]> git.proxmox.com Git - ceph.git/blob - ceph/src/rocksdb/tools/trace_analyzer_tool.h
update sources to ceph Nautilus 14.2.1
[ceph.git] / ceph / src / rocksdb / tools / trace_analyzer_tool.h
1 // Copyright (c) 2011-present, Facebook, Inc. All rights reserved.
2 // This source code is licensed under both the GPLv2 (found in the
3 // COPYING file in the root directory) and Apache 2.0 License
4 // (found in the LICENSE.Apache file in the root directory).
5
6 #pragma once
7 #ifndef ROCKSDB_LITE
8
9 #include <list>
10 #include <map>
11 #include <queue>
12 #include <set>
13 #include <utility>
14 #include <vector>
15
16 #include "rocksdb/env.h"
17 #include "rocksdb/trace_reader_writer.h"
18 #include "rocksdb/write_batch.h"
19 #include "util/trace_replay.h"
20
21 namespace rocksdb {
22
23 class DBImpl;
24 class WriteBatch;
25
26 enum TraceOperationType : int {
27 kGet = 0,
28 kPut = 1,
29 kDelete = 2,
30 kSingleDelete = 3,
31 kRangeDelete = 4,
32 kMerge = 5,
33 kIteratorSeek = 6,
34 kIteratorSeekForPrev = 7,
35 kTaTypeNum = 8
36 };
37
38 struct TraceUnit {
39 uint64_t ts;
40 uint32_t type;
41 uint32_t cf_id;
42 size_t value_size;
43 std::string key;
44 };
45
46 struct TypeCorrelation {
47 uint64_t count;
48 uint64_t total_ts;
49 };
50
51 struct StatsUnit {
52 uint64_t key_id;
53 uint64_t access_count;
54 uint64_t latest_ts;
55 uint64_t succ_count; // current only used to count Get if key found
56 uint32_t cf_id;
57 size_t value_size;
58 std::vector<TypeCorrelation> v_correlation;
59 };
60
61 class AnalyzerOptions {
62 public:
63 std::vector<std::vector<int>> correlation_map;
64 std::vector<std::pair<int, int>> correlation_list;
65
66 AnalyzerOptions();
67
68 ~AnalyzerOptions();
69
70 void SparseCorrelationInput(const std::string& in_str);
71 };
72
73 // Note that, for the variable names in the trace_analyzer,
74 // Starting with 'a_' means the variable is used for 'accessed_keys'.
75 // Starting with 'w_' means it is used for 'the whole key space'.
76 // Ending with '_f' means a file write or reader pointer.
77 // For example, 'a_count' means 'accessed_keys_count',
78 // 'w_key_f' means 'whole_key_space_file'.
79
80 struct TraceStats {
81 uint32_t cf_id;
82 std::string cf_name;
83 uint64_t a_count;
84 uint64_t a_succ_count;
85 uint64_t a_key_id;
86 uint64_t a_key_size_sqsum;
87 uint64_t a_key_size_sum;
88 uint64_t a_key_mid;
89 uint64_t a_value_size_sqsum;
90 uint64_t a_value_size_sum;
91 uint64_t a_value_mid;
92 uint32_t a_peak_qps;
93 double a_ave_qps;
94 std::map<std::string, StatsUnit> a_key_stats;
95 std::map<uint64_t, uint64_t> a_count_stats;
96 std::map<uint64_t, uint64_t> a_key_size_stats;
97 std::map<uint64_t, uint64_t> a_value_size_stats;
98 std::map<uint32_t, uint32_t> a_qps_stats;
99 std::map<uint32_t, std::map<std::string, uint32_t>> a_qps_prefix_stats;
100 std::priority_queue<std::pair<uint64_t, std::string>,
101 std::vector<std::pair<uint64_t, std::string>>,
102 std::greater<std::pair<uint64_t, std::string>>>
103 top_k_queue;
104 std::priority_queue<std::pair<uint64_t, std::string>,
105 std::vector<std::pair<uint64_t, std::string>>,
106 std::greater<std::pair<uint64_t, std::string>>>
107 top_k_prefix_access;
108 std::priority_queue<std::pair<double, std::string>,
109 std::vector<std::pair<double, std::string>>,
110 std::greater<std::pair<double, std::string>>>
111 top_k_prefix_ave;
112 std::priority_queue<std::pair<uint32_t, uint32_t>,
113 std::vector<std::pair<uint32_t, uint32_t>>,
114 std::greater<std::pair<uint32_t, uint32_t>>>
115 top_k_qps_sec;
116 std::list<TraceUnit> time_series;
117 std::vector<std::pair<uint64_t, uint64_t>> correlation_output;
118
119 std::unique_ptr<rocksdb::WritableFile> time_series_f;
120 std::unique_ptr<rocksdb::WritableFile> a_key_f;
121 std::unique_ptr<rocksdb::WritableFile> a_count_dist_f;
122 std::unique_ptr<rocksdb::WritableFile> a_prefix_cut_f;
123 std::unique_ptr<rocksdb::WritableFile> a_value_size_f;
124 std::unique_ptr<rocksdb::WritableFile> a_qps_f;
125 std::unique_ptr<rocksdb::WritableFile> a_top_qps_prefix_f;
126 std::unique_ptr<rocksdb::WritableFile> w_key_f;
127 std::unique_ptr<rocksdb::WritableFile> w_prefix_cut_f;
128
129 TraceStats();
130 ~TraceStats();
131 TraceStats(const TraceStats&) = delete;
132 TraceStats& operator=(const TraceStats&) = delete;
133 TraceStats(TraceStats&&) = default;
134 TraceStats& operator=(TraceStats&&) = default;
135 };
136
137 struct TypeUnit {
138 std::string type_name;
139 bool enabled;
140 uint64_t total_keys;
141 uint64_t total_access;
142 uint64_t total_succ_access;
143 std::map<uint32_t, TraceStats> stats;
144 TypeUnit() = default;
145 ~TypeUnit() = default;
146 TypeUnit(const TypeUnit&) = delete;
147 TypeUnit& operator=(const TypeUnit&) = delete;
148 TypeUnit(TypeUnit&&) = default;
149 TypeUnit& operator=(TypeUnit&&) = default;
150 };
151
152 struct CfUnit {
153 uint32_t cf_id;
154 uint64_t w_count; // total keys in this cf if we use the whole key space
155 uint64_t a_count; // the total keys in this cf that are accessed
156 std::map<uint64_t, uint64_t> w_key_size_stats; // whole key space key size
157 // statistic this cf
158 };
159
160 class TraceAnalyzer {
161 public:
162 TraceAnalyzer(std::string& trace_path, std::string& output_path,
163 AnalyzerOptions _analyzer_opts);
164 ~TraceAnalyzer();
165
166 Status PrepareProcessing();
167
168 Status StartProcessing();
169
170 Status MakeStatistics();
171
172 Status ReProcessing();
173
174 Status EndProcessing();
175
176 Status WriteTraceUnit(TraceUnit& unit);
177
178 // The trace processing functions for different type
179 Status HandleGet(uint32_t column_family_id, const std::string& key,
180 const uint64_t& ts, const uint32_t& get_ret);
181 Status HandlePut(uint32_t column_family_id, const Slice& key,
182 const Slice& value);
183 Status HandleDelete(uint32_t column_family_id, const Slice& key);
184 Status HandleSingleDelete(uint32_t column_family_id, const Slice& key);
185 Status HandleDeleteRange(uint32_t column_family_id, const Slice& begin_key,
186 const Slice& end_key);
187 Status HandleMerge(uint32_t column_family_id, const Slice& key,
188 const Slice& value);
189 Status HandleIter(uint32_t column_family_id, const std::string& key,
190 const uint64_t& ts, TraceType& trace_type);
191 std::vector<TypeUnit>& GetTaVector() { return ta_; }
192
193 private:
194 rocksdb::Env* env_;
195 EnvOptions env_options_;
196 std::unique_ptr<TraceReader> trace_reader_;
197 size_t offset_;
198 char buffer_[1024];
199 uint64_t c_time_;
200 std::string trace_name_;
201 std::string output_path_;
202 AnalyzerOptions analyzer_opts_;
203 uint64_t total_requests_;
204 uint64_t total_access_keys_;
205 uint64_t total_gets_;
206 uint64_t total_writes_;
207 uint64_t begin_time_;
208 uint64_t end_time_;
209 uint64_t time_series_start_;
210 std::unique_ptr<rocksdb::WritableFile> trace_sequence_f_; // readable trace
211 std::unique_ptr<rocksdb::WritableFile> qps_f_; // overall qps
212 std::unique_ptr<rocksdb::SequentialFile> wkey_input_f_;
213 std::vector<TypeUnit> ta_; // The main statistic collecting data structure
214 std::map<uint32_t, CfUnit> cfs_; // All the cf_id appears in this trace;
215 std::vector<uint32_t> qps_peak_;
216 std::vector<double> qps_ave_;
217
218 Status ReadTraceHeader(Trace* header);
219 Status ReadTraceFooter(Trace* footer);
220 Status ReadTraceRecord(Trace* trace);
221 Status KeyStatsInsertion(const uint32_t& type, const uint32_t& cf_id,
222 const std::string& key, const size_t value_size,
223 const uint64_t ts);
224 Status StatsUnitCorrelationUpdate(StatsUnit& unit, const uint32_t& type,
225 const uint64_t& ts, const std::string& key);
226 Status OpenStatsOutputFiles(const std::string& type, TraceStats& new_stats);
227 Status CreateOutputFile(const std::string& type, const std::string& cf_name,
228 const std::string& ending,
229 std::unique_ptr<rocksdb::WritableFile>* f_ptr);
230 void CloseOutputFiles();
231
232 void PrintStatistics();
233 Status TraceUnitWriter(std::unique_ptr<rocksdb::WritableFile>& f_ptr,
234 TraceUnit& unit);
235 Status WriteTraceSequence(const uint32_t& type, const uint32_t& cf_id,
236 const std::string& key, const size_t value_size,
237 const uint64_t ts);
238 Status MakeStatisticKeyStatsOrPrefix(TraceStats& stats);
239 Status MakeStatisticCorrelation(TraceStats& stats, StatsUnit& unit);
240 Status MakeStatisticQPS();
241 };
242
243 // write bach handler to be used for WriteBache iterator
244 // when processing the write trace
245 class TraceWriteHandler : public WriteBatch::Handler {
246 public:
247 TraceWriteHandler() { ta_ptr = nullptr; }
248 explicit TraceWriteHandler(TraceAnalyzer* _ta_ptr) { ta_ptr = _ta_ptr; }
249 ~TraceWriteHandler() {}
250
251 virtual Status PutCF(uint32_t column_family_id, const Slice& key,
252 const Slice& value) override {
253 return ta_ptr->HandlePut(column_family_id, key, value);
254 }
255 virtual Status DeleteCF(uint32_t column_family_id,
256 const Slice& key) override {
257 return ta_ptr->HandleDelete(column_family_id, key);
258 }
259 virtual Status SingleDeleteCF(uint32_t column_family_id,
260 const Slice& key) override {
261 return ta_ptr->HandleSingleDelete(column_family_id, key);
262 }
263 virtual Status DeleteRangeCF(uint32_t column_family_id,
264 const Slice& begin_key,
265 const Slice& end_key) override {
266 return ta_ptr->HandleDeleteRange(column_family_id, begin_key, end_key);
267 }
268 virtual Status MergeCF(uint32_t column_family_id, const Slice& key,
269 const Slice& value) override {
270 return ta_ptr->HandleMerge(column_family_id, key, value);
271 }
272
273 private:
274 TraceAnalyzer* ta_ptr;
275 };
276
277 int trace_analyzer_tool(int argc, char** argv);
278
279 } // namespace rocksdb
280
281 #endif // ROCKSDB_LITE