Coverage Report

Created: 2025-06-13 16:38

/root/doris/be/src/olap/cumulative_compaction.cpp
Line
Count
Source (jump to first uncovered line)
1
// Licensed to the Apache Software Foundation (ASF) under one
2
// or more contributor license agreements.  See the NOTICE file
3
// distributed with this work for additional information
4
// regarding copyright ownership.  The ASF licenses this file
5
// to you under the Apache License, Version 2.0 (the
6
// "License"); you may not use this file except in compliance
7
// with the License.  You may obtain a copy of the License at
8
//
9
//   http://www.apache.org/licenses/LICENSE-2.0
10
//
11
// Unless required by applicable law or agreed to in writing,
12
// software distributed under the License is distributed on an
13
// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
14
// KIND, either express or implied.  See the License for the
15
// specific language governing permissions and limitations
16
// under the License.
17
18
#include "olap/cumulative_compaction.h"
19
20
#include <cpp/sync_point.h>
21
22
#include <memory>
23
#include <mutex>
24
#include <ostream>
25
26
#include "common/config.h"
27
#include "common/logging.h"
28
#include "olap/cumulative_compaction_policy.h"
29
#include "olap/cumulative_compaction_time_series_policy.h"
30
#include "olap/olap_define.h"
31
#include "olap/rowset/rowset_meta.h"
32
#include "olap/tablet.h"
33
#include "runtime/thread_context.h"
34
#include "util/doris_metrics.h"
35
#include "util/time.h"
36
#include "util/trace.h"
37
38
namespace doris {
39
using namespace ErrorCode;
40
41
void CumulativeCompaction::find_longest_consecutive_version(std::vector<RowsetSharedPtr>* rowsets,
42
28
                                                            std::vector<Version>* missing_version) {
43
28
    if (rowsets->empty()) {
44
0
        return;
45
0
    }
46
47
28
    RowsetSharedPtr prev_rowset = rowsets->front();
48
28
    size_t i = 1;
49
28
    int max_start = 0;
50
28
    int max_length = 1;
51
52
28
    int start = 0;
53
28
    int length = 1;
54
631
    for (; i < rowsets->size(); ++i) {
55
603
        RowsetSharedPtr rowset = (*rowsets)[i];
56
603
        if (rowset->start_version() != prev_rowset->end_version() + 1) {
57
6
            if (missing_version != nullptr) {
58
6
                missing_version->push_back(prev_rowset->version());
59
6
                missing_version->push_back(rowset->version());
60
6
            }
61
6
            start = i;
62
6
            length = 1;
63
597
        } else {
64
597
            length++;
65
597
        }
66
67
603
        if (length > max_length) {
68
587
            max_start = start;
69
587
            max_length = length;
70
587
        }
71
72
603
        prev_rowset = rowset;
73
603
    }
74
28
    *rowsets = {rowsets->begin() + max_start, rowsets->begin() + max_start + max_length};
75
28
}
76
77
CumulativeCompaction::CumulativeCompaction(StorageEngine& engine, const TabletSharedPtr& tablet)
78
        : CompactionMixin(engine, tablet,
79
33
                          "CumulativeCompaction:" + std::to_string(tablet->tablet_id())) {}
80
81
33
CumulativeCompaction::~CumulativeCompaction() = default;
82
83
21
Status CumulativeCompaction::prepare_compact() {
84
21
    Status st;
85
21
    Defer defer_set_st([&] {
86
21
        if (!st.ok()) {
87
0
            tablet()->set_last_cumu_compaction_status(st.to_string());
88
0
        }
89
21
    });
90
91
21
    if (!tablet()->init_succeeded()) {
92
0
        st = Status::Error<CUMULATIVE_INVALID_PARAMETERS, false>("_tablet init failed");
93
0
        return st;
94
0
    }
95
96
21
    std::unique_lock<std::mutex> lock(tablet()->get_cumulative_compaction_lock(), std::try_to_lock);
97
21
    if (!lock.owns_lock()) {
98
0
        st = Status::Error<TRY_LOCK_FAILED, false>(
99
0
                "The tablet is under cumulative compaction. tablet={}", _tablet->tablet_id());
100
0
        return st;
101
0
    }
102
103
21
    tablet()->calculate_cumulative_point();
104
21
    VLOG_CRITICAL << "after calculate, current cumulative point is "
105
0
                  << tablet()->cumulative_layer_point() << ", tablet=" << _tablet->tablet_id();
106
107
21
    st = pick_rowsets_to_compact();
108
21
    RETURN_IF_ERROR(st);
109
110
21
    COUNTER_UPDATE(_input_rowsets_counter, _input_rowsets.size());
111
112
21
    st = Status::OK();
113
21
    return st;
114
21
}
115
116
1
Status CumulativeCompaction::execute_compact() {
117
1
    DBUG_EXECUTE_IF("CumulativeCompaction::execute_compact.block", {
118
1
        auto target_tablet_id = dp->param<int64_t>("tablet_id", -1);
119
1
        if (target_tablet_id == _tablet->tablet_id()) {
120
1
            LOG(INFO) << "start debug block "
121
1
                      << "CumulativeCompaction::execute_compact.block";
122
1
            while (DebugPoints::instance()->is_enable(
123
1
                    "CumulativeCompaction::execute_compact.block")) {
124
1
                std::this_thread::sleep_for(std::chrono::milliseconds(200));
125
1
            }
126
1
            LOG(INFO) << "end debug block "
127
1
                      << "CumulativeCompaction::execute_compact.block";
128
1
        }
129
1
    })
130
131
1
    Status st;
132
1
    Defer defer_set_st([&] {
133
1
        tablet()->set_last_cumu_compaction_status(st.to_string());
134
1
        if (!st.ok()) {
135
0
            tablet()->set_last_cumu_compaction_failure_time(UnixMillis());
136
1
        } else {
137
            // TIME_SERIES_POLICY, generating an empty rowset doesn't need to update the timestamp.
138
1
            if (!(tablet()->tablet_meta()->compaction_policy() == CUMULATIVE_TIME_SERIES_POLICY &&
139
1
                  _output_rowset->num_segments() == 0)) {
140
1
                tablet()->set_last_cumu_compaction_success_time(UnixMillis());
141
1
            }
142
1
        }
143
1
    });
144
1
    std::unique_lock<std::mutex> lock(tablet()->get_cumulative_compaction_lock(), std::try_to_lock);
145
1
    if (!lock.owns_lock()) {
146
0
        st = Status::Error<TRY_LOCK_FAILED, false>(
147
0
                "The tablet is under cumulative compaction. tablet={}", _tablet->tablet_id());
148
0
        return st;
149
0
    }
150
151
1
    SCOPED_ATTACH_TASK(_mem_tracker);
152
153
1
    st = CompactionMixin::execute_compact();
154
1
    RETURN_IF_ERROR(st);
155
156
1
    TEST_SYNC_POINT_RETURN_WITH_VALUE(
157
0
            "cumulative_compaction::CumulativeCompaction::execute_compact", Status::OK());
158
159
0
    DCHECK_EQ(_state, CompactionState::SUCCESS);
160
161
0
    tablet()->cumulative_compaction_policy()->update_cumulative_point(
162
0
            tablet(), _input_rowsets, _output_rowset, _last_delete_version);
163
0
    VLOG_CRITICAL << "after cumulative compaction, current cumulative point is "
164
0
                  << tablet()->cumulative_layer_point() << ", tablet=" << _tablet->tablet_id();
165
0
    DorisMetrics::instance()->cumulative_compaction_deltas_total->increment(_input_rowsets.size());
166
0
    DorisMetrics::instance()->cumulative_compaction_bytes_total->increment(
167
0
            _input_rowsets_total_size);
168
169
0
    st = Status::OK();
170
0
    return st;
171
1
}
172
173
21
Status CumulativeCompaction::pick_rowsets_to_compact() {
174
21
    auto candidate_rowsets = tablet()->pick_candidate_rowsets_to_cumulative_compaction();
175
21
    if (candidate_rowsets.empty()) {
176
0
        return Status::Error<CUMULATIVE_NO_SUITABLE_VERSION>("candidate_rowsets is empty");
177
0
    }
178
179
    // candidate_rowsets may not be continuous
180
    // So we need to choose the longest continuous path from it.
181
21
    std::vector<Version> missing_versions;
182
21
    find_longest_consecutive_version(&candidate_rowsets, &missing_versions);
183
21
    if (!missing_versions.empty()) {
184
0
        DCHECK(missing_versions.size() % 2 == 0);
185
0
        LOG(WARNING) << "There are missed versions among rowsets. "
186
0
                     << "total missed version size: " << missing_versions.size() / 2
187
0
                     << ", first missed version prev rowset verison=" << missing_versions[0]
188
0
                     << ", first missed version next rowset version=" << missing_versions[1]
189
0
                     << ", tablet=" << _tablet->tablet_id();
190
0
    }
191
192
21
    int64_t max_score = config::cumulative_compaction_max_deltas;
193
21
    auto process_memory_usage = doris::GlobalMemoryArbitrator::process_memory_usage();
194
21
    bool memory_usage_high = process_memory_usage > MemInfo::soft_mem_limit() * 0.8;
195
21
    if (tablet()->last_compaction_status.is<ErrorCode::MEM_LIMIT_EXCEEDED>() || memory_usage_high) {
196
0
        max_score = std::max(config::cumulative_compaction_max_deltas /
197
0
                                     config::cumulative_compaction_max_deltas_factor,
198
0
                             config::cumulative_compaction_min_deltas + 1);
199
0
    }
200
201
21
    size_t compaction_score = 0;
202
21
    tablet()->cumulative_compaction_policy()->pick_input_rowsets(
203
21
            tablet(), candidate_rowsets, max_score, config::cumulative_compaction_min_deltas,
204
21
            &_input_rowsets, &_last_delete_version, &compaction_score,
205
21
            _allow_delete_in_cumu_compaction);
206
207
    // Cumulative compaction will process with at least 1 rowset.
208
    // So when there is no rowset being chosen, we should return Status::Error<CUMULATIVE_NO_SUITABLE_VERSION>():
209
21
    if (_input_rowsets.empty()) {
210
0
        if (_last_delete_version.first != -1) {
211
            // we meet a delete version, should increase the cumulative point to let base compaction handle the delete version.
212
            // plus 1 to skip the delete version.
213
            // NOTICE: after that, the cumulative point may be larger than max version of this tablet, but it doesn't matter.
214
0
            tablet()->set_cumulative_layer_point(_last_delete_version.first + 1);
215
0
            LOG_INFO(
216
0
                    "cumulative compaction meet delete rowset, increase cumu point without "
217
0
                    "other "
218
0
                    "operation.")
219
0
                    .tag("tablet id:", tablet()->tablet_id())
220
0
                    .tag("after cumulative compaction, cumu point:",
221
0
                         tablet()->cumulative_layer_point());
222
0
            return Status::Error<CUMULATIVE_MEET_DELETE_VERSION>(
223
0
                    "cumulative compaction meet delete version");
224
0
        }
225
226
        // we did not meet any delete version. which means compaction_score is not enough to do cumulative compaction.
227
        // We should wait until there are more rowsets to come, and keep the cumulative point unchanged.
228
        // But in order to avoid the stall of compaction because no new rowset arrives later, we should increase
229
        // the cumulative point after waiting for a long time, to ensure that the base compaction can continue.
230
231
        // check both last success time of base and cumulative compaction
232
0
        int64_t now = UnixMillis();
233
0
        int64_t last_cumu = tablet()->last_cumu_compaction_success_time();
234
0
        int64_t last_base = tablet()->last_base_compaction_success_time();
235
0
        if (last_cumu != 0 || last_base != 0) {
236
0
            int64_t interval_threshold = config::pick_rowset_to_compact_interval_sec * 1000;
237
0
            int64_t cumu_interval = now - last_cumu;
238
0
            int64_t base_interval = now - last_base;
239
0
            if (cumu_interval > interval_threshold && base_interval > interval_threshold) {
240
                // before increasing cumulative point, we should make sure all rowsets are non-overlapping.
241
                // if at least one rowset is overlapping, we should compact them first.
242
0
                for (auto& rs : candidate_rowsets) {
243
0
                    if (rs->rowset_meta()->is_segments_overlapping()) {
244
0
                        _input_rowsets = candidate_rowsets;
245
0
                        return Status::OK();
246
0
                    }
247
0
                }
248
249
                // all candidate rowsets are non-overlapping, increase the cumulative point
250
0
                tablet()->set_cumulative_layer_point(candidate_rowsets.back()->start_version() + 1);
251
0
            }
252
0
        } else {
253
            // init the compaction success time for first time
254
0
            if (last_cumu == 0) {
255
0
                tablet()->set_last_cumu_compaction_success_time(now);
256
0
            }
257
258
0
            if (last_base == 0) {
259
0
                tablet()->set_last_base_compaction_success_time(now);
260
0
            }
261
0
        }
262
263
0
        return Status::Error<CUMULATIVE_NO_SUITABLE_VERSION>("_input_rowsets is empty");
264
0
    }
265
266
21
    return Status::OK();
267
21
}
268
269
} // namespace doris