_cids.insert(schema.columns()[schema.delete_sign_idx()]->unique_id());
94
4.20k
    if (target_rs_column_id == -1) {
95
        // no row store columns
96
4.04k
        return;
97
4.04k
    }
98
154
    const TabletColumn& target_rs_column = schema.column_by_uid(target_rs_column_id);
99
154
    DCHECK(target_rs_column.is_row_store_column());
100
    // The full column group is considered a full match, thus no missing cids
101
154
    if (schema.row_columns_uids().empty()) {
102
144
        missing_cids.clear();
103
144
        return;
104
144
    }
105
18
    for (int cid : schema.row_columns_uids()) {
106
18
        missing_cids.erase(cid);
107
18
        include_cids.insert(cid);
108
18
    }
109
10
}
110
111
constexpr static int s_preallocted_blocks_num = 32;
112
113
static void extract_slot_ref(const VExprSPtr& expr, TupleDescriptor* tuple_desc,
114
5.41k
                             std::vector<SlotDescriptor*>& slots) {
115
5.41k
    const auto& children = expr->children();
116
5.41k
    for (const auto& i : children) {
117
210
        extract_slot_ref(i, tuple_desc, slots);
118
210
    }
119
120
5.41k
    auto node_type = expr->node_type();
121
5.41k
    if (node_type == TExprNodeType::SLOT_REF) {
122
5.20k
        int column_id = static_cast<const VSlotRef*>(expr.get())->column_id();
123
5.20k
        auto* slot_desc = tuple_desc->slots()[column_id];
124
5.20k
        slots.push_back(slot_desc);
125
5.20k
    }
126
5.41k
}
127
128
Status Reusable::init(const TDescriptorTable& t_desc_tbl, const std::vector<TExpr>& output_exprs,
129
                      const TQueryOptions& query_options, const TabletSchema& schema,
130
4.20k
                      size_t block_size) {
131
4.20k
    _runtime_state = RuntimeState::create_unique();
132
4.20k
    _runtime_state->set_query_options(query_options);
133
4.20k
    RETURN_IF_ERROR(DescriptorTbl::create(_runtime_state->obj_pool(), t_desc_tbl, &_desc_tbl));
134
4.20k
    _runtime_state->set_desc_tbl(_desc_tbl);
135
4.20k
    _block_pool.resize(block_size);
136
10.9k
    for (auto& i : _block_pool) {
137
10.9k
        i = Block::create_unique(tuple_desc()->slots(), 2);
138
        // Name is useless but cost space
139
10.9k
        i->clear_names();
140
10.9k
    }
141
142
4.20k
    RETURN_IF_ERROR(VExpr::create_expr_trees(output_exprs, _output_exprs_ctxs));
143
4.20k
    RowDescriptor row_desc(tuple_desc());
144
    // Prepare the exprs to run.
145
4.20k
    RETURN_IF_ERROR(VExpr::prepare(_output_exprs_ctxs, _runtime_state.get(), row_desc));
146
4.20k
    RETURN_IF_ERROR(VExpr::open(_output_exprs_ctxs, _runtime_state.get()));
147
4.20k
    _create_timestamp = butil::gettimeofday_ms();
148
4.20k
    _data_type_serdes = create_data_type_serdes(tuple_desc()->slots());
149
4.20k
    _col_default_values.resize(tuple_desc()->slots().size());
150
4.20k
    bool has_delete_sign = false;
151
94.6k
    for (int i = 0; i < tuple_desc()->slots().size(); ++i) {
152
90.4k
        auto* slot = tuple_desc()->slots()[i];
153
90.4k
        if (slot->col_name() == DELETE_SIGN) {
154
153
            has_delete_sign = true;
155
153
        }
156
90.4k
        _col_uid_to_idx[slot->col_unique_id()] = i;
157
90.4k
        _col_default_values[i] = slot->col_default_value();
158
90.4k
    }
159
160
    // Get the output slot descriptors
161
4.20k
    std::vector<SlotDescriptor*> output_slot_descs;
162
5.20k
    for (const auto& expr : _output_exprs_ctxs) {
163
5.20k
        extract_slot_ref(expr->root(), tuple_desc(), output_slot_descs);
164
5.20k
    }
165
166
    // get the delete sign idx in block
167
4.20k
    if (has_delete_sign) {
168
153
        _delete_sign_idx = _col_uid_to_idx[schema.columns()[schema.delete_sign_idx()]->unique_id()];
169
153
    }
170
171
4.20k
    if (schema.have_column(BeConsts::ROW_STORE_COL)) {
172
154
        const auto& column = *DORIS_TRY(schema.column(BeConsts::ROW_STORE_COL));
173
154
        _row_store_column_ids = column.unique_id();
174
154
    }
175
4.20k
    get_missing_and_include_cids(schema, output_slot_descs, _row_store_column_ids,
176
4.20k
                                 _missing_col_uids, _include_col_uids);
177
178
4.20k
    return Status::OK();
179
4.20k
}
180
181
263
std::unique_ptr<Block> Reusable::get_block() {
182
263
    std::lock_guard lock(_block_mutex);
183
263
    if (_block_pool.empty()) {
184
0
        auto block = Block::create_unique(tuple_desc()->slots(), 2);
185
        // Name is useless but cost space
186
0
        block->clear_names();
187
0
        return block;
188
0
    }
189
263
    auto block = std::move(_block_pool.back());
190
263
    CHECK(block != nullptr);
191
263
    _block_pool.pop_back();
192
263
    return block;
193
263
}
194
195
260
void Reusable::return_block(std::unique_ptr<Block>& block) {
196
260
    std::lock_guard lock(_block_mutex);
197
260
    if (block == nullptr) {
198
0
        return;
199
0
    }
200
260
    block->clear_column_data();
201
260
    _block_pool.push_back(std::move(block));
202
260
    if (_block_pool.size() > s_preallocted_blocks_num) {
203
0
        _block_pool.resize(s_preallocted_blocks_num);
204
0
    }
205
260
}
206
207
7
LookupConnectionCache* LookupConnectionCache::create_global_instance(size_t capacity) {
208
7
    DCHECK(ExecEnv::GetInstance()->get_lookup_connection_cache() == nullptr);
209
7
    auto* res = new LookupConnectionCache(capacity);
210
7
    return res;
211
7
}
212
213
RowCache::RowCache(int64_t capacity, int num_shards)
214
10
        : LRUCachePolicy(CachePolicy::CacheType::POINT_QUERY_ROW_CACHE, capacity,
215
10
                         LRUCacheType::SIZE, config::point_query_row_cache_stale_sweep_time_sec,
216
10
                         num_shards, /*element count capacity */ 0,
217
10
                         /*enable prune*/ true, /*is lru-k*/ true) {}
218
219
// Create global instance of this class
220
7
RowCache* RowCache::create_global_cache(int64_t capacity, uint32_t num_shards) {
221
7
    DCHECK(ExecEnv::GetInstance()->get_row_cache() == nullptr);
222
7
    auto* res = new RowCache(capacity, num_shards);
223
7
    return res;
224
7
}
225
226
307
RowCache* RowCache::instance() {
227
307
    return ExecEnv::GetInstance()->get_row_cache();
228
307
}
229
230
152
bool RowCache::lookup(const RowCacheKey& key, CacheHandle* handle) {
231
152
    const std::string& encoded_key = key.encode();
232
152
    auto* lru_handle = LRUCachePolicy::lookup(encoded_key);
233
152
    if (!lru_handle) {
234
        // cache miss
235
74
        return false;
236
74
    }
237
78
    *handle = CacheHandle(this, lru_handle);
238
78
    return true;
239
152
}
240
241
76
void RowCache::insert(const RowCacheKey& key, const Slice& value) {
242
76
    char* cache_value = static_cast<char*>(malloc(value.size));
243
76
    memcpy(cache_value, value.data, value.size);
244
76
    auto* row_cache_value = new RowCacheValue;
245
76
    row_cache_value->cache_value = cache_value;
246
76
    const std::string& encoded_key = key.encode();
247
76
    auto* handle = LRUCachePolicy::insert(encoded_key, row_cache_value, value.size, value.size,
248
76
                                          CachePriority::NORMAL);
249
    // handle will released
250
76
    auto tmp = CacheHandle {this, handle};
251
76
}
252
253
95
void RowCache::erase(const RowCacheKey& key) {
254
95
    const std::string& encoded_key = key.encode();
255
95
    LRUCachePolicy::erase(encoded_key);
256
95
}
257
258
2.09k
LookupConnectionCache::CacheValue::~CacheValue() {
259
2.09k
    SCOPED_SWITCH_THREAD_MEM_TRACKER_LIMITER(
260
2.09k
            ExecEnv::GetInstance()->point_query_executor_mem_tracker());
261
2.09k
    item.reset();
262
2.09k
}
263
264
263
PointQueryExecutor::~PointQueryExecutor() {
265
263
    SCOPED_SWITCH_THREAD_MEM_TRACKER_LIMITER(
266
263
            ExecEnv::GetInstance()->point_query_executor_mem_tracker());
267
263
    _tablet.reset();
268
263
    _reusable.reset();
269
263
    _result_block.reset();
270
263
    _row_read_ctxs.clear();
271
263
}
272
273
Status PointQueryExecutor::init(const PTabletKeyLookupRequest* request,
274
263
                                PTabletKeyLookupResponse* response) {
275
263
    SCOPED_TIMER(&_profile_metrics.init_ns);
276
263
    _response = response;
277
    // using cache
278
263
    __int128_t uuid =
279
263
            static_cast<__int128_t>(request->uuid().uuid_high()) << 64 | request->uuid().uuid_low();
280
263
    SCOPED_ATTACH_TASK(ExecEnv::GetInstance()->point_query_executor_mem_tracker());
281
263
    auto cache_handle = LookupConnectionCache::instance()->get(uuid);
282
263
    _binary_row_format = request->is_binary_row();
283
263
    _tablet = DORIS_TRY(ExecEnv::get_tablet(request->tablet_id()));
284
263
    if (cache_handle != nullptr) {
285
109
        _reusable = cache_handle;
286
109
        _profile_metrics.hit_lookup_cache = true;
287
154
    } else {
288
        // init handle
289
154
        auto reusable_ptr = std::make_shared<Reusable>();
290
154
        TDescriptorTable t_desc_tbl;
291
154
        TExprList t_output_exprs;
292
154
        auto len = cast_set<uint32_t>(request->desc_tbl().size());
293
154
        RETURN_IF_ERROR(
294
154
                deserialize_thrift_msg(reinterpret_cast<const uint8_t*>(request->desc_tbl().data()),
295
154
                                       &len, false, &t_desc_tbl));
296
154
        len = cast_set<uint32_t>(request->output_expr().size());
297
154
        RETURN_IF_ERROR(deserialize_thrift_msg(
298
154
                reinterpret_cast<const uint8_t*>(request->output_expr().data()), &len, false,
299
154
                &t_output_exprs));
300
154
        _reusable = reusable_ptr;
301
154
        TQueryOptions t_query_options;
302
154
        len = cast_set<uint32_t>(request->query_options().size());
303
154
        if (request->has_query_options()) {
304
154
            RETURN_IF_ERROR(deserialize_thrift_msg(
305
154
                    reinterpret_cast<const uint8_t*>(request->query_options().data()), &len, false,
306
154
                    &t_query_options));
307
154
        }
308
154
        if (uuid != 0) {
309
            // could be reused by requests after, pre allocte more blocks
310
88
            RETURN_IF_ERROR(reusable_ptr->init(t_desc_tbl, t_output_exprs.exprs, t_query_options,
311
88
                                               *_tablet->tablet_schema(),
312
88
                                               s_preallocted_blocks_num));
313
88
            LookupConnectionCache::instance()->add(uuid, reusable_ptr);
314
88
        } else {
315
66
            RETURN_IF_ERROR(reusable_ptr->init(t_desc_tbl, t_output_exprs.exprs, t_query_options,
316
66
                                               *_tablet->tablet_schema(), 1));
317
66
        }
318
154
    }
319
    // Set timezone from request for functions like from_unixtime()
320
263
    if (request->has_time_zone() && !request->time_zone().empty()) {
321
263
        _reusable->runtime_state()->set_timezone(request->time_zone());
322
263
    }
323
263
    if (request->has_version() && request->version() >= 0) {
324
263
        _version = request->version();
325
263
    }
326
263
    RETURN_IF_ERROR(_init_keys(request));
327
263
    _result_block = _reusable->get_block();
328
263
    CHECK(_result_block != nullptr);
329
330
263
    return Status::OK();
331
263
}
332
333
263
Status PointQueryExecutor::lookup_up() {
334
263
    SCOPED_ATTACH_TASK(ExecEnv::GetInstance()->point_query_executor_mem_tracker());
335
263
    RETURN_IF_ERROR(_lookup_row_key());
336
263
    RETURN_IF_ERROR(_lookup_row_data());
337
260
    RETURN_IF_ERROR(_output_data());
338
260
    return Status::OK();
339
260
}
340
341
260
void PointQueryExecutor::print_profile() {
342
260
    auto init_us = _profile_metrics.init_ns.value() / 1000;
343
260
    auto init_key_us = _profile_metrics.init_key_ns.value() / 1000;
344
260
    auto lookup_key_us = _profile_metrics.lookup_key_ns.value() / 1000;
345
260
    auto lookup_data_us = _profile_metrics.lookup_data_ns.value() / 1000;
346
260
    auto output_data_us = _profile_metrics.output_data_ns.value() / 1000;
347
260
    auto load_segments_key_us = _profile_metrics.load_segment_key_stage_ns.value() / 1000;
348
260
    auto load_segments_data_us = _profile_metrics.load_segment_data_stage_ns.value() / 1000;
349
260
    auto total_us = init_us + lookup_key_us + lookup_data_us + output_data_us;
350
260
    auto read_stats = _profile_metrics.read_stats;
351
260
    const std::string stats_str = fmt::format(
352
260
            "[lookup profile:{}us] init:{}us, init_key:{}us,"
353
260
            " lookup_key:{}us, load_segments_key:{}us, lookup_data:{}us, load_segments_data:{}us,"
354
260
            " output_data:{}us, "
355
260
            "hit_lookup_cache:{}"
356
260
            ", is_binary_row:{}, output_columns:{}, total_keys:{}, row_cache_hits:{}"
357
260
            ", hit_cached_pages:{}, total_pages_read:{}, compressed_bytes_read:{}, "
358
260
            "io_latency:{}ns, "
359
260
            "uncompressed_bytes_read:{}, result_data_bytes:{}, row_hits:{}"
360
260
            ", rs_column_uid:{}, bytes_read_from_local:{}, bytes_read_from_remote:{}, "
361
260
            "local_io_timer:{}, remote_io_timer:{}, local_write_timer:{}",
362
260
            total_us, init_us, init_key_us, lookup_key_us, load_segments_key_us, lookup_data_us,
363
260
            load_segments_data_us, output_data_us, _profile_metrics.hit_lookup_cache,
364
260
            _binary_row_format, _reusable->output_exprs().size(), _row_read_ctxs.size(),
365
260
            _profile_metrics.row_cache_hits, read_stats.cached_pages_num,
366
260
            read_stats.total_pages_num, read_stats.compressed_bytes_read, read_stats.io_ns,
367
260
            read_stats.uncompressed_bytes_read, _profile_metrics.result_data_bytes, _row_hits,
368
260
            _reusable->rs_column_uid(),
369
260
            _profile_metrics.read_stats.file_cache_stats.bytes_read_from_local,
370
260
            _profile_metrics.read_stats.file_cache_stats.bytes_read_from_remote,
371
260
            _profile_metrics.read_stats.file_cache_stats.local_io_timer,
372
260
            _profile_metrics.read_stats.file_cache_stats.remote_io_timer,
373
260
            _profile_metrics.read_stats.file_cache_stats.write_cache_io_timer);
374
375
260
    constexpr static int kSlowThreholdUs = 50 * 1000; // 50ms
376
260
    if (total_us > kSlowThreholdUs) {
377
6
        LOG(WARNING) << "slow query, " << stats_str;
378
254
    } else if (VLOG_DEBUG_IS_ON) {
379
0
        VLOG_DEBUG << stats_str;
380
254
    } else {
381
254
        LOG_EVERY_N(INFO, 1000) << stats_str;
382
254
    }
383
260
}
384
385
263
Status PointQueryExecutor::_init_keys(const PTabletKeyLookupRequest* request) {
386
263
    SCOPED_TIMER(&_profile_metrics.init_key_ns);
387
263
    const auto& schema = _tablet->tablet_schema();
388
    // Point query is only supported on merge-on-write unique key tables.
389
263
    DCHECK(schema->keys_type() == UNIQUE_KEYS && _tablet->enable_unique_key_merge_on_write());
390
263
    if (schema->keys_type() != UNIQUE_KEYS || !_tablet->enable_unique_key_merge_on_write()) {
391
0
        return Status::InvalidArgument(
392
0
                "Point query is only supported on merge-on-write unique key tables, "
393
0
                "tablet_id={}",
394
0
                _tablet->tablet_id());
395
0
    }
396
    // 1. get primary key from conditions
397
263
    _row_read_ctxs.resize(request->key_tuples().size());
398
    // get row cursor and encode keys
399
526
    for (int i = 0; i < request->key_tuples().size(); ++i) {
400
263
        const KeyTuple& key_tuple = request->key_tuples(i);
401
263
        if (UNLIKELY(cast_set<size_t>(key_tuple.key_column_literals_size()) !=
402
263
                     schema->num_key_columns())) {
403
0
            return Status::InvalidArgument(
404
0
                    "Key column count mismatch. expected={}, actual={}, tablet_id={}",
405
0
                    schema->num_key_columns(), key_tuple.key_column_literals_size(),
406
0
                    _tablet->tablet_id());
407
0
        }
408
263
        RowCursor cursor;
409
263
        std::vector<Field> key_fields;
410
263
        key_fields.reserve(key_tuple.key_column_literals_size());
411
1.20k
        for (int j = 0; j < key_tuple.key_column_literals_size(); ++j) {
412
946
            const auto& literal_bytes = key_tuple.key_column_literals(j);
413
946
            TExprNode expr_node;
414
946
            auto len = cast_set<uint32_t>(literal_bytes.size());
415
946
            RETURN_IF_ERROR(
416
946
                    deserialize_thrift_msg(reinterpret_cast<const uint8_t*>(literal_bytes.data()),
417
946
                                           &len, false, &expr_node));
418
946
            const auto& col = schema->column(j);
419
946
            auto data_type = DataTypeFactory::instance().create_data_type(
420
946
                    col.type(), col.precision(), col.frac(), col.length());
421
946
            key_fields.push_back(data_type->get_field(expr_node));
422
946
        }
423
263
        RETURN_IF_ERROR(cursor.init_scan_key(_tablet->tablet_schema(), std::move(key_fields)));
424
263
        cursor.encode_key_with_padding<true>(&_row_read_ctxs[i]._primary_key,
425
263
                                             _tablet->tablet_schema()->num_key_columns(), true);
426
263
    }
427
263
    return Status::OK();
428
263
}
429
430
263
Status PointQueryExecutor::_lookup_row_key() {
431
263
    SCOPED_TIMER(&_profile_metrics.lookup_key_ns);
432
    // 2. lookup row location
433
263
    Status st;
434
263
    if (_version >= 0) {
435
263
        CHECK(config::is_cloud_mode()) << "Only cloud mode support snapshot read at present";
436
263
        SyncOptions options;
437
263
        options.query_version = _version;
438
263
        RETURN_IF_ERROR(std::dynamic_pointer_cast<CloudTablet>(_tablet)->sync_rowsets(options));
439
263
    }
440
263
    std::vector<RowsetSharedPtr> specified_rowsets;
441
263
    {
442
263
        std::shared_lock rlock(_tablet->get_header_lock());
443
263
        specified_rowsets = _tablet->get_rowset_by_ids(nullptr);
444
263
    }
445
263
    std::vector<std::unique_ptr<SegmentCacheHandle>> segment_caches(specified_rowsets.size());
446
526
    for (size_t i = 0; i < _row_read_ctxs.size(); ++i) {
447
263
        RowLocation location;
448
263
        if (!config::disable_storage_row_cache) {
449
148
            RowCache::CacheHandle cache_handle;
450
148
            auto hit_cache = RowCache::instance()->lookup(
451
148
                    {_tablet->tablet_id(), _row_read_ctxs[i]._primary_key}, &cache_handle);
452
148
            if (hit_cache) {
453
77
                _row_read_ctxs[i]._cached_row_data = std::move(cache_handle);
454
77
                ++_profile_metrics.row_cache_hits;
455
77
                continue;
456
77
            }
457
148
        }
458
        // Get rowlocation and rowset, ctx._rowset_ptr will acquire wrap this ptr
459
186
        auto rowset_ptr = std::make_unique<RowsetSharedPtr>();
460
186
        st = (_tablet->lookup_row_key(_row_read_ctxs[i]._primary_key, nullptr, false,
461
186
                                      specified_rowsets, &location, INT32_MAX /*rethink?*/,
462
186
                                      segment_caches, rowset_ptr.get(), false, nullptr,
463
186
                                      &_profile_metrics.read_stats));
464
186
        if (st.is<ErrorCode::KEY_NOT_FOUND>()) {
465
9
            continue;
466
9
        }
467
177
        RETURN_IF_ERROR(st);
468
177
        _row_read_ctxs[i]._row_location = location;
469
        // acquire and wrap this rowset
470
177
        (*rowset_ptr)->acquire();
471
177
        VLOG_DEBUG << "aquire rowset " << (*rowset_ptr)->rowset_id();
472
177
        _row_read_ctxs[i]._rowset_ptr = std::unique_ptr<RowsetSharedPtr, decltype(&release_rowset)>(
473
177
                rowset_ptr.release(), &release_rowset);
474
177
        _row_hits++;
475
177
    }
476
263
    return Status::OK();
477
263
}
478
479
263
Status PointQueryExecutor::_lookup_row_data() {
480
    // 3. get values
481
263
    SCOPED_TIMER(&_profile_metrics.lookup_data_ns);
482
523
    for (size_t i = 0; i < _row_read_ctxs.size(); ++i) {
483
263
        if (_row_read_ctxs[i]._cached_row_data.valid()) {
484
77
            RETURN_IF_ERROR(JsonbSerializeUtil::jsonb_to_block(
485
77
                    _reusable->get_data_type_serdes(),
486
77
                    _row_read_ctxs[i]._cached_row_data.data().data,
487
77
                    _row_read_ctxs[i]._cached_row_data.data().size, _reusable->get_col_uid_to_idx(),
488
77
                    *_result_block, _reusable->get_col_default_values(),
489
77
                    _reusable->include_col_uids()));
490
77
            continue;
491
77
        }
492
186
        if (!_row_read_ctxs[i]._row_location.has_value()) {
493
9
            continue;
494
9
        }
495
177
        std::string value;
496
        // fill block by row store
497
177
        if (_reusable->rs_column_uid() != -1) {
498
177
            bool use_row_cache = !config::disable_storage_row_cache;
499
177
            RETURN_IF_ERROR(_tablet->lookup_row_data(
500
177
                    _row_read_ctxs[i]._primary_key, _row_read_ctxs[i]._row_location.value(),
501
177
                    *(_row_read_ctxs[i]._rowset_ptr), _profile_metrics.read_stats, value,
502
177
                    use_row_cache));
503
            // serilize value to block, currently only jsonb row formt
504
177
            RETURN_IF_ERROR(JsonbSerializeUtil::jsonb_to_block(
505
177
                    _reusable->get_data_type_serdes(), value.data(), value.size(),
506
177
                    _reusable->get_col_uid_to_idx(), *_result_block,
507
177
                    _reusable->get_col_default_values(), _reusable->include_col_uids()));
508
177
        }
509
177
        if (!_reusable->missing_col_uids().empty()) {
510
10
            if (!_reusable->runtime_state()->enable_short_circuit_query_access_column_store()) {
511
3
                std::string missing_columns;
512
22
                for (int cid : _reusable->missing_col_uids()) {
513
22
                    missing_columns += _tablet->tablet_schema()->column_by_uid(cid).name() + ",";
514
22
                }
515
3
                return Status::InternalError(
516
3
                        "Not support column store, set store_row_column=true or row_store_columns "
517
3
                        "in table "
518
3
                        "properties, missing columns: " +
519
3
                        missing_columns + " should be added to row store");
520
3
            }
521
            // fill missing columns by column store
522
7
            RowLocation row_loc = _row_read_ctxs[i]._row_location.value();
523
7
            BetaRowsetSharedPtr rowset =
524
7
                    std::static_pointer_cast<BetaRowset>(_tablet->get_rowset(row_loc.rowset_id));
525
7
            SegmentCacheHandle segment_cache;
526
7
            {
527
7
                SCOPED_TIMER(&_profile_metrics.load_segment_data_stage_ns);
528
7
                RETURN_IF_ERROR(
529
7
                        SegmentLoader::instance()->load_segments(rowset, &segment_cache, true));
530
7
            }
531
            // find segment
532
7
            auto it = std::find_if(segment_cache.get_segments().cbegin(),
533
7
                                   segment_cache.get_segments().cend(),
534
7
                                   [&](const segment_v2::SegmentSharedPtr& seg) {
535
7
                                       return seg->id() == row_loc.segment_id;
536
7
                                   });
537
7
            const auto& segment = *it;
538
8
            for (int cid : _reusable->missing_col_uids()) {
539
8
                int pos = _reusable->get_col_uid_to_idx().at(cid);
540
8
                auto row_id = static_cast<segment_v2::rowid_t>(row_loc.row_id);
541
8
                MutableColumnPtr column =
542
8
                        _result_block->get_by_position(pos).column->assume_mutable();
543
8
                std::unique_ptr<ColumnIterator> iter;
544
8
                SlotDescriptor* slot = _reusable->tuple_desc()->slots()[pos];
545
8
                StorageReadOptions storage_read_options;
546
8
                storage_read_options.stats = &_read_stats;
547
8
                storage_read_options.io_ctx.reader_type = ReaderType::READER_QUERY;
548
8
                RETURN_IF_ERROR(segment->seek_and_read_by_rowid(*_tablet->tablet_schema(), slot,
549
8
                                                                row_id, column,
550
8
                                                                storage_read_options, iter));
551
8
                if (_tablet->tablet_schema()
552
8
                            ->column_by_uid(slot->col_unique_id())
553
8
                            .has_char_type()) {
554
1
                    column->shrink_padding_chars();
555
1
                }
556
8
            }
557
7
        }
558
177
    }
559
260
    if (_result_block->columns() > _reusable->include_col_uids().size()) {
560
        // Padding rows for some columns that no need to output to mysql client
561
        // eg. SELECT k1,v1,v2 FROM TABLE WHERE k1 = 1, k1 is not in output slots, tuple as bellow
562
        // TupleDescriptor{id=1, tbl=table_with_column_group}
563
        // SlotDescriptor{id=8, col=v1, colUniqueId=1 ...}
564
        // SlotDescriptor{id=9, col=v2, colUniqueId=2 ...}
565
        // thus missing in include_col_uids and missing_col_uids
566
2.90k
        for (size_t i = 0; i < _result_block->columns(); ++i) {
567
2.64k
            auto column = _result_block->get_by_position(i).column;
568
2.64k
            int padding_rows = _row_hits - cast_set<int>(column->size());
569
2.64k
            if (padding_rows > 0) {
570
1
                column->assume_mutable()->insert_many_defaults(padding_rows);
571
1
            }
572
2.64k
        }
573
260
    }
574
    // filter rows by delete sign
575
260
    if (_row_hits > 0 && _reusable->delete_sign_idx() != -1) {
576
173
        size_t filtered = 0;
577
173
        size_t total = 0;
578
173
        {
579
            // clear_column_data will check reference of ColumnPtr, so we need to release
580
            // reference before clear_column_data
581
173
            ColumnPtr delete_filter_columns =
582
173
                    _result_block->get_columns()[_reusable->delete_sign_idx()];
583
173
            const auto& filter =
584
173
                    assert_cast<const ColumnInt8*>(delete_filter_columns.get())->get_data();
585
173
            filtered = filter.size() - simd::count_zero_num((int8_t*)filter.data(), filter.size());
586
173
            total = filter.size();
587
173
        }
588
589
173
        if (filtered == total) {
590
2
            _result_block->clear_column_data();
591
171
        } else if (filtered > 0) {
592
0
            return Status::NotSupported("Not implemented since only single row at present");
593
0
        }
594
173
    }
595
260
    return Status::OK();
596
260
}
597
598
249
Status serialize_block(std::shared_ptr<TFetchDataResult> res, PTabletKeyLookupResponse* response) {
599
249
    uint8_t* buf = nullptr;
600
249
    uint32_t len = 0;
601
249
    ThriftSerializer ser(false, 4096);
602
249
    RETURN_IF_ERROR(ser.serialize(&(res->result_batch), &len, &buf));
603
249
    response->set_row_batch(std::string((const char*)buf, len));
604
249
    return Status::OK();
605
249
}
606
607
260
Status PointQueryExecutor::_output_data() {
608
    // 4. exprs exec and serialize to mysql row batches
609
260
    SCOPED_TIMER(&_profile_metrics.output_data_ns);
610
260
    if (_result_block->rows()) {
611
249
        RuntimeState state;
612
249
        auto buffer = std::make_shared<PointQueryResultBlockBuffer>(&state);
613
        // TODO reuse mysql_writer
614
249
        VMysqlResultWriter mysql_writer(buffer, _reusable->output_exprs(), nullptr,
615
249
                                        _binary_row_format);
616
249
        RETURN_IF_ERROR(mysql_writer.init(_reusable->runtime_state()));
617
249
        _result_block->clear_names();
618
249
        RETURN_IF_ERROR(mysql_writer.write(_reusable->runtime_state(), *_result_block));
619
249
        RETURN_IF_ERROR(serialize_block(buffer->get_block(), _response));
620
249
        VLOG_DEBUG << "dump block " << _result_block->dump_data();
621
249
    } else {
622
11
        _response->set_empty_batch(true);
623
11
    }
624
260
    _profile_metrics.result_data_bytes = _result_block->bytes();
625
260
    _reusable->return_block(_result_block);
626
260
    return Status::OK();
627
260
}
628
629
} // namespace doris