Coverage Report

Created: 2025-07-24 22:14

next uncovered line (L), next uncovered region (R), next uncovered branch (B)
/root/doris/be/src/olap/cumulative_compaction.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 "olap/cumulative_compaction.h"
19
20
#include <cpp/sync_point.h>
21
#include <gen_cpp/AgentService_types.h>
22
#include <gen_cpp/Types_types.h>
23
24
#include <memory>
25
#include <mutex>
26
#include <ostream>
27
#include <vector>
28
29
#include "common/config.h"
30
#include "common/logging.h"
31
#include "olap/cumulative_compaction_policy.h"
32
#include "olap/cumulative_compaction_time_series_policy.h"
33
#include "olap/olap_define.h"
34
#include "olap/rowset/rowset_meta.h"
35
#include "olap/storage_engine.h"
36
#include "olap/tablet.h"
37
#include "runtime/exec_env.h"
38
#include "runtime/thread_context.h"
39
#include "util/doris_metrics.h"
40
#include "util/time.h"
41
#include "util/trace.h"
42
43
namespace doris {
44
#include "common/compile_check_begin.h"
45
using namespace ErrorCode;
46
47
void CumulativeCompaction::find_longest_consecutive_version(std::vector<RowsetSharedPtr>* rowsets,
48
28
                                                            std::vector<Version>* missing_version) {
49
28
    if (rowsets->empty()) {
50
0
        return;
51
0
    }
52
53
28
    RowsetSharedPtr prev_rowset = rowsets->front();
54
28
    int i = 1;
55
28
    int max_start = 0;
56
28
    int max_length = 1;
57
58
28
    int start = 0;
59
28
    int length = 1;
60
631
    for (; i < rowsets->size(); ++i) {
61
603
        RowsetSharedPtr rowset = (*rowsets)[i];
62
603
        if (rowset->start_version() != prev_rowset->end_version() + 1) {
63
6
            if (missing_version != nullptr) {
64
6
                missing_version->push_back(prev_rowset->version());
65
6
                missing_version->push_back(rowset->version());
66
6
            }
67
6
            start = i;
68
6
            length = 1;
69
597
        } else {
70
597
            length++;
71
597
        }
72
73
603
        if (length > max_length) {
74
587
            max_start = start;
75
587
            max_length = length;
76
587
        }
77
78
603
        prev_rowset = rowset;
79
603
    }
80
28
    *rowsets = {rowsets->begin() + max_start, rowsets->begin() + max_start + max_length};
81
28
}
82
83
CumulativeCompaction::CumulativeCompaction(StorageEngine& engine, const TabletSharedPtr& tablet)
84
33
        : CompactionMixin(engine, tablet,
85
33
                          "CumulativeCompaction:" + std::to_string(tablet->tablet_id())) {}
86
87
33
CumulativeCompaction::~CumulativeCompaction() = default;
88
89
21
Status CumulativeCompaction::prepare_compact() {
90
21
    Status st;
91
21
    Defer defer_set_st([&] {
92
21
        if (!st.ok()) {
93
0
            tablet()->set_last_cumu_compaction_status(st.to_string());
94
0
        }
95
21
    });
96
97
21
    if (!tablet()->init_succeeded()) {
98
0
        st = Status::Error<CUMULATIVE_INVALID_PARAMETERS, false>("_tablet init failed");
99
0
        return st;
100
0
    }
101
102
21
    std::unique_lock<std::mutex> lock(tablet()->get_cumulative_compaction_lock(), std::try_to_lock);
103
21
    if (!lock.owns_lock()) {
104
0
        st = Status::Error<TRY_LOCK_FAILED, false>(
105
0
                "The tablet is under cumulative compaction. tablet={}", _tablet->tablet_id());
106
0
        return st;
107
0
    }
108
109
21
    tablet()->calculate_cumulative_point();
110
21
    VLOG_CRITICAL << "after calculate, current cumulative point is "
111
0
                  << tablet()->cumulative_layer_point() << ", tablet=" << _tablet->tablet_id();
112
113
21
    st = pick_rowsets_to_compact();
114
21
    RETURN_IF_ERROR(st);
115
116
21
    COUNTER_UPDATE(_input_rowsets_counter, _input_rowsets.size());
117
118
21
    st = Status::OK();
119
21
    return st;
120
21
}
121
122
1
Status CumulativeCompaction::execute_compact() {
123
1
    DBUG_EXECUTE_IF("CumulativeCompaction::execute_compact.block", {
124
1
        auto target_tablet_id = dp->param<int64_t>("tablet_id", -1);
125
1
        if (target_tablet_id == _tablet->tablet_id()) {
126
1
            LOG(INFO) << "start debug block "
127
1
                      << "CumulativeCompaction::execute_compact.block";
128
1
            while (DebugPoints::instance()->is_enable(
129
1
                    "CumulativeCompaction::execute_compact.block")) {
130
1
                std::this_thread::sleep_for(std::chrono::milliseconds(200));
131
1
            }
132
1
            LOG(INFO) << "end debug block "
133
1
                      << "CumulativeCompaction::execute_compact.block";
134
1
        }
135
1
    })
136
137
1
    Status st;
138
1
    Defer defer_set_st([&] {
139
1
        tablet()->set_last_cumu_compaction_status(st.to_string());
140
1
        if (!st.ok()) {
141
0
            tablet()->set_last_cumu_compaction_failure_time(UnixMillis());
142
1
        } else {
143
            // TIME_SERIES_POLICY, generating an empty rowset doesn't need to update the timestamp.
144
1
            if (!(tablet()->tablet_meta()->compaction_policy() == CUMULATIVE_TIME_SERIES_POLICY &&
145
1
                  _output_rowset->num_segments() == 0)) {
146
1
                tablet()->set_last_cumu_compaction_success_time(UnixMillis());
147
1
            }
148
1
        }
149
1
    });
150
1
    std::unique_lock<std::mutex> lock(tablet()->get_cumulative_compaction_lock(), std::try_to_lock);
151
1
    if (!lock.owns_lock()) {
152
0
        st = Status::Error<TRY_LOCK_FAILED, false>(
153
0
                "The tablet is under cumulative compaction. tablet={}", _tablet->tablet_id());
154
0
        return st;
155
0
    }
156
157
1
    SCOPED_ATTACH_TASK(_mem_tracker);
158
159
1
    st = CompactionMixin::execute_compact();
160
1
    RETURN_IF_ERROR(st);
161
162
1
    TEST_SYNC_POINT_RETURN_WITH_VALUE(
163
0
            "cumulative_compaction::CumulativeCompaction::execute_compact", Status::OK());
164
165
0
    DCHECK_EQ(_state, CompactionState::SUCCESS);
166
167
0
    tablet()->cumulative_compaction_policy()->update_cumulative_point(
168
0
            tablet(), _input_rowsets, _output_rowset, _last_delete_version);
169
0
    VLOG_CRITICAL << "after cumulative compaction, current cumulative point is "
170
0
                  << tablet()->cumulative_layer_point() << ", tablet=" << _tablet->tablet_id();
171
0
    DorisMetrics::instance()->cumulative_compaction_deltas_total->increment(_input_rowsets.size());
172
0
    DorisMetrics::instance()->cumulative_compaction_bytes_total->increment(
173
0
            _input_rowsets_total_size);
174
175
0
    st = Status::OK();
176
0
    return st;
177
1
}
178
179
21
Status CumulativeCompaction::pick_rowsets_to_compact() {
180
21
    auto candidate_rowsets = tablet()->pick_candidate_rowsets_to_cumulative_compaction();
181
21
    if (candidate_rowsets.empty()) {
182
0
        return Status::Error<CUMULATIVE_NO_SUITABLE_VERSION>("candidate_rowsets is empty");
183
0
    }
184
185
    // candidate_rowsets may not be continuous
186
    // So we need to choose the longest continuous path from it.
187
21
    std::vector<Version> missing_versions;
188
21
    find_longest_consecutive_version(&candidate_rowsets, &missing_versions);
189
21
    if (!missing_versions.empty()) {
190
0
        DCHECK(missing_versions.size() % 2 == 0);
191
0
        LOG(WARNING) << "There are missed versions among rowsets. "
192
0
                     << "total missed version size: " << missing_versions.size() / 2
193
0
                     << ", first missed version prev rowset verison=" << missing_versions[0]
194
0
                     << ", first missed version next rowset version=" << missing_versions[1]
195
0
                     << ", tablet=" << _tablet->tablet_id();
196
0
        if (config::enable_auto_clone_on_compaction_missing_version) {
197
0
            LOG_INFO("cumulative compaction submit missing rowset clone task.")
198
0
                    .tag("tablet_id", _tablet->tablet_id())
199
0
                    .tag("version", missing_versions.back().first)
200
0
                    .tag("replica_id", tablet()->replica_id())
201
0
                    .tag("partition_id", _tablet->partition_id())
202
0
                    .tag("table_id", _tablet->table_id());
203
0
            Status st = _engine.submit_clone_task(tablet(), missing_versions.back().first);
204
0
            if (!st) {
205
0
                LOG_WARNING("cumulative compaction failed to submit missing rowset clone task.")
206
0
                        .tag("st", st.msg())
207
0
                        .tag("tablet_id", _tablet->tablet_id())
208
0
                        .tag("version", missing_versions.back().first)
209
0
                        .tag("replica_id", tablet()->replica_id())
210
0
                        .tag("partition_id", _tablet->partition_id())
211
0
                        .tag("table_id", _tablet->table_id());
212
0
            }
213
0
        }
214
0
    }
215
216
21
    int64_t max_score = config::cumulative_compaction_max_deltas;
217
21
    int64_t process_memory_usage = doris::GlobalMemoryArbitrator::process_memory_usage();
218
21
    bool memory_usage_high = process_memory_usage > MemInfo::soft_mem_limit() * 8 / 10;
219
21
    if (tablet()->last_compaction_status.is<ErrorCode::MEM_LIMIT_EXCEEDED>() || memory_usage_high) {
220
0
        max_score = std::max(config::cumulative_compaction_max_deltas /
221
0
                                     config::cumulative_compaction_max_deltas_factor,
222
0
                             config::cumulative_compaction_min_deltas + 1);
223
0
    }
224
225
21
    size_t compaction_score = 0;
226
21
    tablet()->cumulative_compaction_policy()->pick_input_rowsets(
227
21
            tablet(), candidate_rowsets, max_score, config::cumulative_compaction_min_deltas,
228
21
            &_input_rowsets, &_last_delete_version, &compaction_score,
229
21
            _allow_delete_in_cumu_compaction);
230
231
    // Cumulative compaction will process with at least 1 rowset.
232
    // So when there is no rowset being chosen, we should return Status::Error<CUMULATIVE_NO_SUITABLE_VERSION>():
233
21
    if (_input_rowsets.empty()) {
234
0
        if (_last_delete_version.first != -1) {
235
            // we meet a delete version, should increase the cumulative point to let base compaction handle the delete version.
236
            // plus 1 to skip the delete version.
237
            // NOTICE: after that, the cumulative point may be larger than max version of this tablet, but it doesn't matter.
238
0
            tablet()->set_cumulative_layer_point(_last_delete_version.first + 1);
239
0
            LOG_INFO(
240
0
                    "cumulative compaction meet delete rowset, increase cumu point without "
241
0
                    "other "
242
0
                    "operation.")
243
0
                    .tag("tablet id:", tablet()->tablet_id())
244
0
                    .tag("after cumulative compaction, cumu point:",
245
0
                         tablet()->cumulative_layer_point());
246
0
            return Status::Error<CUMULATIVE_MEET_DELETE_VERSION>(
247
0
                    "cumulative compaction meet delete version");
248
0
        }
249
250
        // we did not meet any delete version. which means compaction_score is not enough to do cumulative compaction.
251
        // We should wait until there are more rowsets to come, and keep the cumulative point unchanged.
252
        // But in order to avoid the stall of compaction because no new rowset arrives later, we should increase
253
        // the cumulative point after waiting for a long time, to ensure that the base compaction can continue.
254
255
        // check both last success time of base and cumulative compaction
256
0
        int64_t now = UnixMillis();
257
0
        int64_t last_cumu = tablet()->last_cumu_compaction_success_time();
258
0
        int64_t last_base = tablet()->last_base_compaction_success_time();
259
0
        if (last_cumu != 0 || last_base != 0) {
260
0
            int64_t interval_threshold = config::pick_rowset_to_compact_interval_sec * 1000;
261
0
            int64_t cumu_interval = now - last_cumu;
262
0
            int64_t base_interval = now - last_base;
263
0
            if (cumu_interval > interval_threshold && base_interval > interval_threshold) {
264
                // before increasing cumulative point, we should make sure all rowsets are non-overlapping.
265
                // if at least one rowset is overlapping, we should compact them first.
266
0
                for (auto& rs : candidate_rowsets) {
267
0
                    if (rs->rowset_meta()->is_segments_overlapping()) {
268
0
                        _input_rowsets = candidate_rowsets;
269
0
                        return Status::OK();
270
0
                    }
271
0
                }
272
273
                // all candidate rowsets are non-overlapping, increase the cumulative point
274
0
                tablet()->set_cumulative_layer_point(candidate_rowsets.back()->start_version() + 1);
275
0
            }
276
0
        } else {
277
            // init the compaction success time for first time
278
0
            if (last_cumu == 0) {
279
0
                tablet()->set_last_cumu_compaction_success_time(now);
280
0
            }
281
282
0
            if (last_base == 0) {
283
0
                tablet()->set_last_base_compaction_success_time(now);
284
0
            }
285
0
        }
286
287
0
        return Status::Error<CUMULATIVE_NO_SUITABLE_VERSION>("_input_rowsets is empty");
288
0
    }
289
290
21
    return Status::OK();
291
21
}
292
#include "common/compile_check_end.h"
293
294
} // namespace doris