Coverage Report

Created: 2026-06-02 13:57

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/defer_op.h"
58
#include "util/jsonb/serialize.h"
59
#include "util/lru_cache.h"
60
#include "util/simd/bits.h"
61
#include "util/thrift_util.h"
62
63
namespace doris {
64
65
class PointQueryResultBlockBuffer final : public MySQLResultBlockBuffer {
66
public:
67
12
    PointQueryResultBlockBuffer(RuntimeState* state) : MySQLResultBlockBuffer(state) {}
68
    ~PointQueryResultBlockBuffer() override = default;
69
12
    std::shared_ptr<TFetchDataResult> get_block() {
70
12
        std::lock_guard<std::mutex> l(_lock);
71
12
        DCHECK_EQ(_result_batch_queue.size(), 1);
72
12
        auto result = std::move(_result_batch_queue.front());
73
12
        _result_batch_queue.pop_front();
74
12
        return result;
75
12
    }
76
};
77
78
4.11k
Reusable::~Reusable() = default;
79
80
// get missing and include column ids
81
// input include_cids : the output expr slots columns unique ids
82
// missing_cids : the output expr columns that not in row columns cids
83
static void get_missing_and_include_cids(const TabletSchema& schema,
84
                                         const std::vector<SlotDescriptor*>& slots,
85
                                         int target_rs_column_id,
86
                                         std::unordered_set<int>& missing_cids,
87
4.06k
                                         std::unordered_set<int>& include_cids) {
88
4.06k
    missing_cids.clear();
89
4.06k
    include_cids.clear();
90
4.12k
    for (auto* slot : slots) {
91
4.12k
        missing_cids.insert(slot->col_unique_id());
92
4.12k
    }
93
    // insert delete sign column id
94
4.06k
    missing_cids.insert(schema.columns()[schema.delete_sign_idx()]->unique_id());
95
4.06k
    if (target_rs_column_id == -1) {
96
        // no row store columns
97
4.04k
        return;
98
4.04k
    }
99
15
    const TabletColumn& target_rs_column = schema.column_by_uid(target_rs_column_id);
100
15
    DCHECK(target_rs_column.is_row_store_column());
101
    // The full column group is considered a full match, thus no missing cids
102
15
    if (schema.row_columns_uids().empty()) {
103
9
        missing_cids.clear();
104
9
        return;
105
9
    }
106
11
    for (int cid : schema.row_columns_uids()) {
107
11
        missing_cids.erase(cid);
108
11
        include_cids.insert(cid);
109
11
    }
110
6
}
111
112
constexpr static int s_preallocted_blocks_num = 32;
113
114
static void extract_slot_ref(const VExprSPtr& expr, TupleDescriptor* tuple_desc,
115
4.12k
                             std::vector<SlotDescriptor*>& slots) {
116
4.12k
    const auto& children = expr->children();
117
4.12k
    for (const auto& i : children) {
118
0
        extract_slot_ref(i, tuple_desc, slots);
119
0
    }
120
121
4.12k
    auto node_type = expr->node_type();
122
4.12k
    if (node_type == TExprNodeType::SLOT_REF) {
123
4.12k
        int column_id = static_cast<const VSlotRef*>(expr.get())->column_id();
124
4.12k
        auto* slot_desc = tuple_desc->slots()[column_id];
125
4.12k
        slots.push_back(slot_desc);
126
4.12k
    }
127
4.12k
}
128
129
Status Reusable::init(const TDescriptorTable& t_desc_tbl, const std::vector<TExpr>& output_exprs,
130
                      const TQueryOptions& query_options, const TabletSchema& schema,
131
4.06k
                      size_t block_size) {
132
4.06k
    _runtime_state = RuntimeState::create_unique();
133
4.06k
    _runtime_state->set_query_options(query_options);
134
4.06k
    RETURN_IF_ERROR(DescriptorTbl::create(_runtime_state->obj_pool(), t_desc_tbl, &_desc_tbl));
135
4.06k
    _runtime_state->set_desc_tbl(_desc_tbl);
136
4.06k
    _block_pool.resize(block_size);
137
8.10k
    for (auto& i : _block_pool) {
138
8.10k
        i = Block::create_unique(tuple_desc()->slots(), 2);
139
        // Name is useless but cost space
140
8.10k
        i->clear_names();
141
8.10k
    }
142
143
4.06k
    RETURN_IF_ERROR(VExpr::create_expr_trees(output_exprs, _output_exprs_ctxs));
144
4.06k
    RowDescriptor row_desc(tuple_desc());
145
    // Prepare the exprs to run.
146
4.06k
    RETURN_IF_ERROR(VExpr::prepare(_output_exprs_ctxs, _runtime_state.get(), row_desc));
147
4.06k
    RETURN_IF_ERROR(VExpr::open(_output_exprs_ctxs, _runtime_state.get()));
148
4.06k
    _create_timestamp = butil::gettimeofday_ms();
149
4.06k
    _data_type_serdes = create_data_type_serdes(tuple_desc()->slots());
150
4.06k
    _col_default_values.resize(tuple_desc()->slots().size());
151
4.06k
    bool has_delete_sign = false;
152
93.1k
    for (int i = 0; i < tuple_desc()->slots().size(); ++i) {
153
89.1k
        auto* slot = tuple_desc()->slots()[i];
154
89.1k
        if (slot->col_name() == DELETE_SIGN) {
155
15
            has_delete_sign = true;
156
15
        }
157
89.1k
        _col_uid_to_idx[slot->col_unique_id()] = i;
158
89.1k
        _col_default_values[i] = slot->col_default_value();
159
89.1k
    }
160
161
    // Get the output slot descriptors
162
4.06k
    std::vector<SlotDescriptor*> output_slot_descs;
163
4.12k
    for (const auto& expr : _output_exprs_ctxs) {
164
4.12k
        extract_slot_ref(expr->root(), tuple_desc(), output_slot_descs);
165
4.12k
    }
166
167
    // get the delete sign idx in block
168
4.06k
    if (has_delete_sign) {
169
15
        _delete_sign_idx = _col_uid_to_idx[schema.columns()[schema.delete_sign_idx()]->unique_id()];
170
15
    }
171
172
4.06k
    if (schema.have_column(BeConsts::ROW_STORE_COL)) {
173
15
        const auto& column = *DORIS_TRY(schema.column(BeConsts::ROW_STORE_COL));
174
15
        _row_store_column_ids = column.unique_id();
175
15
    }
176
4.06k
    get_missing_and_include_cids(schema, output_slot_descs, _row_store_column_ids,
177
4.06k
                                 _missing_col_uids, _include_col_uids);
178
179
4.06k
    return Status::OK();
180
4.06k
}
181
182
15
std::unique_ptr<Block> Reusable::get_block() {
183
15
    std::lock_guard lock(_block_mutex);
184
15
    if (_block_pool.empty()) {
185
0
        auto block = Block::create_unique(tuple_desc()->slots(), 2);
186
        // Name is useless but cost space
187
0
        block->clear_names();
188
0
        return block;
189
0
    }
190
15
    auto block = std::move(_block_pool.back());
191
15
    CHECK(block != nullptr);
192
15
    _block_pool.pop_back();
193
15
    return block;
194
15
}
195
196
13
void Reusable::return_block(std::unique_ptr<Block>& block) {
197
13
    std::lock_guard lock(_block_mutex);
198
13
    if (block == nullptr) {
199
0
        return;
200
0
    }
201
13
    block->clear_column_data();
202
13
    _block_pool.push_back(std::move(block));
203
13
    if (_block_pool.size() > s_preallocted_blocks_num) {
204
0
        _block_pool.resize(s_preallocted_blocks_num);
205
0
    }
206
13
}
207
208
7
LookupConnectionCache* LookupConnectionCache::create_global_instance(size_t capacity) {
209
7
    DCHECK(ExecEnv::GetInstance()->get_lookup_connection_cache() == nullptr);
210
7
    auto* res = new LookupConnectionCache(capacity);
211
7
    return res;
212
7
}
213
214
RowCache::RowCache(int64_t capacity, int num_shards)
215
10
        : LRUCachePolicy(CachePolicy::CacheType::POINT_QUERY_ROW_CACHE, capacity,
216
10
                         LRUCacheType::SIZE, config::point_query_row_cache_stale_sweep_time_sec,
217
10
                         num_shards, /*element count capacity */ 0,
218
10
                         /*enable prune*/ true, /*is lru-k*/ true) {}
219
220
// Create global instance of this class
221
7
RowCache* RowCache::create_global_cache(int64_t capacity, uint32_t num_shards) {
222
7
    DCHECK(ExecEnv::GetInstance()->get_row_cache() == nullptr);
223
7
    auto* res = new RowCache(capacity, num_shards);
224
7
    return res;
225
7
}
226
227
0
RowCache* RowCache::instance() {
228
0
    return ExecEnv::GetInstance()->get_row_cache();
229
0
}
230
231
4
bool RowCache::lookup(const RowCacheKey& key, CacheHandle* handle) {
232
4
    const std::string& encoded_key = key.encode();
233
4
    auto* lru_handle = LRUCachePolicy::lookup(encoded_key);
234
4
    if (!lru_handle) {
235
        // cache miss
236
3
        return false;
237
3
    }
238
1
    *handle = CacheHandle(this, lru_handle);
239
1
    return true;
240
4
}
241
242
11
void RowCache::insert(const RowCacheKey& key, const Slice& value) {
243
11
    char* cache_value = static_cast<char*>(malloc(value.size));
244
11
    memcpy(cache_value, value.data, value.size);
245
11
    auto* row_cache_value = new RowCacheValue;
246
11
    row_cache_value->cache_value = cache_value;
247
11
    const std::string& encoded_key = key.encode();
248
11
    auto* handle = LRUCachePolicy::insert(encoded_key, row_cache_value, value.size, value.size,
249
11
                                          CachePriority::NORMAL);
250
    // handle will released
251
11
    auto tmp = CacheHandle {this, handle};
252
11
}
253
254
1
void RowCache::erase(const RowCacheKey& key) {
255
1
    const std::string& encoded_key = key.encode();
256
1
    LRUCachePolicy::erase(encoded_key);
257
1
}
258
259
2.09k
LookupConnectionCache::CacheValue::~CacheValue() {
260
2.09k
    SCOPED_SWITCH_THREAD_MEM_TRACKER_LIMITER(
261
2.09k
            ExecEnv::GetInstance()->point_query_executor_mem_tracker());
262
2.09k
    item.reset();
263
2.09k
}
264
265
15
PointQueryExecutor::~PointQueryExecutor() {
266
15
    SCOPED_SWITCH_THREAD_MEM_TRACKER_LIMITER(
267
15
            ExecEnv::GetInstance()->point_query_executor_mem_tracker());
268
15
    _tablet.reset();
269
15
    _reusable.reset();
270
15
    _result_block.reset();
271
15
    _row_read_ctxs.clear();
272
15
}
273
274
Status PointQueryExecutor::init(const PTabletKeyLookupRequest* request,
275
15
                                PTabletKeyLookupResponse* response) {
276
15
    SCOPED_TIMER(&_profile_metrics.init_ns);
277
15
    _response = response;
278
    // using cache
279
15
    __int128_t uuid =
280
15
            static_cast<__int128_t>(request->uuid().uuid_high()) << 64 | request->uuid().uuid_low();
281
15
    SCOPED_ATTACH_TASK(ExecEnv::GetInstance()->point_query_executor_mem_tracker());
282
15
    auto cache_handle = LookupConnectionCache::instance()->get(uuid);
283
15
    _binary_row_format = request->is_binary_row();
284
15
    _tablet = DORIS_TRY(ExecEnv::get_tablet(request->tablet_id()));
285
15
    if (cache_handle != nullptr) {
286
0
        _reusable = cache_handle;
287
0
        _profile_metrics.hit_lookup_cache = true;
288
15
    } else {
289
        // Lightweight request: FE may omit reusable query context and rely on uuid cache.
290
        // If cache miss and required parameters are absent, ask FE to resend a full request.
291
15
        if (uuid != 0 && (!request->has_desc_tbl() || request->desc_tbl().empty() ||
292
0
                          !request->has_output_expr() || request->output_expr().empty() ||
293
0
                          !request->has_query_options() || request->query_options().empty())) {
294
0
            if (VLOG_DEBUG_IS_ON) {
295
0
                VLOG_DEBUG << "lookup connection cache miss, ask FE to resend query context"
296
0
                           << ", tablet_id=" << request->tablet_id()
297
0
                           << ", uuid_high=" << request->uuid().uuid_high()
298
0
                           << ", uuid_low=" << request->uuid().uuid_low();
299
0
            }
300
0
            response->set_need_resend_query_context(true);
301
0
            return Status::OK();
302
0
        }
303
15
        if (uuid == 0 && (!request->has_desc_tbl() || request->desc_tbl().empty() ||
304
15
                          !request->has_output_expr() || request->output_expr().empty())) {
305
0
            return Status::InvalidArgument(
306
0
                    "tablet_fetch_data requires desc_tbl/output_expr when uuid is not set");
307
0
        }
308
        // init handle
309
15
        auto reusable_ptr = std::make_shared<Reusable>();
310
15
        TDescriptorTable t_desc_tbl;
311
15
        TExprList t_output_exprs;
312
15
        auto len = cast_set<uint32_t>(request->desc_tbl().size());
313
15
        RETURN_IF_ERROR(
314
15
                deserialize_thrift_msg(reinterpret_cast<const uint8_t*>(request->desc_tbl().data()),
315
15
                                       &len, false, &t_desc_tbl));
316
15
        len = cast_set<uint32_t>(request->output_expr().size());
317
15
        RETURN_IF_ERROR(deserialize_thrift_msg(
318
15
                reinterpret_cast<const uint8_t*>(request->output_expr().data()), &len, false,
319
15
                &t_output_exprs));
320
15
        _reusable = reusable_ptr;
321
15
        TQueryOptions t_query_options;
322
15
        len = cast_set<uint32_t>(request->query_options().size());
323
15
        if (request->has_query_options()) {
324
15
            RETURN_IF_ERROR(deserialize_thrift_msg(
325
15
                    reinterpret_cast<const uint8_t*>(request->query_options().data()), &len, false,
326
15
                    &t_query_options));
327
15
        }
328
15
        if (uuid != 0) {
329
            // could be reused by requests after, pre allocte more blocks
330
0
            RETURN_IF_ERROR(reusable_ptr->init(t_desc_tbl, t_output_exprs.exprs, t_query_options,
331
0
                                               *_tablet->tablet_schema(),
332
0
                                               s_preallocted_blocks_num));
333
0
            LookupConnectionCache::instance()->add(uuid, reusable_ptr);
334
15
        } else {
335
15
            RETURN_IF_ERROR(reusable_ptr->init(t_desc_tbl, t_output_exprs.exprs, t_query_options,
336
15
                                               *_tablet->tablet_schema(), 1));
337
15
        }
338
15
    }
339
    // Set timezone from request for functions like from_unixtime()
340
15
    if (request->has_time_zone() && !request->time_zone().empty()) {
341
15
        _reusable->runtime_state()->set_timezone(request->time_zone());
342
15
    }
343
15
    if (request->has_version() && request->version() >= 0) {
344
15
        _version = request->version();
345
15
    }
346
15
    RETURN_IF_ERROR(_init_keys(request));
347
15
    _result_block = _reusable->get_block();
348
15
    CHECK(_result_block != nullptr);
349
350
15
    return Status::OK();
351
15
}
352
353
15
Status PointQueryExecutor::lookup_up() {
354
15
    SCOPED_ATTACH_TASK(ExecEnv::GetInstance()->point_query_executor_mem_tracker());
355
15
    RETURN_IF_ERROR(_lookup_row_key());
356
15
    RETURN_IF_ERROR(_lookup_row_data());
357
13
    RETURN_IF_ERROR(_output_data());
358
13
    return Status::OK();
359
13
}
360
361
13
void PointQueryExecutor::print_profile() {
362
13
    auto init_us = _profile_metrics.init_ns.value() / 1000;
363
13
    auto init_key_us = _profile_metrics.init_key_ns.value() / 1000;
364
13
    auto lookup_key_us = _profile_metrics.lookup_key_ns.value() / 1000;
365
13
    auto lookup_data_us = _profile_metrics.lookup_data_ns.value() / 1000;
366
13
    auto output_data_us = _profile_metrics.output_data_ns.value() / 1000;
367
13
    auto load_segments_key_us = _profile_metrics.load_segment_key_stage_ns.value() / 1000;
368
13
    auto load_segments_data_us = _profile_metrics.load_segment_data_stage_ns.value() / 1000;
369
13
    auto total_us = init_us + lookup_key_us + lookup_data_us + output_data_us;
370
13
    auto read_stats = _profile_metrics.read_stats;
371
13
    const std::string stats_str = fmt::format(
372
13
            "[lookup profile:{}us] init:{}us, init_key:{}us,"
373
13
            " lookup_key:{}us, load_segments_key:{}us, lookup_data:{}us, load_segments_data:{}us,"
374
13
            " output_data:{}us, "
375
13
            "hit_lookup_cache:{}"
376
13
            ", is_binary_row:{}, output_columns:{}, total_keys:{}, row_cache_hits:{}"
377
13
            ", hit_cached_pages:{}, total_pages_read:{}, compressed_bytes_read:{}, "
378
13
            "io_latency:{}ns, "
379
13
            "uncompressed_bytes_read:{}, result_data_bytes:{}, row_hits:{}"
380
13
            ", rs_column_uid:{}, bytes_read_from_local:{}, bytes_read_from_remote:{}, "
381
13
            "local_io_timer:{}, remote_io_timer:{}, local_write_timer:{}",
382
13
            total_us, init_us, init_key_us, lookup_key_us, load_segments_key_us, lookup_data_us,
383
13
            load_segments_data_us, output_data_us, _profile_metrics.hit_lookup_cache,
384
13
            _binary_row_format, _reusable->output_exprs().size(), _row_read_ctxs.size(),
385
13
            _profile_metrics.row_cache_hits, read_stats.cached_pages_num,
386
13
            read_stats.total_pages_num, read_stats.compressed_bytes_read, read_stats.io_ns,
387
13
            read_stats.uncompressed_bytes_read, _profile_metrics.result_data_bytes, _row_hits,
388
13
            _reusable->rs_column_uid(),
389
13
            _profile_metrics.read_stats.file_cache_stats.bytes_read_from_local,
390
13
            _profile_metrics.read_stats.file_cache_stats.bytes_read_from_remote,
391
13
            _profile_metrics.read_stats.file_cache_stats.local_io_timer,
392
13
            _profile_metrics.read_stats.file_cache_stats.remote_io_timer,
393
13
            _profile_metrics.read_stats.file_cache_stats.write_cache_io_timer);
394
395
13
    constexpr static int kSlowThreholdUs = 50 * 1000; // 50ms
396
13
    if (total_us > kSlowThreholdUs) {
397
2
        LOG(WARNING) << "slow query, " << stats_str;
398
11
    } else if (VLOG_DEBUG_IS_ON) {
399
0
        VLOG_DEBUG << stats_str;
400
11
    } else {
401
11
        LOG_EVERY_N(INFO, 1000) << stats_str;
402
11
    }
403
13
}
404
405
15
Status PointQueryExecutor::_init_keys(const PTabletKeyLookupRequest* request) {
406
15
    SCOPED_TIMER(&_profile_metrics.init_key_ns);
407
15
    const auto& schema = _tablet->tablet_schema();
408
    // Point query is only supported on merge-on-write unique key tables.
409
15
    DCHECK(schema->keys_type() == UNIQUE_KEYS && _tablet->enable_unique_key_merge_on_write());
410
15
    if (schema->keys_type() != UNIQUE_KEYS || !_tablet->enable_unique_key_merge_on_write()) {
411
0
        return Status::InvalidArgument(
412
0
                "Point query is only supported on merge-on-write unique key tables, "
413
0
                "tablet_id={}",
414
0
                _tablet->tablet_id());
415
0
    }
416
    // 1. get primary key from conditions
417
15
    _row_read_ctxs.resize(request->key_tuples().size());
418
    // get row cursor and encode keys
419
30
    for (int i = 0; i < request->key_tuples().size(); ++i) {
420
15
        const KeyTuple& key_tuple = request->key_tuples(i);
421
15
        if (UNLIKELY(cast_set<size_t>(key_tuple.key_column_literals_size()) !=
422
15
                     schema->num_key_columns())) {
423
0
            return Status::InvalidArgument(
424
0
                    "Key column count mismatch. expected={}, actual={}, tablet_id={}",
425
0
                    schema->num_key_columns(), key_tuple.key_column_literals_size(),
426
0
                    _tablet->tablet_id());
427
0
        }
428
15
        RowCursor cursor;
429
15
        std::vector<Field> key_fields;
430
15
        key_fields.reserve(key_tuple.key_column_literals_size());
431
30
        for (int j = 0; j < key_tuple.key_column_literals_size(); ++j) {
432
15
            const auto& literal_bytes = key_tuple.key_column_literals(j);
433
15
            TExprNode expr_node;
434
15
            auto len = cast_set<uint32_t>(literal_bytes.size());
435
15
            RETURN_IF_ERROR(
436
15
                    deserialize_thrift_msg(reinterpret_cast<const uint8_t*>(literal_bytes.data()),
437
15
                                           &len, false, &expr_node));
438
15
            const auto& col = schema->column(j);
439
15
            auto data_type = DataTypeFactory::instance().create_data_type(
440
15
                    col.type(), col.precision(), col.frac(), col.length());
441
15
            key_fields.push_back(data_type->get_field(expr_node));
442
15
        }
443
15
        RETURN_IF_ERROR(cursor.init_scan_key(_tablet->tablet_schema(), std::move(key_fields)));
444
15
        cursor.encode_key_with_padding<true>(&_row_read_ctxs[i]._primary_key,
445
15
                                             _tablet->tablet_schema()->num_key_columns(), true);
446
15
    }
447
15
    return Status::OK();
448
15
}
449
450
15
Status PointQueryExecutor::_lookup_row_key() {
451
15
    SCOPED_TIMER(&_profile_metrics.lookup_key_ns);
452
    // 2. lookup row location
453
15
    Status st;
454
15
    if (_version >= 0) {
455
15
        CHECK(config::is_cloud_mode()) << "Only cloud mode support snapshot read at present";
456
15
        SyncOptions options;
457
15
        options.query_version = _version;
458
15
        RETURN_IF_ERROR(std::dynamic_pointer_cast<CloudTablet>(_tablet)->sync_rowsets(options));
459
15
    }
460
15
    std::vector<RowsetSharedPtr> specified_rowsets;
461
15
    {
462
15
        std::shared_lock rlock(_tablet->get_header_lock());
463
15
        specified_rowsets = _tablet->get_rowset_by_ids(nullptr);
464
15
    }
465
15
    std::vector<std::unique_ptr<SegmentCacheHandle>> segment_caches(specified_rowsets.size());
466
30
    for (size_t i = 0; i < _row_read_ctxs.size(); ++i) {
467
15
        RowLocation location;
468
15
        if (!config::disable_storage_row_cache) {
469
0
            RowCache::CacheHandle cache_handle;
470
0
            auto hit_cache = RowCache::instance()->lookup(
471
0
                    {_tablet->tablet_id(), _row_read_ctxs[i]._primary_key}, &cache_handle);
472
0
            if (hit_cache) {
473
0
                _row_read_ctxs[i]._cached_row_data = std::move(cache_handle);
474
0
                ++_profile_metrics.row_cache_hits;
475
0
                continue;
476
0
            }
477
0
        }
478
        // Get rowlocation and rowset, ctx._rowset_ptr will acquire wrap this ptr
479
15
        auto rowset_ptr = std::make_unique<RowsetSharedPtr>();
480
15
        st = (_tablet->lookup_row_key(_row_read_ctxs[i]._primary_key, nullptr, false,
481
15
                                      specified_rowsets, &location, INT32_MAX /*rethink?*/,
482
15
                                      segment_caches, rowset_ptr.get(), false, nullptr,
483
15
                                      &_profile_metrics.read_stats));
484
15
        if (st.is<ErrorCode::KEY_NOT_FOUND>()) {
485
0
            continue;
486
0
        }
487
15
        RETURN_IF_ERROR(st);
488
15
        _row_read_ctxs[i]._row_location = location;
489
        // acquire and wrap this rowset
490
15
        (*rowset_ptr)->acquire();
491
15
        VLOG_DEBUG << "aquire rowset " << (*rowset_ptr)->rowset_id();
492
15
        _row_read_ctxs[i]._rowset_ptr = std::unique_ptr<RowsetSharedPtr, decltype(&release_rowset)>(
493
15
                rowset_ptr.release(), &release_rowset);
494
15
        _row_hits++;
495
15
    }
496
15
    return Status::OK();
497
15
}
498
499
15
Status PointQueryExecutor::_lookup_row_data() {
500
    // 3. get values
501
15
    SCOPED_TIMER(&_profile_metrics.lookup_data_ns);
502
15
    {
503
15
        auto result_columns_guard = _result_block->mutate_columns_scoped();
504
15
        MutableColumns& result_columns = result_columns_guard.mutable_columns();
505
28
        for (size_t i = 0; i < _row_read_ctxs.size(); ++i) {
506
15
            if (_row_read_ctxs[i]._cached_row_data.valid()) {
507
0
                RETURN_IF_ERROR(JsonbSerializeUtil::jsonb_to_columns(
508
0
                        _reusable->get_data_type_serdes(),
509
0
                        _row_read_ctxs[i]._cached_row_data.data().data,
510
0
                        _row_read_ctxs[i]._cached_row_data.data().size,
511
0
                        _reusable->get_col_uid_to_idx(), result_columns,
512
0
                        _reusable->get_col_default_values(), _reusable->include_col_uids()));
513
0
                continue;
514
0
            }
515
15
            if (!_row_read_ctxs[i]._row_location.has_value()) {
516
0
                continue;
517
0
            }
518
15
            std::string value;
519
            // fill block by row store
520
15
            if (_reusable->rs_column_uid() != -1) {
521
15
                bool use_row_cache = !config::disable_storage_row_cache;
522
15
                RETURN_IF_ERROR(_tablet->lookup_row_data(
523
15
                        _row_read_ctxs[i]._primary_key, _row_read_ctxs[i]._row_location.value(),
524
15
                        *(_row_read_ctxs[i]._rowset_ptr), _profile_metrics.read_stats, value,
525
15
                        use_row_cache));
526
                // serialize value to block, currently only jsonb row format
527
15
                RETURN_IF_ERROR(JsonbSerializeUtil::jsonb_to_columns(
528
15
                        _reusable->get_data_type_serdes(), value.data(), value.size(),
529
15
                        _reusable->get_col_uid_to_idx(), result_columns,
530
15
                        _reusable->get_col_default_values(), _reusable->include_col_uids()));
531
15
            }
532
15
            if (!_reusable->missing_col_uids().empty()) {
533
6
                if (!_reusable->runtime_state()->enable_short_circuit_query_access_column_store()) {
534
2
                    std::string missing_columns;
535
20
                    for (int cid : _reusable->missing_col_uids()) {
536
20
                        missing_columns +=
537
20
                                _tablet->tablet_schema()->column_by_uid(cid).name() + ",";
538
20
                    }
539
2
                    return Status::InternalError(
540
2
                            "Not support column store, set store_row_column=true or "
541
2
                            "row_store_columns in table properties, missing columns: " +
542
2
                            missing_columns + " should be added to row store");
543
2
                }
544
                // fill missing columns by column store
545
4
                RowLocation row_loc = _row_read_ctxs[i]._row_location.value();
546
4
                BetaRowsetSharedPtr rowset = std::static_pointer_cast<BetaRowset>(
547
4
                        _tablet->get_rowset(row_loc.rowset_id));
548
4
                SegmentCacheHandle segment_cache;
549
4
                {
550
4
                    SCOPED_TIMER(&_profile_metrics.load_segment_data_stage_ns);
551
4
                    RETURN_IF_ERROR(
552
4
                            SegmentLoader::instance()->load_segments(rowset, &segment_cache, true));
553
4
                }
554
                // find segment
555
4
                auto it = std::find_if(segment_cache.get_segments().cbegin(),
556
4
                                       segment_cache.get_segments().cend(),
557
4
                                       [&](const segment_v2::SegmentSharedPtr& seg) {
558
4
                                           return seg->id() == row_loc.segment_id;
559
4
                                       });
560
4
                const auto& segment = *it;
561
4
                for (int cid : _reusable->missing_col_uids()) {
562
4
                    int pos = _reusable->get_col_uid_to_idx().at(cid);
563
4
                    std::vector<segment_v2::rowid_t> row_ids {
564
4
                            static_cast<segment_v2::rowid_t>(row_loc.row_id)};
565
4
                    auto& column = result_columns[pos];
566
4
                    std::unique_ptr<ColumnIterator> iter;
567
4
                    SlotDescriptor* slot = _reusable->tuple_desc()->slots()[pos];
568
4
                    StorageReadOptions storage_read_options;
569
4
                    storage_read_options.stats = &_read_stats;
570
4
                    storage_read_options.io_ctx.reader_type = ReaderType::READER_QUERY;
571
4
                    RETURN_IF_ERROR(segment->seek_and_read_by_rowid(*_tablet->tablet_schema(), slot,
572
4
                                                                    row_ids, column,
573
4
                                                                    storage_read_options, iter));
574
4
                }
575
4
            }
576
15
        }
577
13
        if (result_columns.size() > _reusable->include_col_uids().size()) {
578
            // Padding rows for some columns that no need to output to mysql client
579
            // eg. SELECT k1,v1,v2 FROM TABLE WHERE k1 = 1, k1 is not in output slots, tuple as bellow
580
            // TupleDescriptor{id=1, tbl=table_with_column_group}
581
            // SlotDescriptor{id=8, col=v1, colUniqueId=1 ...}
582
            // SlotDescriptor{id=9, col=v2, colUniqueId=2 ...}
583
            // thus missing in include_col_uids and missing_col_uids
584
74
            for (auto& column : result_columns) {
585
74
                int padding_rows = _row_hits - cast_set<int>(column->size());
586
74
                if (padding_rows > 0) {
587
1
                    column->insert_many_defaults(padding_rows);
588
1
                }
589
74
            }
590
13
        }
591
13
    }
592
    // filter rows by delete sign
593
13
    if (_row_hits > 0 && _reusable->delete_sign_idx() != -1) {
594
13
        size_t filtered = 0;
595
13
        size_t total = 0;
596
13
        {
597
            // clear_column_data will check reference of ColumnPtr, so we need to release
598
            // reference before clear_column_data
599
13
            ColumnPtr delete_filter_columns =
600
13
                    _result_block->get_columns()[_reusable->delete_sign_idx()];
601
13
            const auto& filter =
602
13
                    assert_cast<const ColumnInt8*>(delete_filter_columns.get())->get_data();
603
13
            filtered = filter.size() - simd::count_zero_num((int8_t*)filter.data(), filter.size());
604
13
            total = filter.size();
605
13
        }
606
607
13
        if (filtered == total) {
608
1
            _result_block->clear_column_data();
609
12
        } else if (filtered > 0) {
610
0
            return Status::NotSupported("Not implemented since only single row at present");
611
0
        }
612
13
    }
613
13
    return Status::OK();
614
13
}
615
616
12
Status serialize_block(std::shared_ptr<TFetchDataResult> res, PTabletKeyLookupResponse* response) {
617
12
    uint8_t* buf = nullptr;
618
12
    uint32_t len = 0;
619
12
    ThriftSerializer ser(false, 4096);
620
12
    RETURN_IF_ERROR(ser.serialize(&(res->result_batch), &len, &buf));
621
12
    response->set_row_batch(std::string((const char*)buf, len));
622
12
    return Status::OK();
623
12
}
624
625
13
Status PointQueryExecutor::_output_data() {
626
    // 4. exprs exec and serialize to mysql row batches
627
13
    SCOPED_TIMER(&_profile_metrics.output_data_ns);
628
13
    if (_result_block->rows()) {
629
12
        RuntimeState state;
630
12
        auto buffer = std::make_shared<PointQueryResultBlockBuffer>(&state);
631
        // TODO reuse mysql_writer
632
12
        VMysqlResultWriter mysql_writer(buffer, _reusable->output_exprs(), nullptr,
633
12
                                        _binary_row_format);
634
12
        RETURN_IF_ERROR(mysql_writer.init(_reusable->runtime_state()));
635
12
        _result_block->clear_names();
636
12
        RETURN_IF_ERROR(mysql_writer.write(_reusable->runtime_state(), *_result_block));
637
12
        RETURN_IF_ERROR(serialize_block(buffer->get_block(), _response));
638
12
        VLOG_DEBUG << "dump block " << _result_block->dump_data();
639
12
    } else {
640
1
        _response->set_empty_batch(true);
641
1
    }
642
13
    _profile_metrics.result_data_bytes = _result_block->bytes();
643
13
    _reusable->return_block(_result_block);
644
13
    return Status::OK();
645
13
}
646
647
} // namespace doris