Coverage Report

Created: 2026-03-19 07:34

next uncovered line (L), next uncovered region (R), next uncovered branch (B)
be/src/cloud/cloud_tablet_mgr.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 "cloud/cloud_tablet_mgr.h"
19
20
#include <bthread/countdown_event.h>
21
22
#include <chrono>
23
24
#include "cloud/cloud_cluster_info.h"
25
#include "cloud/cloud_meta_mgr.h"
26
#include "cloud/cloud_storage_engine.h"
27
#include "cloud/cloud_tablet.h"
28
#include "cloud/config.h"
29
#include "common/status.h"
30
#include "cpp/sync_point.h"
31
#include "runtime/memory/cache_policy.h"
32
#include "util/debug_points.h"
33
#include "util/lru_cache.h"
34
#include "util/stack_util.h"
35
36
namespace doris {
37
uint64_t g_tablet_report_inactive_duration_ms = 0;
38
bvar::Adder<uint64_t> g_base_compaction_not_frozen_tablet_num(
39
        "base_compaction_not_frozen_tablet_num");
40
bvar::Adder<uint64_t> g_cumu_compaction_not_frozen_tablet_num(
41
        "cumu_compaction_not_frozen_tablet_num");
42
namespace {
43
44
// port from
45
// https://github.com/golang/groupcache/blob/master/singleflight/singleflight.go
46
template <typename Key, typename Val>
47
class SingleFlight {
48
public:
49
9
    SingleFlight() = default;
50
51
    SingleFlight(const SingleFlight&) = delete;
52
    void operator=(const SingleFlight&) = delete;
53
54
    using Loader = std::function<Val(const Key&)>;
55
56
    // Do executes and returns the results of the given function, making
57
    // sure that only one execution is in-flight for a given key at a
58
    // time. If a duplicate comes in, the duplicate caller waits for the
59
    // original to complete and receives the same results.
60
132k
    Val load(const Key& key, Loader loader) {
61
132k
        std::unique_lock lock(_call_map_mtx);
62
63
132k
        auto it = _call_map.find(key);
64
132k
        if (it != _call_map.end()) {
65
171
            auto call = it->second;
66
171
            lock.unlock();
67
171
            if (int ec = call->event.wait(); ec != 0) {
68
0
                throw std::system_error(std::error_code(ec, std::system_category()),
69
0
                                        "CountdownEvent wait failed");
70
0
            }
71
171
            return call->val;
72
171
        }
73
132k
        auto call = std::make_shared<Call>();
74
132k
        _call_map.emplace(key, call);
75
132k
        lock.unlock();
76
77
132k
        call->val = loader(key);
78
132k
        call->event.signal();
79
80
132k
        lock.lock();
81
132k
        _call_map.erase(key);
82
132k
        lock.unlock();
83
84
132k
        return call->val;
85
132k
    }
86
87
private:
88
    // `Call` is an in-flight or completed `load` call
89
    struct Call {
90
        bthread::CountdownEvent event;
91
        Val val;
92
    };
93
94
    std::mutex _call_map_mtx;
95
    std::unordered_map<Key, std::shared_ptr<Call>> _call_map;
96
};
97
98
// tablet_id -> load tablet function
99
SingleFlight<int64_t, Result<std::shared_ptr<CloudTablet>>> s_singleflight_load_tablet;
100
101
} // namespace
102
103
// tablet_id -> cached tablet
104
// This map owns all cached tablets. The lifetime of tablet can be longer than the LRU handle.
105
// It's also used for scenarios where users want to access the tablet by `tablet_id` without changing the LRU order.
106
// TODO(plat1ko): multi shard to increase concurrency
107
class CloudTabletMgr::TabletMap {
108
public:
109
130k
    void put(std::shared_ptr<CloudTablet> tablet) {
110
130k
        std::lock_guard lock(_mtx);
111
130k
        _map[tablet->tablet_id()] = std::move(tablet);
112
130k
    }
113
114
32.4k
    void erase(CloudTablet* tablet) {
115
32.4k
        std::lock_guard lock(_mtx);
116
32.4k
        auto it = _map.find(tablet->tablet_id());
117
        // According to the implementation of `LRUCache`, `deleter` may be called after a tablet
118
        // with same tablet id insert into cache and `TabletMap`. So we MUST check if the tablet
119
        // instance to be erased is the same one in the map.
120
32.7k
        if (it != _map.end() && it->second.get() == tablet) {
121
32.7k
            _map.erase(it);
122
32.7k
        }
123
32.4k
    }
124
125
0
    std::shared_ptr<CloudTablet> get(int64_t tablet_id) {
126
0
        std::lock_guard lock(_mtx);
127
0
        if (auto it = _map.find(tablet_id); it != _map.end()) {
128
0
            return it->second;
129
0
        }
130
0
        return nullptr;
131
0
    }
132
133
19.8k
    size_t size() { return _map.size(); }
134
135
19.8k
    void traverse(std::function<void(const std::shared_ptr<CloudTablet>&)> visitor) {
136
19.8k
        std::lock_guard lock(_mtx);
137
689M
        for (auto& [_, tablet] : _map) {
138
689M
            visitor(tablet);
139
689M
        }
140
19.8k
    }
141
142
private:
143
    std::mutex _mtx;
144
    std::unordered_map<int64_t, std::shared_ptr<CloudTablet>> _map;
145
};
146
147
// TODO(plat1ko): Prune cache
148
CloudTabletMgr::CloudTabletMgr(CloudStorageEngine& engine)
149
157
        : _engine(engine),
150
157
          _tablet_map(std::make_unique<TabletMap>()),
151
157
          _cache(std::make_unique<LRUCachePolicy>(
152
157
                  CachePolicy::CacheType::CLOUD_TABLET_CACHE, config::tablet_cache_capacity,
153
157
                  LRUCacheType::NUMBER, /*sweep time*/ 0, config::tablet_cache_shards,
154
157
                  /*element_count_capacity*/ 0, /*enable_prune*/ false,
155
157
                  /*is_lru_k*/ false)) {}
156
157
156
CloudTabletMgr::~CloudTabletMgr() = default;
158
159
3.56M
void set_tablet_access_time_ms(CloudTablet* tablet) {
160
3.56M
    using namespace std::chrono;
161
3.56M
    int64_t now = duration_cast<milliseconds>(system_clock::now().time_since_epoch()).count();
162
3.56M
    tablet->last_access_time_ms = now;
163
3.56M
}
164
165
Result<std::shared_ptr<CloudTablet>> CloudTabletMgr::get_tablet(int64_t tablet_id, bool warmup_data,
166
                                                                bool sync_delete_bitmap,
167
                                                                SyncRowsetStats* sync_stats,
168
                                                                bool force_use_only_cached,
169
1.77M
                                                                bool cache_on_miss) {
170
1.77M
    DBUG_EXECUTE_IF("CloudTabletMgr::get_tablet.block", DBUG_BLOCK);
171
    // LRU value type. `Value`'s lifetime MUST NOT be longer than `CloudTabletMgr`
172
1.77M
    class Value : public LRUCacheValueBase {
173
1.77M
    public:
174
1.77M
        Value(const std::shared_ptr<CloudTablet>& tablet, TabletMap& tablet_map)
175
1.77M
                : tablet(tablet), tablet_map(tablet_map) {}
176
1.77M
        ~Value() override { tablet_map.erase(tablet.get()); }
177
178
        // FIXME(plat1ko): The ownership of tablet seems to belong to 'TabletMap', while `Value`
179
        // only requires a reference.
180
1.77M
        std::shared_ptr<CloudTablet> tablet;
181
1.77M
        TabletMap& tablet_map;
182
1.77M
    };
183
184
18.4E
    VLOG_DEBUG << "get_tablet tablet_id=" << tablet_id << " stack: " << get_stack_trace();
185
186
1.77M
    auto tablet_id_str = std::to_string(tablet_id);
187
1.77M
    CacheKey key(tablet_id_str);
188
1.77M
    auto* handle = _cache->lookup(key);
189
190
1.77M
    if (handle == nullptr) {
191
132k
        if (force_use_only_cached) {
192
1
            LOG(INFO) << "tablet=" << tablet_id
193
1
                      << "does not exists in local tablet cache, because param "
194
1
                         "force_use_only_cached=true, "
195
1
                         "treat it as an error";
196
1
            return ResultError(Status::InternalError(
197
1
                    "tablet={} does not exists in local tablet cache, because param "
198
1
                    "force_use_only_cached=true, "
199
1
                    "treat it as an error",
200
1
                    tablet_id));
201
1
        }
202
132k
        TEST_SYNC_POINT("CloudTabletMgr::get_tablet.not_found_in_cache");
203
132k
        if (sync_stats) {
204
19.6k
            ++sync_stats->tablet_meta_cache_miss;
205
19.6k
        }
206
        // Insert into cache and tablet_map inside SingleFlight lambda to ensure
207
        // only the leader caller does this. Moving these outside the lambda causes
208
        // a race condition: when multiple concurrent callers share the same CloudTablet*
209
        // from SingleFlight, each creates a competing LRU cache entry. Delayed Value
210
        // destructors then erase the tablet_map entry (the raw pointer safety check
211
        // passes since all callers share the same pointer), and the tablet permanently
212
        // disappears from tablet_map. Subsequent get_tablet() calls hit the LRU cache
213
        // directly (cache hit path) which never re-inserts into tablet_map, making the
214
        // tablet invisible to the compaction scheduler.
215
132k
        auto load_tablet = [this, &key, warmup_data, sync_delete_bitmap, sync_stats, cache_on_miss](
216
132k
                                   int64_t tablet_id) -> Result<std::shared_ptr<CloudTablet>> {
217
132k
            TabletMetaSharedPtr tablet_meta;
218
132k
            auto start = std::chrono::steady_clock::now();
219
132k
            auto st = _engine.meta_mgr().get_tablet_meta(tablet_id, &tablet_meta);
220
132k
            auto end = std::chrono::steady_clock::now();
221
132k
            if (sync_stats) {
222
19.6k
                sync_stats->get_remote_tablet_meta_rpc_ns +=
223
19.6k
                        std::chrono::duration_cast<std::chrono::nanoseconds>(end - start).count();
224
19.6k
            }
225
132k
            if (!st.ok()) {
226
110
                LOG(WARNING) << "failed to tablet " << tablet_id << ": " << st;
227
110
                return ResultError(st);
228
110
            }
229
230
132k
            auto tablet = std::make_shared<CloudTablet>(_engine, std::move(tablet_meta));
231
            // MUST sync stats to let compaction scheduler work correctly
232
132k
            SyncOptions options;
233
132k
            options.warmup_delta_data = warmup_data;
234
132k
            options.sync_delete_bitmap = sync_delete_bitmap;
235
132k
            st = _engine.meta_mgr().sync_tablet_rowsets(tablet.get(), options, sync_stats);
236
132k
            if (!st.ok()) {
237
0
                LOG(WARNING) << "failed to sync tablet " << tablet_id << ": " << st;
238
0
                return ResultError(st);
239
0
            }
240
241
132k
            if (!cache_on_miss) {
242
0
                set_tablet_access_time_ms(tablet.get());
243
0
                return tablet;
244
0
            }
245
246
132k
            auto value = std::make_unique<Value>(tablet, *_tablet_map);
247
132k
            auto* insert_handle = _cache->insert(key, value.release(), 1, sizeof(CloudTablet),
248
132k
                                                 CachePriority::NORMAL);
249
132k
            auto ret = std::shared_ptr<CloudTablet>(tablet.get(),
250
132k
                                                    [this, insert_handle](CloudTablet* tablet_ptr) {
251
132k
                                                        set_tablet_access_time_ms(tablet_ptr);
252
132k
                                                        _cache->release(insert_handle);
253
132k
                                                    });
254
132k
            _tablet_map->put(std::move(tablet));
255
132k
            return ret;
256
132k
        };
257
258
132k
        auto load_result = s_singleflight_load_tablet.load(tablet_id, std::move(load_tablet));
259
132k
        if (!load_result.has_value()) {
260
110
            return ResultError(Status::InternalError("failed to get tablet {}, msg={}", tablet_id,
261
110
                                                     load_result.error()));
262
110
        }
263
132k
        auto tablet = load_result.value();
264
132k
        set_tablet_access_time_ms(tablet.get());
265
132k
        return tablet;
266
132k
    }
267
1.63M
    if (sync_stats) {
268
1.24M
        ++sync_stats->tablet_meta_cache_hit;
269
1.24M
    }
270
1.63M
    CloudTablet* tablet_raw_ptr = reinterpret_cast<Value*>(_cache->value(handle))->tablet.get();
271
1.63M
    set_tablet_access_time_ms(tablet_raw_ptr);
272
1.65M
    auto tablet = std::shared_ptr<CloudTablet>(tablet_raw_ptr, [this, handle](CloudTablet* tablet) {
273
1.65M
        set_tablet_access_time_ms(tablet);
274
1.65M
        _cache->release(handle);
275
1.65M
    });
276
1.63M
    return tablet;
277
1.77M
}
278
279
0
bool CloudTabletMgr::peek_tablet_meta(int64_t tablet_id, TabletMetaSharedPtr* tablet_meta) {
280
0
    if (tablet_meta == nullptr) {
281
0
        return false;
282
0
    }
283
0
    auto tablet = _tablet_map->get(tablet_id);
284
0
    if (!tablet) {
285
0
        return false;
286
0
    }
287
0
    *tablet_meta = tablet->tablet_meta();
288
0
    return true;
289
0
}
290
291
481
void CloudTabletMgr::erase_tablet(int64_t tablet_id) {
292
481
    auto tablet_id_str = std::to_string(tablet_id);
293
481
    CacheKey key(tablet_id_str.data(), tablet_id_str.size());
294
481
    _cache->erase(key);
295
481
}
296
297
13
void CloudTabletMgr::vacuum_stale_rowsets(const CountDownLatch& stop_latch) {
298
13
    LOG_INFO("begin to vacuum stale rowsets");
299
13
    std::vector<std::shared_ptr<CloudTablet>> tablets_to_vacuum;
300
13
    tablets_to_vacuum.reserve(_tablet_map->size());
301
716k
    _tablet_map->traverse([&tablets_to_vacuum](auto&& t) {
302
716k
        if (t->has_stale_rowsets()) {
303
4.87k
            tablets_to_vacuum.push_back(t);
304
4.87k
        }
305
716k
    });
306
13
    int num_vacuumed = 0;
307
4.87k
    for (auto& t : tablets_to_vacuum) {
308
4.87k
        if (stop_latch.count() <= 0) {
309
0
            break;
310
0
        }
311
312
4.87k
        num_vacuumed += t->delete_expired_stale_rowsets();
313
4.87k
    }
314
13
    LOG_INFO("finish vacuum stale rowsets")
315
13
            .tag("num_vacuumed", num_vacuumed)
316
13
            .tag("num_tablets", tablets_to_vacuum.size());
317
318
13
    {
319
13
        LOG_INFO("begin to remove unused rowsets");
320
13
        std::vector<std::shared_ptr<CloudTablet>> tablets_to_remove_unused_rowsets;
321
13
        tablets_to_remove_unused_rowsets.reserve(_tablet_map->size());
322
717k
        _tablet_map->traverse([&tablets_to_remove_unused_rowsets](auto&& t) {
323
717k
            if (t->need_remove_unused_rowsets()) {
324
2.46k
                tablets_to_remove_unused_rowsets.push_back(t);
325
2.46k
            }
326
717k
        });
327
2.46k
        for (auto& t : tablets_to_remove_unused_rowsets) {
328
2.46k
            t->remove_unused_rowsets();
329
2.46k
        }
330
13
        LOG_INFO("finish remove unused rowsets")
331
13
                .tag("num_tablets", tablets_to_remove_unused_rowsets.size());
332
13
        if (config::enable_check_agg_and_remove_pre_rowsets_delete_bitmap) {
333
0
            int64_t max_useless_rowset_count = 0;
334
0
            int64_t tablet_id_with_max_useless_rowset_count = 0;
335
0
            int64_t max_useless_rowset_version_count = 0;
336
0
            int64_t tablet_id_with_max_useless_rowset_version_count = 0;
337
0
            OlapStopWatch watch;
338
0
            _tablet_map->traverse([&](auto&& tablet) {
339
0
                int64_t useless_rowset_count = 0;
340
0
                int64_t useless_rowset_version_count = 0;
341
0
                tablet->check_agg_delete_bitmap_for_stale_rowsets(useless_rowset_count,
342
0
                                                                  useless_rowset_version_count);
343
0
                if (useless_rowset_count > max_useless_rowset_count) {
344
0
                    max_useless_rowset_count = useless_rowset_count;
345
0
                    tablet_id_with_max_useless_rowset_count = tablet->tablet_id();
346
0
                }
347
0
                if (useless_rowset_version_count > max_useless_rowset_version_count) {
348
0
                    max_useless_rowset_version_count = useless_rowset_version_count;
349
0
                    tablet_id_with_max_useless_rowset_version_count = tablet->tablet_id();
350
0
                }
351
0
            });
352
0
            g_max_rowsets_with_useless_delete_bitmap.set_value(max_useless_rowset_count);
353
0
            g_max_rowsets_with_useless_delete_bitmap_version.set_value(
354
0
                    max_useless_rowset_version_count);
355
0
            LOG(INFO) << "finish check_agg_delete_bitmap_for_stale_rowsets, cost(us)="
356
0
                      << watch.get_elapse_time_us()
357
0
                      << ". max useless rowset count=" << max_useless_rowset_count
358
0
                      << ", tablet_id=" << tablet_id_with_max_useless_rowset_count
359
0
                      << ", max useless rowset version count=" << max_useless_rowset_version_count
360
0
                      << ", tablet_id=" << tablet_id_with_max_useless_rowset_version_count;
361
0
        }
362
13
    }
363
13
    {
364
13
        _tablet_map->traverse(
365
717k
                [](auto&& tablet) { tablet->clear_unused_visible_pending_rowsets(); });
366
13
    }
367
13
}
368
369
19.8k
std::vector<std::weak_ptr<CloudTablet>> CloudTabletMgr::get_weak_tablets() {
370
19.8k
    std::vector<std::weak_ptr<CloudTablet>> weak_tablets;
371
19.8k
    weak_tablets.reserve(_tablet_map->size());
372
687M
    _tablet_map->traverse([&weak_tablets](auto& t) { weak_tablets.push_back(t); });
373
19.8k
    return weak_tablets;
374
19.8k
}
375
376
6
void CloudTabletMgr::sync_tablets(const CountDownLatch& stop_latch) {
377
6
    LOG_INFO("begin to sync tablets");
378
6
    int64_t last_sync_time_bound = ::time(nullptr) - config::tablet_sync_interval_s;
379
380
6
    auto weak_tablets = get_weak_tablets();
381
382
    // sort by last_sync_time
383
925k
    static auto cmp = [](const auto& a, const auto& b) { return a.first < b.first; };
384
6
    std::multiset<std::pair<int64_t, std::weak_ptr<CloudTablet>>, decltype(cmp)>
385
6
            sync_time_tablet_set(cmp);
386
387
332k
    for (auto& weak_tablet : weak_tablets) {
388
332k
        if (auto tablet = weak_tablet.lock()) {
389
332k
            int64_t last_sync_time = tablet->last_sync_time_s;
390
332k
            if (last_sync_time <= last_sync_time_bound) {
391
48.1k
                sync_time_tablet_set.emplace(last_sync_time, weak_tablet);
392
48.1k
            }
393
332k
        }
394
332k
    }
395
396
6
    int num_sync = 0;
397
48.1k
    for (auto&& [_, weak_tablet] : sync_time_tablet_set) {
398
48.1k
        if (stop_latch.count() <= 0) {
399
0
            break;
400
0
        }
401
402
48.1k
        if (auto tablet = weak_tablet.lock()) {
403
48.1k
            if (tablet->last_sync_time_s > last_sync_time_bound) {
404
0
                continue;
405
0
            }
406
407
48.1k
            ++num_sync;
408
48.1k
            auto st = tablet->sync_meta();
409
48.1k
            if (!st) {
410
0
                LOG_WARNING("failed to sync tablet meta {}", tablet->tablet_id()).error(st);
411
0
                if (st.is<ErrorCode::NOT_FOUND>()) {
412
0
                    continue;
413
0
                }
414
0
            }
415
48.1k
            SyncOptions options;
416
48.1k
            options.query_version = -1;
417
48.1k
            options.merge_schema = true;
418
48.1k
            st = tablet->sync_rowsets(options);
419
48.1k
            if (!st) {
420
0
                LOG_WARNING("failed to sync tablet rowsets {}", tablet->tablet_id()).error(st);
421
0
            }
422
48.1k
        }
423
48.1k
    }
424
6
    LOG_INFO("finish sync tablets").tag("num_sync", num_sync);
425
6
}
426
427
Status CloudTabletMgr::get_topn_tablets_to_compact(
428
        int n, CompactionType compaction_type, const std::function<bool(CloudTablet*)>& filter_out,
429
19.6k
        std::vector<std::shared_ptr<CloudTablet>>* tablets, int64_t* max_score) {
430
19.6k
    DCHECK(compaction_type == CompactionType::BASE_COMPACTION ||
431
19.6k
           compaction_type == CompactionType::CUMULATIVE_COMPACTION);
432
19.6k
    *max_score = 0;
433
19.6k
    int64_t max_score_tablet_id = 0;
434
    // clang-format off
435
681M
    auto score = [compaction_type](CloudTablet* t) {
436
681M
        return compaction_type == CompactionType::BASE_COMPACTION ? t->get_cloud_base_compaction_score()
437
681M
               : compaction_type == CompactionType::CUMULATIVE_COMPACTION ? t->get_cloud_cumu_compaction_score()
438
613M
               : 0;
439
681M
    };
440
441
19.6k
    using namespace std::chrono;
442
19.6k
    auto now = duration_cast<milliseconds>(system_clock::now().time_since_epoch()).count();
443
300M
    auto skip = [now, compaction_type](CloudTablet* t) {
444
300M
        auto* cloud_cluster_info = static_cast<CloudClusterInfo*>(ExecEnv::GetInstance()->cluster_info());
445
446
300M
        if (config::enable_standby_passive_compaction && cloud_cluster_info->is_in_standby()) {
447
0
            if (t->fetch_add_approximate_num_rowsets(0) < config::max_tablet_version_num * config::standby_compaction_version_ratio) {
448
0
                return true;
449
0
            }
450
0
        }
451
452
        // Compaction read-write separation: skip tablets that should be compacted by other clusters.
453
        // Placed after standby check so standby invariants (version count threshold) are preserved.
454
300M
        if (cloud_cluster_info->should_skip_compaction(t)) {
455
0
            return true;
456
0
        }
457
458
300M
        int32_t max_version_config = t->max_version_config();
459
300M
        if (compaction_type == CompactionType::BASE_COMPACTION) {
460
67.4M
            bool is_recent_failure = now - t->last_base_compaction_failure_time() < config::min_compaction_failure_interval_ms;
461
67.4M
            bool is_frozen = (now - t->last_load_time_ms > config::compaction_load_max_freeze_interval_s * 1000
462
67.4M
                   && now - t->last_base_compaction_success_time_ms < config::base_compaction_freeze_interval_s * 1000
463
67.4M
                   && t->fetch_add_approximate_num_rowsets(0) < max_version_config / 2);
464
67.4M
            g_base_compaction_not_frozen_tablet_num << !is_frozen;
465
67.4M
            return is_recent_failure || is_frozen;
466
67.4M
        }
467
468
        // If tablet has too many rowsets but not be compacted for a long time, compaction should be performed
469
        // regardless of whether there is a load job recently.
470
233M
        bool is_recent_failure = now - t->last_cumu_compaction_failure_time() < config::min_compaction_failure_interval_ms;
471
233M
        bool is_recent_no_suitable_version = now - t->last_cumu_no_suitable_version_ms < config::min_compaction_failure_interval_ms;
472
233M
        bool is_frozen = (now - t->last_load_time_ms > config::compaction_load_max_freeze_interval_s * 1000
473
233M
               && now - t->last_cumu_compaction_success_time_ms < config::cumu_compaction_interval_s * 1000
474
233M
               && t->fetch_add_approximate_num_rowsets(0) < max_version_config / 2);
475
233M
        g_cumu_compaction_not_frozen_tablet_num << !is_frozen;
476
233M
        return is_recent_failure || is_recent_no_suitable_version || is_frozen;
477
300M
    };
478
    // We don't schedule tablets that are disabled for compaction
479
305M
    auto disable = [](CloudTablet* t) { return t->tablet_meta()->tablet_schema()->disable_auto_compaction(); };
480
481
19.6k
    auto [num_filtered, num_disabled, num_skipped] = std::make_tuple(0, 0, 0);
482
483
19.6k
    auto weak_tablets = get_weak_tablets();
484
19.6k
    std::vector<std::pair<std::shared_ptr<CloudTablet>, int64_t>> buf;
485
19.6k
    buf.reserve(n + 1);
486
681M
    for (auto& weak_tablet : weak_tablets) {
487
681M
        auto t = weak_tablet.lock();
488
681M
        if (t == nullptr) { continue; }
489
490
681M
        int64_t s = score(t.get());
491
681M
        if (s <= 0) { continue; }
492
305M
        if (s > *max_score) {
493
103k
            max_score_tablet_id = t->tablet_id();
494
103k
            *max_score = s;
495
103k
        }
496
497
305M
        if (filter_out(t.get())) { ++num_filtered; continue; }
498
305M
        if (disable(t.get())) { ++num_disabled; continue; }
499
300M
        if (skip(t.get())) { ++num_skipped; continue; }
500
501
295M
        buf.emplace_back(std::move(t), s);
502
3.45G
        std::sort(buf.begin(), buf.end(), [](auto& a, auto& b) { return a.second > b.second; });
503
295M
        if (buf.size() > n) { buf.pop_back(); }
504
295M
    }
505
506
19.6k
    LOG_EVERY_N(INFO, 1000) << "get_topn_compaction_score, n=" << n << " type=" << compaction_type
507
21
               << " num_tablets=" << weak_tablets.size() << " num_skipped=" << num_skipped
508
21
               << " num_disabled=" << num_disabled << " num_filtered=" << num_filtered
509
21
               << " max_score=" << *max_score << " max_score_tablet=" << max_score_tablet_id
510
125
               << " tablets=[" << [&buf] { std::stringstream ss; for (auto& i : buf) ss << i.first->tablet_id() << ":" << i.second << ","; return ss.str(); }() << "]"
511
19.6k
               ;
512
    // clang-format on
513
514
19.6k
    tablets->clear();
515
19.6k
    tablets->reserve(n + 1);
516
119k
    for (auto& [t, _] : buf) {
517
119k
        tablets->emplace_back(std::move(t));
518
119k
    }
519
520
19.6k
    return Status::OK();
521
19.6k
}
522
523
void CloudTabletMgr::build_all_report_tablets_info(std::map<TTabletId, TTablet>* tablets_info,
524
85
                                                   uint64_t* tablet_num) {
525
85
    DCHECK(tablets_info != nullptr);
526
85
    VLOG_NOTICE << "begin to build all report cloud tablets info";
527
528
85
    HistogramStat tablet_version_num_hist;
529
530
4.88M
    auto handler = [&](const std::weak_ptr<CloudTablet>& tablet_wk) {
531
4.88M
        auto tablet = tablet_wk.lock();
532
4.88M
        if (!tablet) return;
533
4.88M
        (*tablet_num)++;
534
4.88M
        TTabletInfo tablet_info;
535
4.88M
        tablet->build_tablet_report_info(&tablet_info);
536
4.88M
        using namespace std::chrono;
537
4.88M
        int64_t now = duration_cast<milliseconds>(system_clock::now().time_since_epoch()).count();
538
4.88M
        if (now - g_tablet_report_inactive_duration_ms < tablet->last_access_time_ms) {
539
            // the tablet is still being accessed and used in recently, so not report it
540
4.88M
            return;
541
4.88M
        }
542
0
        auto& t_tablet = (*tablets_info)[tablet->tablet_id()];
543
        // On the cloud, a specific BE has only one tablet replica;
544
        // there are no multiple replicas for a specific BE.
545
        // This is only to reuse the non-cloud report protocol.
546
0
        tablet_version_num_hist.add(tablet_info.total_version_count);
547
0
        t_tablet.tablet_infos.emplace_back(std::move(tablet_info));
548
0
    };
549
550
85
    auto weak_tablets = get_weak_tablets();
551
85
    std::for_each(weak_tablets.begin(), weak_tablets.end(), handler);
552
553
85
    DorisMetrics::instance()->tablet_version_num_distribution->set_histogram(
554
85
            tablet_version_num_hist);
555
85
    LOG(INFO) << "success to build all cloud report tablets info. all_tablet_count=" << *tablet_num
556
85
              << " exceed drop time limit count=" << tablets_info->size();
557
85
}
558
559
0
void CloudTabletMgr::get_tablet_info(int64_t num_tablets, std::vector<TabletInfo>* tablets_info) {
560
0
    auto weak_tablets = get_weak_tablets();
561
0
    for (auto& weak_tablet : weak_tablets) {
562
0
        auto tablet = weak_tablet.lock();
563
0
        if (tablet == nullptr) {
564
0
            continue;
565
0
        }
566
0
        if (tablets_info->size() >= num_tablets) {
567
0
            return;
568
0
        }
569
0
        tablets_info->push_back(tablet->get_tablet_info());
570
0
    }
571
0
}
572
573
void CloudTabletMgr::get_topn_tablet_delete_bitmap_score(
574
13
        uint64_t* max_delete_bitmap_score, uint64_t* max_base_rowset_delete_bitmap_score) {
575
13
    int64_t max_delete_bitmap_score_tablet_id = 0;
576
13
    OlapStopWatch watch;
577
13
    uint64_t total_delete_map_count = 0;
578
13
    int64_t max_base_rowset_delete_bitmap_score_tablet_id = 0;
579
13
    int n = config::check_tablet_delete_bitmap_score_top_n;
580
13
    std::vector<std::pair<std::shared_ptr<CloudTablet>, int64_t>> buf;
581
13
    buf.reserve(n + 1);
582
716k
    auto handler = [&](const std::weak_ptr<CloudTablet>& tablet_wk) {
583
716k
        auto t = tablet_wk.lock();
584
716k
        if (!t) return;
585
716k
        uint64_t delete_bitmap_count =
586
716k
                t.get()->tablet_meta()->delete_bitmap().get_delete_bitmap_count();
587
716k
        total_delete_map_count += delete_bitmap_count;
588
716k
        if (delete_bitmap_count > *max_delete_bitmap_score) {
589
82
            max_delete_bitmap_score_tablet_id = t->tablet_id();
590
82
            *max_delete_bitmap_score = delete_bitmap_count;
591
82
        }
592
716k
        buf.emplace_back(std::move(t), delete_bitmap_count);
593
14.3M
        std::sort(buf.begin(), buf.end(), [](auto& a, auto& b) { return a.second > b.second; });
594
716k
        if (buf.size() > n) {
595
716k
            buf.pop_back();
596
716k
        }
597
716k
    };
598
13
    auto weak_tablets = get_weak_tablets();
599
13
    std::for_each(weak_tablets.begin(), weak_tablets.end(), handler);
600
130
    for (auto& [t, _] : buf) {
601
130
        t->get_base_rowset_delete_bitmap_count(max_base_rowset_delete_bitmap_score,
602
130
                                               &max_base_rowset_delete_bitmap_score_tablet_id);
603
130
    }
604
13
    std::stringstream ss;
605
130
    for (auto& i : buf) {
606
130
        ss << i.first->tablet_id() << ": " << i.second << ", ";
607
130
    }
608
13
    LOG(INFO) << "get_topn_tablet_delete_bitmap_score, n=" << n
609
13
              << ", tablet size=" << weak_tablets.size()
610
13
              << ", total_delete_map_count=" << total_delete_map_count
611
13
              << ", cost(us)=" << watch.get_elapse_time_us()
612
13
              << ", max_delete_bitmap_score=" << *max_delete_bitmap_score
613
13
              << ", max_delete_bitmap_score_tablet_id=" << max_delete_bitmap_score_tablet_id
614
13
              << ", max_base_rowset_delete_bitmap_score=" << *max_base_rowset_delete_bitmap_score
615
13
              << ", max_base_rowset_delete_bitmap_score_tablet_id="
616
13
              << max_base_rowset_delete_bitmap_score_tablet_id << ", tablets=[" << ss.str() << "]";
617
13
}
618
619
2
std::vector<std::shared_ptr<CloudTablet>> CloudTabletMgr::get_all_tablet() {
620
2
    std::vector<std::shared_ptr<CloudTablet>> tablets;
621
2
    tablets.reserve(_tablet_map->size());
622
12.4k
    _tablet_map->traverse([&tablets](auto& t) { tablets.push_back(t); });
623
2
    return tablets;
624
2
}
625
626
2
void CloudTabletMgr::put_tablet_for_UT(std::shared_ptr<CloudTablet> tablet) {
627
2
    _tablet_map->put(tablet);
628
2
}
629
630
} // namespace doris