]>
Commit | Line | Data |
---|---|---|
f67539c2 TL |
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 | #ifndef ROCKSDB_LITE | |
7 | #ifndef GFLAGS | |
8 | #include <cstdio> | |
9 | int main() { | |
10 | fprintf(stderr, | |
11 | "Please install gflags to run block_cache_trace_analyzer_test\n"); | |
12 | return 1; | |
13 | } | |
14 | #else | |
15 | ||
16 | #include <fstream> | |
17 | #include <iostream> | |
18 | #include <map> | |
19 | #include <vector> | |
20 | ||
21 | #include "rocksdb/env.h" | |
22 | #include "rocksdb/status.h" | |
23 | #include "rocksdb/trace_reader_writer.h" | |
24 | #include "test_util/testharness.h" | |
25 | #include "test_util/testutil.h" | |
26 | #include "tools/block_cache_analyzer/block_cache_trace_analyzer.h" | |
27 | #include "trace_replay/block_cache_tracer.h" | |
28 | ||
29 | namespace ROCKSDB_NAMESPACE { | |
30 | ||
31 | namespace { | |
32 | const uint64_t kBlockSize = 1024; | |
33 | const std::string kBlockKeyPrefix = "test-block-"; | |
34 | const uint32_t kCFId = 0; | |
35 | const uint32_t kLevel = 1; | |
36 | const uint64_t kSSTStoringEvenKeys = 100; | |
37 | const uint64_t kSSTStoringOddKeys = 101; | |
38 | const std::string kRefKeyPrefix = "test-get-"; | |
39 | const uint64_t kNumKeysInBlock = 1024; | |
40 | const int kMaxArgCount = 100; | |
41 | const size_t kArgBufferSize = 100000; | |
42 | } // namespace | |
43 | ||
44 | class BlockCacheTracerTest : public testing::Test { | |
45 | public: | |
46 | BlockCacheTracerTest() { | |
47 | test_path_ = test::PerThreadDBPath("block_cache_tracer_test"); | |
48 | env_ = ROCKSDB_NAMESPACE::Env::Default(); | |
49 | EXPECT_OK(env_->CreateDir(test_path_)); | |
50 | trace_file_path_ = test_path_ + "/block_cache_trace"; | |
51 | block_cache_sim_config_path_ = test_path_ + "/block_cache_sim_config"; | |
52 | timeline_labels_ = | |
53 | "block,all,cf,sst,level,bt,caller,cf_sst,cf_level,cf_bt,cf_caller"; | |
54 | reuse_distance_labels_ = | |
55 | "block,all,cf,sst,level,bt,caller,cf_sst,cf_level,cf_bt,cf_caller"; | |
56 | reuse_distance_buckets_ = "1,1K,1M,1G"; | |
57 | reuse_interval_labels_ = "block,all,cf,sst,level,bt,cf_sst,cf_level,cf_bt"; | |
58 | reuse_interval_buckets_ = "1,10,100,1000"; | |
59 | reuse_lifetime_labels_ = "block,all,cf,sst,level,bt,cf_sst,cf_level,cf_bt"; | |
60 | reuse_lifetime_buckets_ = "1,10,100,1000"; | |
61 | analyzing_callers_ = "Get,Iterator"; | |
62 | access_count_buckets_ = "2,3,4,5,10"; | |
63 | analyze_get_spatial_locality_labels_ = "all"; | |
64 | analyze_get_spatial_locality_buckets_ = "10,20,30,40,50,60,70,80,90,100"; | |
65 | } | |
66 | ||
67 | ~BlockCacheTracerTest() override { | |
68 | if (getenv("KEEP_DB")) { | |
69 | printf("The trace file is still at %s\n", trace_file_path_.c_str()); | |
70 | return; | |
71 | } | |
72 | EXPECT_OK(env_->DeleteFile(trace_file_path_)); | |
73 | EXPECT_OK(env_->DeleteDir(test_path_)); | |
74 | } | |
75 | ||
76 | TableReaderCaller GetCaller(uint32_t key_id) { | |
77 | uint32_t n = key_id % 5; | |
78 | switch (n) { | |
79 | case 0: | |
80 | return TableReaderCaller::kPrefetch; | |
81 | case 1: | |
82 | return TableReaderCaller::kCompaction; | |
83 | case 2: | |
84 | return TableReaderCaller::kUserGet; | |
85 | case 3: | |
86 | return TableReaderCaller::kUserMultiGet; | |
87 | case 4: | |
88 | return TableReaderCaller::kUserIterator; | |
89 | } | |
90 | // This cannot happend. | |
91 | assert(false); | |
92 | return TableReaderCaller::kMaxBlockCacheLookupCaller; | |
93 | } | |
94 | ||
95 | void WriteBlockAccess(BlockCacheTraceWriter* writer, uint32_t from_key_id, | |
96 | TraceType block_type, uint32_t nblocks) { | |
97 | assert(writer); | |
98 | for (uint32_t i = 0; i < nblocks; i++) { | |
99 | uint32_t key_id = from_key_id + i; | |
100 | uint64_t timestamp = (key_id + 1) * kMicrosInSecond; | |
101 | BlockCacheTraceRecord record; | |
102 | record.block_type = block_type; | |
103 | record.block_size = kBlockSize + key_id; | |
104 | record.block_key = kBlockKeyPrefix + std::to_string(key_id); | |
105 | record.access_timestamp = timestamp; | |
106 | record.cf_id = kCFId; | |
107 | record.cf_name = kDefaultColumnFamilyName; | |
108 | record.caller = GetCaller(key_id); | |
109 | record.level = kLevel; | |
110 | if (key_id % 2 == 0) { | |
111 | record.sst_fd_number = kSSTStoringEvenKeys; | |
112 | } else { | |
113 | record.sst_fd_number = kSSTStoringOddKeys; | |
114 | } | |
115 | record.is_cache_hit = Boolean::kFalse; | |
116 | record.no_insert = Boolean::kFalse; | |
117 | // Provide these fields for all block types. | |
118 | // The writer should only write these fields for data blocks and the | |
119 | // caller is either GET or MGET. | |
120 | record.referenced_key = | |
121 | kRefKeyPrefix + std::to_string(key_id) + std::string(8, 0); | |
122 | record.referenced_key_exist_in_block = Boolean::kTrue; | |
123 | record.num_keys_in_block = kNumKeysInBlock; | |
124 | ASSERT_OK(writer->WriteBlockAccess( | |
125 | record, record.block_key, record.cf_name, record.referenced_key)); | |
126 | } | |
127 | } | |
128 | ||
129 | void AssertBlockAccessInfo( | |
130 | uint32_t key_id, TraceType type, | |
131 | const std::map<std::string, BlockAccessInfo>& block_access_info_map) { | |
132 | auto key_id_str = kBlockKeyPrefix + std::to_string(key_id); | |
133 | ASSERT_TRUE(block_access_info_map.find(key_id_str) != | |
134 | block_access_info_map.end()); | |
135 | auto& block_access_info = block_access_info_map.find(key_id_str)->second; | |
136 | ASSERT_EQ(1, block_access_info.num_accesses); | |
137 | ASSERT_EQ(kBlockSize + key_id, block_access_info.block_size); | |
138 | ASSERT_GT(block_access_info.first_access_time, 0); | |
139 | ASSERT_GT(block_access_info.last_access_time, 0); | |
140 | ASSERT_EQ(1, block_access_info.caller_num_access_map.size()); | |
141 | TableReaderCaller expected_caller = GetCaller(key_id); | |
142 | ASSERT_TRUE(block_access_info.caller_num_access_map.find(expected_caller) != | |
143 | block_access_info.caller_num_access_map.end()); | |
144 | ASSERT_EQ( | |
145 | 1, | |
146 | block_access_info.caller_num_access_map.find(expected_caller)->second); | |
147 | ||
148 | if ((expected_caller == TableReaderCaller::kUserGet || | |
149 | expected_caller == TableReaderCaller::kUserMultiGet) && | |
150 | type == TraceType::kBlockTraceDataBlock) { | |
151 | ASSERT_EQ(kNumKeysInBlock, block_access_info.num_keys); | |
152 | ASSERT_EQ(1, block_access_info.key_num_access_map.size()); | |
153 | ASSERT_EQ(0, block_access_info.non_exist_key_num_access_map.size()); | |
154 | ASSERT_EQ(1, block_access_info.num_referenced_key_exist_in_block); | |
155 | } | |
156 | } | |
157 | ||
158 | void RunBlockCacheTraceAnalyzer() { | |
159 | std::vector<std::string> params = { | |
160 | "./block_cache_trace_analyzer", | |
161 | "-block_cache_trace_path=" + trace_file_path_, | |
162 | "-block_cache_sim_config_path=" + block_cache_sim_config_path_, | |
163 | "-block_cache_analysis_result_dir=" + test_path_, | |
164 | "-print_block_size_stats", | |
165 | "-print_access_count_stats", | |
166 | "-print_data_block_access_count_stats", | |
167 | "-cache_sim_warmup_seconds=0", | |
168 | "-analyze_bottom_k_access_count_blocks=5", | |
169 | "-analyze_top_k_access_count_blocks=5", | |
170 | "-analyze_blocks_reuse_k_reuse_window=5", | |
171 | "-timeline_labels=" + timeline_labels_, | |
172 | "-reuse_distance_labels=" + reuse_distance_labels_, | |
173 | "-reuse_distance_buckets=" + reuse_distance_buckets_, | |
174 | "-reuse_interval_labels=" + reuse_interval_labels_, | |
175 | "-reuse_interval_buckets=" + reuse_interval_buckets_, | |
176 | "-reuse_lifetime_labels=" + reuse_lifetime_labels_, | |
177 | "-reuse_lifetime_buckets=" + reuse_lifetime_buckets_, | |
178 | "-analyze_callers=" + analyzing_callers_, | |
179 | "-access_count_buckets=" + access_count_buckets_, | |
180 | "-analyze_get_spatial_locality_labels=" + | |
181 | analyze_get_spatial_locality_labels_, | |
182 | "-analyze_get_spatial_locality_buckets=" + | |
183 | analyze_get_spatial_locality_buckets_, | |
184 | "-analyze_correlation_coefficients_labels=all", | |
185 | "-skew_labels=all", | |
186 | "-skew_buckets=10,50,100"}; | |
187 | char arg_buffer[kArgBufferSize]; | |
188 | char* argv[kMaxArgCount]; | |
189 | int argc = 0; | |
190 | int cursor = 0; | |
191 | for (const auto& arg : params) { | |
192 | ASSERT_LE(cursor + arg.size() + 1, kArgBufferSize); | |
193 | ASSERT_LE(argc + 1, kMaxArgCount); | |
194 | snprintf(arg_buffer + cursor, arg.size() + 1, "%s", arg.c_str()); | |
195 | ||
196 | argv[argc++] = arg_buffer + cursor; | |
197 | cursor += static_cast<int>(arg.size()) + 1; | |
198 | } | |
199 | ASSERT_EQ(0, | |
200 | ROCKSDB_NAMESPACE::block_cache_trace_analyzer_tool(argc, argv)); | |
201 | } | |
202 | ||
203 | Env* env_; | |
204 | EnvOptions env_options_; | |
205 | std::string block_cache_sim_config_path_; | |
206 | std::string trace_file_path_; | |
207 | std::string test_path_; | |
208 | std::string timeline_labels_; | |
209 | std::string reuse_distance_labels_; | |
210 | std::string reuse_distance_buckets_; | |
211 | std::string reuse_interval_labels_; | |
212 | std::string reuse_interval_buckets_; | |
213 | std::string reuse_lifetime_labels_; | |
214 | std::string reuse_lifetime_buckets_; | |
215 | std::string analyzing_callers_; | |
216 | std::string access_count_buckets_; | |
217 | std::string analyze_get_spatial_locality_labels_; | |
218 | std::string analyze_get_spatial_locality_buckets_; | |
219 | }; | |
220 | ||
221 | TEST_F(BlockCacheTracerTest, BlockCacheAnalyzer) { | |
222 | { | |
223 | // Generate a trace file. | |
224 | TraceOptions trace_opt; | |
225 | std::unique_ptr<TraceWriter> trace_writer; | |
226 | ASSERT_OK(NewFileTraceWriter(env_, env_options_, trace_file_path_, | |
227 | &trace_writer)); | |
228 | BlockCacheTraceWriter writer(env_, trace_opt, std::move(trace_writer)); | |
229 | ASSERT_OK(writer.WriteHeader()); | |
230 | WriteBlockAccess(&writer, 0, TraceType::kBlockTraceDataBlock, 50); | |
231 | ASSERT_OK(env_->FileExists(trace_file_path_)); | |
232 | } | |
233 | { | |
234 | // Generate a cache sim config. | |
235 | std::string config = "lru,1,0,1K,1M,1G"; | |
236 | std::ofstream out(block_cache_sim_config_path_); | |
237 | ASSERT_TRUE(out.is_open()); | |
238 | out << config << std::endl; | |
239 | out.close(); | |
240 | } | |
241 | RunBlockCacheTraceAnalyzer(); | |
242 | { | |
243 | // Validate the cache miss ratios. | |
244 | std::vector<uint64_t> expected_capacities{1024, 1024 * 1024, | |
245 | 1024 * 1024 * 1024}; | |
246 | const std::string mrc_path = test_path_ + "/49_50_mrc"; | |
247 | std::ifstream infile(mrc_path); | |
248 | uint32_t config_index = 0; | |
249 | std::string line; | |
250 | // Read header. | |
251 | ASSERT_TRUE(getline(infile, line)); | |
252 | while (getline(infile, line)) { | |
253 | std::stringstream ss(line); | |
254 | std::vector<std::string> result_strs; | |
255 | while (ss.good()) { | |
256 | std::string substr; | |
257 | getline(ss, substr, ','); | |
258 | result_strs.push_back(substr); | |
259 | } | |
260 | ASSERT_EQ(6, result_strs.size()); | |
261 | ASSERT_LT(config_index, expected_capacities.size()); | |
262 | ASSERT_EQ("lru", result_strs[0]); // cache_name | |
263 | ASSERT_EQ("1", result_strs[1]); // num_shard_bits | |
264 | ASSERT_EQ("0", result_strs[2]); // ghost_cache_capacity | |
265 | ASSERT_EQ(std::to_string(expected_capacities[config_index]), | |
266 | result_strs[3]); // cache_capacity | |
267 | ASSERT_EQ("100.0000", result_strs[4]); // miss_ratio | |
268 | ASSERT_EQ("50", result_strs[5]); // number of accesses. | |
269 | config_index++; | |
270 | } | |
271 | ASSERT_EQ(expected_capacities.size(), config_index); | |
272 | infile.close(); | |
273 | ASSERT_OK(env_->DeleteFile(mrc_path)); | |
274 | ||
275 | const std::vector<std::string> time_units{"1", "60", "3600"}; | |
276 | expected_capacities.push_back(port::kMaxUint64); | |
277 | for (auto const& expected_capacity : expected_capacities) { | |
278 | for (auto const& time_unit : time_units) { | |
279 | const std::string miss_ratio_timeline_path = | |
280 | test_path_ + "/" + std::to_string(expected_capacity) + "_" + | |
281 | time_unit + "_miss_ratio_timeline"; | |
282 | std::ifstream mrt_file(miss_ratio_timeline_path); | |
283 | // Read header. | |
284 | ASSERT_TRUE(getline(mrt_file, line)); | |
285 | ASSERT_TRUE(getline(mrt_file, line)); | |
286 | std::stringstream ss(line); | |
287 | bool read_header = false; | |
288 | while (ss.good()) { | |
289 | std::string substr; | |
290 | getline(ss, substr, ','); | |
291 | if (!read_header) { | |
292 | if (expected_capacity == port::kMaxUint64) { | |
293 | ASSERT_EQ("trace", substr); | |
294 | } else { | |
295 | ASSERT_EQ("lru-1-0", substr); | |
296 | } | |
297 | read_header = true; | |
298 | continue; | |
299 | } | |
300 | ASSERT_DOUBLE_EQ(100.0, ParseDouble(substr)); | |
301 | } | |
302 | ASSERT_FALSE(getline(mrt_file, line)); | |
303 | mrt_file.close(); | |
304 | ASSERT_OK(env_->DeleteFile(miss_ratio_timeline_path)); | |
305 | } | |
306 | for (auto const& time_unit : time_units) { | |
307 | const std::string miss_timeline_path = | |
308 | test_path_ + "/" + std::to_string(expected_capacity) + "_" + | |
309 | time_unit + "_miss_timeline"; | |
310 | std::ifstream mt_file(miss_timeline_path); | |
311 | // Read header. | |
312 | ASSERT_TRUE(getline(mt_file, line)); | |
313 | ASSERT_TRUE(getline(mt_file, line)); | |
314 | std::stringstream ss(line); | |
315 | uint32_t num_misses = 0; | |
316 | while (ss.good()) { | |
317 | std::string substr; | |
318 | getline(ss, substr, ','); | |
319 | if (num_misses == 0) { | |
320 | if (expected_capacity == port::kMaxUint64) { | |
321 | ASSERT_EQ("trace", substr); | |
322 | } else { | |
323 | ASSERT_EQ("lru-1-0", substr); | |
324 | } | |
325 | num_misses++; | |
326 | continue; | |
327 | } | |
328 | num_misses += ParseInt(substr); | |
329 | } | |
330 | ASSERT_EQ(51, num_misses); | |
331 | ASSERT_FALSE(getline(mt_file, line)); | |
332 | mt_file.close(); | |
333 | ASSERT_OK(env_->DeleteFile(miss_timeline_path)); | |
334 | } | |
335 | } | |
336 | } | |
337 | { | |
338 | // Validate the skewness csv file. | |
339 | const std::string skewness_file_path = test_path_ + "/all_skewness"; | |
340 | std::ifstream skew_file(skewness_file_path); | |
341 | // Read header. | |
342 | std::string line; | |
343 | ASSERT_TRUE(getline(skew_file, line)); | |
344 | std::stringstream ss(line); | |
345 | double sum_percent = 0; | |
346 | while (getline(skew_file, line)) { | |
347 | std::stringstream ss_naccess(line); | |
348 | std::string substr; | |
349 | bool read_label = false; | |
350 | while (ss_naccess.good()) { | |
351 | ASSERT_TRUE(getline(ss_naccess, substr, ',')); | |
352 | if (!read_label) { | |
353 | read_label = true; | |
354 | continue; | |
355 | } | |
356 | sum_percent += ParseDouble(substr); | |
357 | } | |
358 | } | |
359 | ASSERT_EQ(100.0, sum_percent); | |
360 | ASSERT_FALSE(getline(skew_file, line)); | |
361 | skew_file.close(); | |
362 | ASSERT_OK(env_->DeleteFile(skewness_file_path)); | |
363 | } | |
364 | { | |
365 | // Validate the timeline csv files. | |
366 | const std::vector<std::string> time_units{"_60", "_3600"}; | |
367 | const std::vector<std::string> user_access_only_flags{"user_access_only_", | |
368 | "all_access_"}; | |
369 | for (auto const& user_access_only : user_access_only_flags) { | |
370 | for (auto const& unit : time_units) { | |
371 | std::stringstream ss(timeline_labels_); | |
372 | while (ss.good()) { | |
373 | std::string l; | |
374 | ASSERT_TRUE(getline(ss, l, ',')); | |
375 | if (l.find("block") == std::string::npos) { | |
376 | if (user_access_only != "all_access_") { | |
377 | continue; | |
378 | } | |
379 | } | |
380 | const std::string timeline_file = test_path_ + "/" + | |
381 | user_access_only + l + unit + | |
382 | "_access_timeline"; | |
383 | std::ifstream infile(timeline_file); | |
384 | std::string line; | |
385 | const uint64_t expected_naccesses = 50; | |
386 | const uint64_t expected_user_accesses = 30; | |
387 | ASSERT_TRUE(getline(infile, line)) << timeline_file; | |
388 | uint32_t naccesses = 0; | |
389 | while (getline(infile, line)) { | |
390 | std::stringstream ss_naccess(line); | |
391 | std::string substr; | |
392 | bool read_label = false; | |
393 | while (ss_naccess.good()) { | |
394 | ASSERT_TRUE(getline(ss_naccess, substr, ',')); | |
395 | if (!read_label) { | |
396 | read_label = true; | |
397 | continue; | |
398 | } | |
399 | naccesses += ParseUint32(substr); | |
400 | } | |
401 | } | |
402 | if (user_access_only == "user_access_only_") { | |
403 | ASSERT_EQ(expected_user_accesses, naccesses) << timeline_file; | |
404 | } else { | |
405 | ASSERT_EQ(expected_naccesses, naccesses) << timeline_file; | |
406 | } | |
407 | ASSERT_OK(env_->DeleteFile(timeline_file)); | |
408 | } | |
409 | } | |
410 | } | |
411 | } | |
412 | { | |
413 | // Validate the reuse_interval and reuse_distance csv files. | |
414 | std::map<std::string, std::string> test_reuse_csv_files; | |
415 | test_reuse_csv_files["_access_reuse_interval"] = reuse_interval_labels_; | |
416 | test_reuse_csv_files["_reuse_distance"] = reuse_distance_labels_; | |
417 | test_reuse_csv_files["_reuse_lifetime"] = reuse_lifetime_labels_; | |
418 | test_reuse_csv_files["_avg_reuse_interval"] = reuse_interval_labels_; | |
419 | test_reuse_csv_files["_avg_reuse_interval_naccesses"] = | |
420 | reuse_interval_labels_; | |
421 | for (auto const& test : test_reuse_csv_files) { | |
422 | const std::string& file_suffix = test.first; | |
423 | const std::string& labels = test.second; | |
424 | const uint32_t expected_num_rows = 5; | |
425 | std::stringstream ss(labels); | |
426 | while (ss.good()) { | |
427 | std::string l; | |
428 | ASSERT_TRUE(getline(ss, l, ',')); | |
429 | const std::string reuse_csv_file = test_path_ + "/" + l + file_suffix; | |
430 | std::ifstream infile(reuse_csv_file); | |
431 | std::string line; | |
432 | ASSERT_TRUE(getline(infile, line)); | |
433 | double npercentage = 0; | |
434 | uint32_t nrows = 0; | |
435 | while (getline(infile, line)) { | |
436 | std::stringstream ss_naccess(line); | |
437 | bool label_read = false; | |
438 | nrows++; | |
439 | while (ss_naccess.good()) { | |
440 | std::string substr; | |
441 | ASSERT_TRUE(getline(ss_naccess, substr, ',')); | |
442 | if (!label_read) { | |
443 | label_read = true; | |
444 | continue; | |
445 | } | |
446 | npercentage += ParseDouble(substr); | |
447 | } | |
448 | } | |
449 | ASSERT_EQ(expected_num_rows, nrows); | |
450 | if ("_reuse_lifetime" == test.first || | |
451 | "_avg_reuse_interval" == test.first || | |
452 | "_avg_reuse_interval_naccesses" == test.first) { | |
453 | ASSERT_EQ(100, npercentage) << reuse_csv_file; | |
454 | } else { | |
455 | ASSERT_LT(npercentage, 0); | |
456 | } | |
457 | ASSERT_OK(env_->DeleteFile(reuse_csv_file)); | |
458 | } | |
459 | } | |
460 | } | |
461 | ||
462 | { | |
463 | // Validate the percentage of accesses summary. | |
464 | const std::string percent_access_summary_file = | |
465 | test_path_ + "/percentage_of_accesses_summary"; | |
466 | std::ifstream infile(percent_access_summary_file); | |
467 | std::string line; | |
468 | ASSERT_TRUE(getline(infile, line)); | |
469 | std::set<std::string> callers; | |
470 | std::set<std::string> expected_callers{"Get", "MultiGet", "Iterator", | |
471 | "Prefetch", "Compaction"}; | |
472 | while (getline(infile, line)) { | |
473 | std::stringstream caller_percent(line); | |
474 | std::string caller; | |
475 | ASSERT_TRUE(getline(caller_percent, caller, ',')); | |
476 | std::string percent; | |
477 | ASSERT_TRUE(getline(caller_percent, percent, ',')); | |
478 | ASSERT_FALSE(caller_percent.good()); | |
479 | callers.insert(caller); | |
480 | ASSERT_EQ(20, ParseDouble(percent)); | |
481 | } | |
482 | ASSERT_EQ(expected_callers.size(), callers.size()); | |
483 | for (auto caller : callers) { | |
484 | ASSERT_TRUE(expected_callers.find(caller) != expected_callers.end()); | |
485 | } | |
486 | ASSERT_OK(env_->DeleteFile(percent_access_summary_file)); | |
487 | } | |
488 | { | |
489 | // Validate the percentage of accesses summary by analyzing callers. | |
490 | std::stringstream analyzing_callers(analyzing_callers_); | |
491 | while (analyzing_callers.good()) { | |
492 | std::string caller; | |
493 | ASSERT_TRUE(getline(analyzing_callers, caller, ',')); | |
494 | std::vector<std::string> breakdowns{"level", "bt"}; | |
495 | for (auto breakdown : breakdowns) { | |
496 | const std::string file_name = test_path_ + "/" + caller + "_" + | |
497 | breakdown + | |
498 | "_percentage_of_accesses_summary"; | |
499 | std::ifstream infile(file_name); | |
500 | std::string line; | |
501 | ASSERT_TRUE(getline(infile, line)); | |
502 | double sum = 0; | |
503 | while (getline(infile, line)) { | |
504 | std::stringstream label_percent(line); | |
505 | std::string label; | |
506 | ASSERT_TRUE(getline(label_percent, label, ',')); | |
507 | std::string percent; | |
508 | ASSERT_TRUE(getline(label_percent, percent, ',')); | |
509 | ASSERT_FALSE(label_percent.good()); | |
510 | sum += ParseDouble(percent); | |
511 | } | |
512 | ASSERT_EQ(100, sum); | |
513 | ASSERT_OK(env_->DeleteFile(file_name)); | |
514 | } | |
515 | } | |
516 | } | |
517 | const std::vector<std::string> access_types{"user_access_only", "all_access"}; | |
518 | const std::vector<std::string> prefix{"bt", "cf"}; | |
519 | for (auto const& pre : prefix) { | |
520 | for (auto const& access_type : access_types) { | |
521 | { | |
522 | // Validate the access count summary. | |
523 | const std::string bt_access_count_summary = test_path_ + "/" + pre + | |
524 | "_" + access_type + | |
525 | "_access_count_summary"; | |
526 | std::ifstream infile(bt_access_count_summary); | |
527 | std::string line; | |
528 | ASSERT_TRUE(getline(infile, line)); | |
529 | double sum_percent = 0; | |
530 | while (getline(infile, line)) { | |
531 | std::stringstream bt_percent(line); | |
532 | std::string bt; | |
533 | ASSERT_TRUE(getline(bt_percent, bt, ',')); | |
534 | std::string percent; | |
535 | ASSERT_TRUE(getline(bt_percent, percent, ',')); | |
536 | sum_percent += ParseDouble(percent); | |
537 | } | |
538 | ASSERT_EQ(100.0, sum_percent); | |
539 | ASSERT_OK(env_->DeleteFile(bt_access_count_summary)); | |
540 | } | |
541 | } | |
542 | } | |
543 | for (auto const& access_type : access_types) { | |
544 | std::vector<std::string> block_types{"Index", "Data", "Filter"}; | |
545 | for (auto block_type : block_types) { | |
546 | // Validate reuse block timeline. | |
547 | const std::string reuse_blocks_timeline = test_path_ + "/" + block_type + | |
548 | "_" + access_type + | |
549 | "_5_reuse_blocks_timeline"; | |
550 | std::ifstream infile(reuse_blocks_timeline); | |
551 | std::string line; | |
552 | ASSERT_TRUE(getline(infile, line)) << reuse_blocks_timeline; | |
553 | uint32_t index = 0; | |
554 | while (getline(infile, line)) { | |
555 | std::stringstream timeline(line); | |
556 | bool start_time = false; | |
557 | double sum = 0; | |
558 | while (timeline.good()) { | |
559 | std::string value; | |
560 | ASSERT_TRUE(getline(timeline, value, ',')); | |
561 | if (!start_time) { | |
562 | start_time = true; | |
563 | continue; | |
564 | } | |
565 | sum += ParseDouble(value); | |
566 | } | |
567 | index++; | |
568 | ASSERT_LT(sum, 100.0 * index + 1) << reuse_blocks_timeline; | |
569 | } | |
570 | ASSERT_OK(env_->DeleteFile(reuse_blocks_timeline)); | |
571 | } | |
572 | } | |
573 | ||
574 | std::stringstream ss(analyze_get_spatial_locality_labels_); | |
575 | while (ss.good()) { | |
576 | std::string l; | |
577 | ASSERT_TRUE(getline(ss, l, ',')); | |
578 | const std::vector<std::string> spatial_locality_files{ | |
579 | "_percent_ref_keys", "_percent_accesses_on_ref_keys", | |
580 | "_percent_data_size_on_ref_keys"}; | |
581 | for (auto const& spatial_locality_file : spatial_locality_files) { | |
582 | const std::string filename = test_path_ + "/" + l + spatial_locality_file; | |
583 | std::ifstream infile(filename); | |
584 | std::string line; | |
585 | ASSERT_TRUE(getline(infile, line)); | |
586 | double sum_percent = 0; | |
587 | uint32_t nrows = 0; | |
588 | while (getline(infile, line)) { | |
589 | std::stringstream bt_percent(line); | |
590 | std::string bt; | |
591 | ASSERT_TRUE(getline(bt_percent, bt, ',')); | |
592 | std::string percent; | |
593 | ASSERT_TRUE(getline(bt_percent, percent, ',')); | |
594 | sum_percent += ParseDouble(percent); | |
595 | nrows++; | |
596 | } | |
597 | ASSERT_EQ(11, nrows); | |
598 | ASSERT_EQ(100.0, sum_percent); | |
599 | ASSERT_OK(env_->DeleteFile(filename)); | |
600 | } | |
601 | } | |
602 | ASSERT_OK(env_->DeleteFile(block_cache_sim_config_path_)); | |
603 | } | |
604 | ||
605 | TEST_F(BlockCacheTracerTest, MixedBlocks) { | |
606 | { | |
607 | // Generate a trace file containing a mix of blocks. | |
608 | // It contains two SST files with 25 blocks of odd numbered block_key in | |
609 | // kSSTStoringOddKeys and 25 blocks of even numbered blocks_key in | |
610 | // kSSTStoringEvenKeys. | |
611 | TraceOptions trace_opt; | |
612 | std::unique_ptr<TraceWriter> trace_writer; | |
613 | ASSERT_OK(NewFileTraceWriter(env_, env_options_, trace_file_path_, | |
614 | &trace_writer)); | |
615 | BlockCacheTraceWriter writer(env_, trace_opt, std::move(trace_writer)); | |
616 | ASSERT_OK(writer.WriteHeader()); | |
617 | // Write blocks of different types. | |
618 | WriteBlockAccess(&writer, 0, TraceType::kBlockTraceUncompressionDictBlock, | |
619 | 10); | |
620 | WriteBlockAccess(&writer, 10, TraceType::kBlockTraceDataBlock, 10); | |
621 | WriteBlockAccess(&writer, 20, TraceType::kBlockTraceFilterBlock, 10); | |
622 | WriteBlockAccess(&writer, 30, TraceType::kBlockTraceIndexBlock, 10); | |
623 | WriteBlockAccess(&writer, 40, TraceType::kBlockTraceRangeDeletionBlock, 10); | |
624 | ASSERT_OK(env_->FileExists(trace_file_path_)); | |
625 | } | |
626 | ||
627 | { | |
628 | // Verify trace file is generated correctly. | |
629 | std::unique_ptr<TraceReader> trace_reader; | |
630 | ASSERT_OK(NewFileTraceReader(env_, env_options_, trace_file_path_, | |
631 | &trace_reader)); | |
632 | BlockCacheTraceReader reader(std::move(trace_reader)); | |
633 | BlockCacheTraceHeader header; | |
634 | ASSERT_OK(reader.ReadHeader(&header)); | |
635 | ASSERT_EQ(kMajorVersion, header.rocksdb_major_version); | |
636 | ASSERT_EQ(kMinorVersion, header.rocksdb_minor_version); | |
637 | // Read blocks. | |
638 | BlockCacheTraceAnalyzer analyzer( | |
639 | trace_file_path_, | |
640 | /*output_miss_ratio_curve_path=*/"", | |
641 | /*human_readable_trace_file_path=*/"", | |
642 | /*compute_reuse_distance=*/true, | |
643 | /*mrc_only=*/false, | |
644 | /*is_block_cache_human_readable_trace=*/false, | |
645 | /*simulator=*/nullptr); | |
646 | // The analyzer ends when it detects an incomplete access record. | |
647 | ASSERT_EQ(Status::Incomplete(""), analyzer.Analyze()); | |
648 | const uint64_t expected_num_cfs = 1; | |
649 | std::vector<uint64_t> expected_fds{kSSTStoringOddKeys, kSSTStoringEvenKeys}; | |
650 | const std::vector<TraceType> expected_types{ | |
651 | TraceType::kBlockTraceUncompressionDictBlock, | |
652 | TraceType::kBlockTraceDataBlock, TraceType::kBlockTraceFilterBlock, | |
653 | TraceType::kBlockTraceIndexBlock, | |
654 | TraceType::kBlockTraceRangeDeletionBlock}; | |
655 | const uint64_t expected_num_keys_per_type = 5; | |
656 | ||
657 | auto& stats = analyzer.TEST_cf_aggregates_map(); | |
658 | ASSERT_EQ(expected_num_cfs, stats.size()); | |
659 | ASSERT_TRUE(stats.find(kDefaultColumnFamilyName) != stats.end()); | |
660 | auto& cf_stats = stats.find(kDefaultColumnFamilyName)->second; | |
661 | ASSERT_EQ(expected_fds.size(), cf_stats.fd_aggregates_map.size()); | |
662 | for (auto fd_id : expected_fds) { | |
663 | ASSERT_TRUE(cf_stats.fd_aggregates_map.find(fd_id) != | |
664 | cf_stats.fd_aggregates_map.end()); | |
665 | ASSERT_EQ(kLevel, cf_stats.fd_aggregates_map.find(fd_id)->second.level); | |
666 | auto& block_type_aggregates_map = cf_stats.fd_aggregates_map.find(fd_id) | |
667 | ->second.block_type_aggregates_map; | |
668 | ASSERT_EQ(expected_types.size(), block_type_aggregates_map.size()); | |
669 | uint32_t key_id = 0; | |
670 | for (auto type : expected_types) { | |
671 | ASSERT_TRUE(block_type_aggregates_map.find(type) != | |
672 | block_type_aggregates_map.end()); | |
673 | auto& block_access_info_map = | |
674 | block_type_aggregates_map.find(type)->second.block_access_info_map; | |
675 | // Each block type has 5 blocks. | |
676 | ASSERT_EQ(expected_num_keys_per_type, block_access_info_map.size()); | |
677 | for (uint32_t i = 0; i < 10; i++) { | |
678 | // Verify that odd numbered blocks are stored in kSSTStoringOddKeys | |
679 | // and even numbered blocks are stored in kSSTStoringEvenKeys. | |
680 | auto key_id_str = kBlockKeyPrefix + std::to_string(key_id); | |
681 | if (fd_id == kSSTStoringOddKeys) { | |
682 | if (key_id % 2 == 1) { | |
683 | AssertBlockAccessInfo(key_id, type, block_access_info_map); | |
684 | } else { | |
685 | ASSERT_TRUE(block_access_info_map.find(key_id_str) == | |
686 | block_access_info_map.end()); | |
687 | } | |
688 | } else { | |
689 | if (key_id % 2 == 1) { | |
690 | ASSERT_TRUE(block_access_info_map.find(key_id_str) == | |
691 | block_access_info_map.end()); | |
692 | } else { | |
693 | AssertBlockAccessInfo(key_id, type, block_access_info_map); | |
694 | } | |
695 | } | |
696 | key_id++; | |
697 | } | |
698 | } | |
699 | } | |
700 | } | |
701 | } | |
702 | ||
703 | } // namespace ROCKSDB_NAMESPACE | |
704 | ||
705 | int main(int argc, char** argv) { | |
706 | ::testing::InitGoogleTest(&argc, argv); | |
707 | return RUN_ALL_TESTS(); | |
708 | } | |
709 | #endif // GFLAG | |
710 | #else | |
711 | #include <stdio.h> | |
712 | int main(int /*argc*/, char** /*argv*/) { | |
713 | fprintf(stderr, | |
714 | "block_cache_trace_analyzer_test is not supported in ROCKSDB_LITE\n"); | |
715 | return 0; | |
716 | } | |
717 | #endif // ROCKSDB_LITE |