db_stress.cc 60.1 KB
Newer Older
1 2 3 4 5
//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
//  This source code is licensed under the BSD-style license found in the
//  LICENSE file in the root directory of this source tree. An additional grant
//  of patent rights can be found in the PATENTS file in the same directory.
//
6 7 8
// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file. See the AUTHORS file for names of contributors.
9
//
X
Xing Jin 已提交
10 11 12 13 14 15 16 17 18 19 20 21
// The test uses an array to compare against values written to the database.
// Keys written to the array are in 1:1 correspondence to the actual values in
// the database according to the formula in the function GenerateValue.

// Space is reserved in the array from 0 to FLAGS_max_key and values are
// randomly written/deleted/read from those positions. During verification we
// compare all the positions in the array. To shorten/elongate the running
// time, you could change the settings: FLAGS_max_key, FLAGS_ops_per_thread,
// (sometimes also FLAGS_threads).
//
// NOTE that if FLAGS_test_batches_snapshots is set, the test will have
// different behavior. See comment of the flag for details.
A
amayank 已提交
22

23 24 25 26 27 28 29 30
#ifndef GFLAGS
#include <cstdio>
int main() {
  fprintf(stderr, "Please install gflags to run rocksdb tools\n");
  return 1;
}
#else

31 32 33
#include <sys/types.h>
#include <stdio.h>
#include <stdlib.h>
34
#include <gflags/gflags.h>
35 36
#include "db/db_impl.h"
#include "db/version_set.h"
I
Igor Canadi 已提交
37
#include "rocksdb/statistics.h"
38
#include "rocksdb/cache.h"
39
#include "rocksdb/utilities/db_ttl.h"
40 41
#include "rocksdb/env.h"
#include "rocksdb/write_batch.h"
42 43
#include "rocksdb/slice.h"
#include "rocksdb/slice_transform.h"
44
#include "port/port.h"
45
#include "util/coding.h"
46 47 48 49 50
#include "util/crc32c.h"
#include "util/histogram.h"
#include "util/mutexlock.h"
#include "util/random.h"
#include "util/testutil.h"
51
#include "util/logging.h"
52
#include "hdfs/env_hdfs.h"
D
Deon Nicholas 已提交
53
#include "utilities/merge_operators.h"
54

55 56 57
using GFLAGS::ParseCommandLineFlags;
using GFLAGS::RegisterFlagValidator;
using GFLAGS::SetUsageMessage;
58

59
static const long KB = 1024;
60

61 62
static bool ValidateUint32Range(const char* flagname, uint64_t value) {
  if (value > std::numeric_limits<uint32_t>::max()) {
K
kailiu 已提交
63 64 65 66
    fprintf(stderr,
            "Invalid value for --%s: %lu, overflow\n",
            flagname,
            (unsigned long)value);
67 68 69 70
    return false;
  }
  return true;
}
71

72
DEFINE_uint64(seed, 2341234, "Seed for PRNG");
I
Igor Canadi 已提交
73
static const bool FLAGS_seed_dummy __attribute__((unused)) =
74
    RegisterFlagValidator(&FLAGS_seed, &ValidateUint32Range);
75

I
Igor Canadi 已提交
76
DEFINE_int64(max_key, 1 * KB* KB,
77
             "Max number of key/values to place in database");
78

I
Igor Canadi 已提交
79 80
DEFINE_int32(column_families, 10, "Number of column families");

81
DEFINE_bool(test_batches_snapshots, false,
I
Igor Canadi 已提交
82
            "If set, the test uses MultiGet(), Multiut() and MultiDelete()"
83 84 85 86 87 88 89 90 91
            " which read/write/delete multiple keys in a batch. In this mode,"
            " we do not verify db content by comparing the content with the "
            "pre-allocated array. Instead, we do partial verification inside"
            " MultiGet() by checking various values in a batch. Benefit of"
            " this mode:\n"
            "\t(a) No need to acquire mutexes during writes (less cache "
            "flushes in multi-core leading to speed up)\n"
            "\t(b) No long validation at the end (more speed up)\n"
            "\t(c) Test snapshot and atomicity of batch writes");
92

93
DEFINE_int32(threads, 32, "Number of concurrent threads to run.");
94

95 96 97 98
DEFINE_int32(ttl, -1,
             "Opens the db with this ttl value if this is not -1. "
             "Carefully specify a large value such that verifications on "
             "deleted values don't fail");
99

100 101
DEFINE_int32(value_size_mult, 8,
             "Size of value will be this number times rand_int(1,3) bytes");
102

103
DEFINE_bool(verify_before_write, false, "Verify before write");
104

105
DEFINE_bool(histogram, false, "Print histogram of operation timings");
106

107 108
DEFINE_bool(destroy_db_initially, true,
            "Destroys the database dir before start if this is true");
109

110 111 112 113
DEFINE_bool(verbose, false, "Verbose");

DEFINE_bool(progress_reports, true,
            "If true, db_stress will report number of finished operations");
114

115 116
DEFINE_int32(write_buffer_size, rocksdb::Options().write_buffer_size,
             "Number of bytes to buffer in memtable before compacting");
117

118 119 120 121
DEFINE_int32(max_write_buffer_number,
             rocksdb::Options().max_write_buffer_number,
             "The number of in-memory memtables. "
             "Each memtable is of size FLAGS_write_buffer_size.");
122

123 124 125 126 127 128 129 130 131 132
DEFINE_int32(min_write_buffer_number_to_merge,
             rocksdb::Options().min_write_buffer_number_to_merge,
             "The minimum number of write buffers that will be merged together "
             "before writing to storage. This is cheap because it is an "
             "in-memory merge. If this feature is not enabled, then all these "
             "write buffers are flushed to L0 as separate files and this "
             "increases read amplification because a get request has to check "
             "in all of these files. Also, an in-memory merge may result in "
             "writing less data to storage if there are duplicate records in"
             " each of these individual write buffers.");
133

134 135 136
DEFINE_int32(open_files, rocksdb::Options().max_open_files,
             "Maximum number of files to keep open at the same time "
             "(use default if == 0)");
137

138 139 140 141
DEFINE_int64(compressed_cache_size, -1,
             "Number of bytes to use as a cache of compressed data."
             " Negative means use default settings.");

142
DEFINE_int32(compaction_style, rocksdb::Options().compaction_style, "");
143

144 145 146
DEFINE_int32(level0_file_num_compaction_trigger,
             rocksdb::Options().level0_file_num_compaction_trigger,
             "Level0 compaction start trigger");
147

148 149 150
DEFINE_int32(level0_slowdown_writes_trigger,
             rocksdb::Options().level0_slowdown_writes_trigger,
             "Number of files in level-0 that will slow down writes");
151

152 153 154
DEFINE_int32(level0_stop_writes_trigger,
             rocksdb::Options().level0_stop_writes_trigger,
             "Number of files in level-0 that will trigger put stop.");
155

156
DEFINE_int32(block_size, rocksdb::BlockBasedTableOptions().block_size,
157
             "Number of bytes in a block.");
158

159 160 161 162
DEFINE_int32(max_background_compactions,
             rocksdb::Options().max_background_compactions,
             "The maximum number of concurrent background compactions "
             "that can occur in parallel.");
163

164 165 166 167 168 169 170 171
DEFINE_int32(compaction_thread_pool_adjust_interval, 0,
             "The interval (in milliseconds) to adjust compaction thread pool "
             "size. Don't change it periodically if the value is 0.");

DEFINE_int32(compaction_thread_pool_varations, 2,
             "Range of bakground thread pool size variations when adjusted "
             "periodically.");

I
Igor Canadi 已提交
172 173 174 175
DEFINE_int32(max_background_flushes, rocksdb::Options().max_background_flushes,
             "The maximum number of concurrent background flushes "
             "that can occur in parallel.");

176 177
DEFINE_int32(universal_size_ratio, 0, "The ratio of file sizes that trigger"
             " compaction in universal style");
178

179 180
DEFINE_int32(universal_min_merge_width, 0, "The minimum number of files to "
             "compact in universal style compaction");
181

182 183
DEFINE_int32(universal_max_merge_width, 0, "The max number of files to compact"
             " in universal style compaction");
184

185 186
DEFINE_int32(universal_max_size_amplification_percent, 0,
             "The max size amplification for universal style compaction");
187

I
Igor Canadi 已提交
188 189 190 191 192
DEFINE_int32(clear_column_family_one_in, 1000000,
             "With a chance of 1/N, delete a column family and then recreate "
             "it again. If N == 0, never drop/create column families. "
             "When test_batches_snapshots is true, this flag has no effect");

193 194
DEFINE_int64(cache_size, 2 * KB * KB * KB,
             "Number of bytes to use as a cache of uncompressed data.");
195

196 197 198 199 200 201 202 203 204
static bool ValidateInt32Positive(const char* flagname, int32_t value) {
  if (value < 0) {
    fprintf(stderr, "Invalid value for --%s: %d, must be >=0\n",
            flagname, value);
    return false;
  }
  return true;
}
DEFINE_int32(reopen, 10, "Number of times database reopens");
I
Igor Canadi 已提交
205
static const bool FLAGS_reopen_dummy __attribute__((unused)) =
206
    RegisterFlagValidator(&FLAGS_reopen, &ValidateInt32Positive);
207

208 209
DEFINE_int32(bloom_bits, 10, "Bloom filter bits per key. "
             "Negative means use default settings.");
210

211
DEFINE_string(db, "", "Use the db with the following name.");
212

213 214
DEFINE_bool(verify_checksum, false,
            "Verify checksum for every block read from storage");
215

216 217
DEFINE_bool(mmap_read, rocksdb::EnvOptions().use_mmap_reads,
            "Allow reads to occur via mmap-ing files");
218

219 220 221
// Database statistics
static std::shared_ptr<rocksdb::Statistics> dbstats;
DEFINE_bool(statistics, false, "Create database statistics");
222

223
DEFINE_bool(sync, false, "Sync all writes to disk");
224

225 226
DEFINE_bool(disable_data_sync, false,
            "If true, do not wait until data is synced to disk.");
A
amayank 已提交
227

228
DEFINE_bool(use_fsync, false, "If true, issue fsync instead of fdatasync");
229

230 231 232
DEFINE_int32(kill_random_test, 0,
             "If non-zero, kill at various points in source code with "
             "probability 1/this");
I
Igor Canadi 已提交
233
static const bool FLAGS_kill_random_test_dummy __attribute__((unused)) =
234
    RegisterFlagValidator(&FLAGS_kill_random_test, &ValidateInt32Positive);
235
extern int rocksdb_kill_odds;
236

237
DEFINE_bool(disable_wal, false, "If true, do not write WAL for write.");
238

239 240
DEFINE_int32(target_file_size_base, 64 * KB,
             "Target level-1 file size for compaction");
241

242 243
DEFINE_int32(target_file_size_multiplier, 1,
             "A multiplier to compute targe level-N file size (N >= 2)");
244

245
DEFINE_uint64(max_bytes_for_level_base, 256 * KB, "Max bytes for level-1");
246

247 248
DEFINE_int32(max_bytes_for_level_multiplier, 2,
             "A multiplier to compute max bytes for level-N (N >= 2)");
249

250 251 252 253 254 255 256 257 258 259
static bool ValidateInt32Percent(const char* flagname, int32_t value) {
  if (value < 0 || value>100) {
    fprintf(stderr, "Invalid value for --%s: %d, 0<= pct <=100 \n",
            flagname, value);
    return false;
  }
  return true;
}
DEFINE_int32(readpercent, 10,
             "Ratio of reads to total workload (expressed as a percentage)");
I
Igor Canadi 已提交
260
static const bool FLAGS_readpercent_dummy __attribute__((unused)) =
261
    RegisterFlagValidator(&FLAGS_readpercent, &ValidateInt32Percent);
262 263 264 265

DEFINE_int32(prefixpercent, 20,
             "Ratio of prefix iterators to total workload (expressed as a"
             " percentage)");
I
Igor Canadi 已提交
266
static const bool FLAGS_prefixpercent_dummy __attribute__((unused)) =
267
    RegisterFlagValidator(&FLAGS_prefixpercent, &ValidateInt32Percent);
268 269 270

DEFINE_int32(writepercent, 45,
             " Ratio of deletes to total workload (expressed as a percentage)");
I
Igor Canadi 已提交
271
static const bool FLAGS_writepercent_dummy __attribute__((unused)) =
272
    RegisterFlagValidator(&FLAGS_writepercent, &ValidateInt32Percent);
273 274 275

DEFINE_int32(delpercent, 15,
             "Ratio of deletes to total workload (expressed as a percentage)");
I
Igor Canadi 已提交
276
static const bool FLAGS_delpercent_dummy __attribute__((unused)) =
277
    RegisterFlagValidator(&FLAGS_delpercent, &ValidateInt32Percent);
278 279 280

DEFINE_int32(iterpercent, 10, "Ratio of iterations to total workload"
             " (expressed as a percentage)");
I
Igor Canadi 已提交
281
static const bool FLAGS_iterpercent_dummy __attribute__((unused)) =
282
    RegisterFlagValidator(&FLAGS_iterpercent, &ValidateInt32Percent);
283 284

DEFINE_uint64(num_iterations, 10, "Number of iterations per MultiIterate run");
I
Igor Canadi 已提交
285
static const bool FLAGS_num_iterations_dummy __attribute__((unused)) =
286
    RegisterFlagValidator(&FLAGS_num_iterations, &ValidateUint32Range);
287

I
Igor Canadi 已提交
288
namespace {
289 290 291 292 293 294 295 296 297 298 299
enum rocksdb::CompressionType StringToCompressionType(const char* ctype) {
  assert(ctype);

  if (!strcasecmp(ctype, "none"))
    return rocksdb::kNoCompression;
  else if (!strcasecmp(ctype, "snappy"))
    return rocksdb::kSnappyCompression;
  else if (!strcasecmp(ctype, "zlib"))
    return rocksdb::kZlibCompression;
  else if (!strcasecmp(ctype, "bzip2"))
    return rocksdb::kBZip2Compression;
A
Albert Strasheim 已提交
300 301 302 303
  else if (!strcasecmp(ctype, "lz4"))
    return rocksdb::kLZ4Compression;
  else if (!strcasecmp(ctype, "lz4hc"))
    return rocksdb::kLZ4HCCompression;
304 305 306 307

  fprintf(stdout, "Cannot parse compression type '%s'\n", ctype);
  return rocksdb::kSnappyCompression; //default value
}
I
Igor Canadi 已提交
308 309
}  // namespace

310 311 312
DEFINE_string(compression_type, "snappy",
              "Algorithm to use to compress the database");
static enum rocksdb::CompressionType FLAGS_compression_type_e =
313
    rocksdb::kSnappyCompression;
314

315
DEFINE_string(hdfs, "", "Name of hdfs environment");
316
// posix or hdfs environment
317
static rocksdb::Env* FLAGS_env = rocksdb::Env::Default();
318

I
Igor Canadi 已提交
319 320
DEFINE_uint64(ops_per_thread, 1200000, "Number of operations per thread.");
static const bool FLAGS_ops_per_thread_dummy __attribute__((unused)) =
321
    RegisterFlagValidator(&FLAGS_ops_per_thread, &ValidateUint32Range);
322

323
DEFINE_uint64(log2_keys_per_lock, 2, "Log2 of number of keys per lock");
I
Igor Canadi 已提交
324
static const bool FLAGS_log2_keys_per_lock_dummy __attribute__((unused)) =
325
    RegisterFlagValidator(&FLAGS_log2_keys_per_lock, &ValidateUint32Range);
326

327 328 329
DEFINE_int32(purge_redundant_percent, 50,
             "Percentage of times we want to purge redundant keys in memory "
             "before flushing");
I
Igor Canadi 已提交
330
static const bool FLAGS_purge_redundant_percent_dummy __attribute__((unused)) =
331 332
    RegisterFlagValidator(&FLAGS_purge_redundant_percent,
                          &ValidateInt32Percent);
333

334 335
DEFINE_bool(filter_deletes, false, "On true, deletes use KeyMayExist to drop"
            " the delete if key not present");
336

J
Jim Paton 已提交
337 338
enum RepFactory {
  kSkipList,
I
Igor Canadi 已提交
339
  kHashSkipList,
J
Jim Paton 已提交
340 341
  kVectorRep
};
I
Igor Canadi 已提交
342 343

namespace {
344 345 346 347 348 349
enum RepFactory StringToRepFactory(const char* ctype) {
  assert(ctype);

  if (!strcasecmp(ctype, "skip_list"))
    return kSkipList;
  else if (!strcasecmp(ctype, "prefix_hash"))
I
Igor Canadi 已提交
350
    return kHashSkipList;
351 352 353 354 355 356
  else if (!strcasecmp(ctype, "vector"))
    return kVectorRep;

  fprintf(stdout, "Cannot parse memreptable %s\n", ctype);
  return kSkipList;
}
I
Igor Canadi 已提交
357 358
}  // namespace

J
Jim Paton 已提交
359
static enum RepFactory FLAGS_rep_factory;
L
Lei Jin 已提交
360
DEFINE_string(memtablerep, "prefix_hash", "");
J
Jim Paton 已提交
361

362
static bool ValidatePrefixSize(const char* flagname, int32_t value) {
L
Lei Jin 已提交
363 364
  if (value < 0 || value > 8) {
    fprintf(stderr, "Invalid value for --%s: %d. 0 <= PrefixSize <= 8\n",
365 366 367 368 369
            flagname, value);
    return false;
  }
  return true;
}
I
Igor Canadi 已提交
370
DEFINE_int32(prefix_size, 7, "Control the prefix size for HashSkipListRep");
371
static const bool FLAGS_prefix_size_dummy =
372
    RegisterFlagValidator(&FLAGS_prefix_size, &ValidatePrefixSize);
373 374 375

DEFINE_bool(use_merge, false, "On true, replaces all writes with a Merge "
            "that behaves like a Put");
J
Jim Paton 已提交
376

D
Deon Nicholas 已提交
377

378
namespace rocksdb {
379

380 381 382 383 384 385 386 387 388 389 390 391 392
// convert long to a big-endian slice key
static std::string Key(long val) {
  std::string little_endian_key;
  std::string big_endian_key;
  PutFixed64(&little_endian_key, val);
  assert(little_endian_key.size() == sizeof(val));
  big_endian_key.resize(sizeof(val));
  for (int i=0; i<(int)sizeof(val); i++) {
    big_endian_key[i] = little_endian_key[sizeof(val) - 1 - i];
  }
  return big_endian_key;
}

393 394 395 396 397 398 399 400 401 402 403
static std::string StringToHex(const std::string& str) {
  std::string result = "0x";
  char buf[10];
  for (size_t i = 0; i < str.length(); i++) {
    snprintf(buf, 10, "%02X", (unsigned char)str[i]);
    result += buf;
  }
  return result;
}


404 405 406 407 408 409 410 411 412
class StressTest;
namespace {

class Stats {
 private:
  double start_;
  double finish_;
  double seconds_;
  long done_;
413 414
  long gets_;
  long prefixes_;
415
  long writes_;
A
amayank 已提交
416
  long deletes_;
417
  long iterator_size_sums_;
418
  long founds_;
419
  long iterations_;
420
  long errors_;
421 422 423
  int next_report_;
  size_t bytes_;
  double last_op_finish_;
424
  HistogramImpl hist_;
425 426 427 428 429 430 431 432

 public:
  Stats() { }

  void Start() {
    next_report_ = 100;
    hist_.Clear();
    done_ = 0;
433 434
    gets_ = 0;
    prefixes_ = 0;
435
    writes_ = 0;
A
amayank 已提交
436
    deletes_ = 0;
437
    iterator_size_sums_ = 0;
438
    founds_ = 0;
439
    iterations_ = 0;
440
    errors_ = 0;
441 442 443 444 445 446 447 448 449 450
    bytes_ = 0;
    seconds_ = 0;
    start_ = FLAGS_env->NowMicros();
    last_op_finish_ = start_;
    finish_ = start_;
  }

  void Merge(const Stats& other) {
    hist_.Merge(other.hist_);
    done_ += other.done_;
451 452
    gets_ += other.gets_;
    prefixes_ += other.prefixes_;
453
    writes_ += other.writes_;
A
amayank 已提交
454
    deletes_ += other.deletes_;
455
    iterator_size_sums_ += other.iterator_size_sums_;
456
    founds_ += other.founds_;
457
    iterations_ += other.iterations_;
458
    errors_ += other.errors_;
459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475
    bytes_ += other.bytes_;
    seconds_ += other.seconds_;
    if (other.start_ < start_) start_ = other.start_;
    if (other.finish_ > finish_) finish_ = other.finish_;
  }

  void Stop() {
    finish_ = FLAGS_env->NowMicros();
    seconds_ = (finish_ - start_) * 1e-6;
  }

  void FinishedSingleOp() {
    if (FLAGS_histogram) {
      double now = FLAGS_env->NowMicros();
      double micros = now - last_op_finish_;
      hist_.Add(micros);
      if (micros > 20000) {
476
        fprintf(stdout, "long op: %.1f micros%30s\r", micros, "");
477 478 479 480
      }
      last_op_finish_ = now;
    }

481
      done_++;
482
    if (FLAGS_progress_reports) {
483 484 485 486 487 488 489 490 491 492
      if (done_ >= next_report_) {
        if      (next_report_ < 1000)   next_report_ += 100;
        else if (next_report_ < 5000)   next_report_ += 500;
        else if (next_report_ < 10000)  next_report_ += 1000;
        else if (next_report_ < 50000)  next_report_ += 5000;
        else if (next_report_ < 100000) next_report_ += 10000;
        else if (next_report_ < 500000) next_report_ += 50000;
        else                            next_report_ += 100000;
        fprintf(stdout, "... finished %ld ops%30s\r", done_, "");
      }
493 494 495
    }
  }

496 497 498 499 500
  void AddBytesForWrites(int nwrites, size_t nbytes) {
    writes_ += nwrites;
    bytes_ += nbytes;
  }

501 502 503
  void AddGets(int ngets, int nfounds) {
    founds_ += nfounds;
    gets_ += ngets;
504 505
  }

506 507 508 509 510
  void AddPrefixes(int nprefixes, int count) {
    prefixes_ += nprefixes;
    iterator_size_sums_ += count;
  }

511 512 513 514
  void AddIterations(int n) {
    iterations_ += n;
  }

515 516 517 518
  void AddDeletes(int n) {
    deletes_ += n;
  }

519 520
  void AddErrors(int n) {
    errors_ += n;
A
amayank 已提交
521 522
  }

523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539
  void Report(const char* name) {
    std::string extra;
    if (bytes_ < 1 || done_ < 1) {
      fprintf(stderr, "No writes or ops?\n");
      return;
    }

    double elapsed = (finish_ - start_) * 1e-6;
    double bytes_mb = bytes_ / 1048576.0;
    double rate = bytes_mb / elapsed;
    double throughput = (double)done_/elapsed;

    fprintf(stdout, "%-12s: ", name);
    fprintf(stdout, "%.3f micros/op %ld ops/sec\n",
            seconds_ * 1e6 / done_, (long)throughput);
    fprintf(stdout, "%-12s: Wrote %.2f MB (%.2f MB/sec) (%ld%% of %ld ops)\n",
            "", bytes_mb, rate, (100*writes_)/done_, done_);
540
    fprintf(stdout, "%-12s: Wrote %ld times\n", "", writes_);
A
amayank 已提交
541
    fprintf(stdout, "%-12s: Deleted %ld times\n", "", deletes_);
X
Xing Jin 已提交
542 543
    fprintf(stdout, "%-12s: %ld read and %ld found the key\n", "",
            gets_, founds_);
544 545 546
    fprintf(stdout, "%-12s: Prefix scanned %ld times\n", "", prefixes_);
    fprintf(stdout, "%-12s: Iterator size sum is %ld\n", "",
            iterator_size_sums_);
547
    fprintf(stdout, "%-12s: Iterated %ld times\n", "", iterations_);
548
    fprintf(stdout, "%-12s: Got errors %ld times\n", "", errors_);
549 550 551 552 553 554 555 556 557 558 559

    if (FLAGS_histogram) {
      fprintf(stdout, "Microseconds per op:\n%s\n", hist_.ToString().c_str());
    }
    fflush(stdout);
  }
};

// State shared by all concurrent executions of the same benchmark.
class SharedState {
 public:
560
  static const uint32_t SENTINEL;
561

562 563 564 565 566 567 568 569 570 571 572 573
  explicit SharedState(StressTest* stress_test)
      : cv_(&mu_),
        seed_(FLAGS_seed),
        max_key_(FLAGS_max_key),
        log2_keys_per_lock_(FLAGS_log2_keys_per_lock),
        num_threads_(FLAGS_threads),
        num_initialized_(0),
        num_populated_(0),
        vote_reopen_(0),
        num_done_(0),
        start_(false),
        start_verify_(false),
574 575
        should_stop_bg_thread_(false),
        bg_thread_finished_(false),
576 577
        stress_test_(stress_test),
        verification_failure_(false) {
578 579 580 581
    if (FLAGS_test_batches_snapshots) {
      fprintf(stdout, "No lock creation because test_batches_snapshots set\n");
      return;
    }
I
Igor Canadi 已提交
582 583 584 585
    values_.resize(FLAGS_column_families);

    for (int i = 0; i < FLAGS_column_families; ++i) {
      values_[i] = std::vector<uint32_t>(max_key_, SENTINEL);
586
    }
587

588 589
    long num_locks = (max_key_ >> log2_keys_per_lock_);
    if (max_key_ & ((1 << log2_keys_per_lock_) - 1)) {
I
Igor Canadi 已提交
590 591 592 593 594 595
      num_locks++;
    }
    fprintf(stdout, "Creating %ld locks\n", num_locks * FLAGS_column_families);
    key_locks_.resize(FLAGS_column_families);
    for (int i = 0; i < FLAGS_column_families; ++i) {
      key_locks_[i] = std::vector<port::Mutex>(num_locks);
596 597 598
    }
  }

I
Igor Canadi 已提交
599
  ~SharedState() {}
600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624

  port::Mutex* GetMutex() {
    return &mu_;
  }

  port::CondVar* GetCondVar() {
    return &cv_;
  }

  StressTest* GetStressTest() const {
    return stress_test_;
  }

  long GetMaxKey() const {
    return max_key_;
  }

  uint32_t GetNumThreads() const {
    return num_threads_;
  }

  void IncInitialized() {
    num_initialized_++;
  }

A
amayank 已提交
625
  void IncOperated() {
626 627 628 629 630 631 632
    num_populated_++;
  }

  void IncDone() {
    num_done_++;
  }

633 634 635 636
  void IncVotedReopen() {
    vote_reopen_ = (vote_reopen_ + 1) % num_threads_;
  }

637 638 639 640
  bool AllInitialized() const {
    return num_initialized_ >= num_threads_;
  }

A
amayank 已提交
641
  bool AllOperated() const {
642 643 644 645 646 647 648
    return num_populated_ >= num_threads_;
  }

  bool AllDone() const {
    return num_done_ >= num_threads_;
  }

649 650 651 652
  bool AllVotedReopen() {
    return (vote_reopen_ == 0);
  }

653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668
  void SetStart() {
    start_ = true;
  }

  void SetStartVerify() {
    start_verify_ = true;
  }

  bool Started() const {
    return start_;
  }

  bool VerifyStarted() const {
    return start_verify_;
  }

I
Igor Canadi 已提交
669 670 671 672
  void SetVerificationFailure() { verification_failure_.store(true); }

  bool HasVerificationFailedYet() { return verification_failure_.load(); }

I
Igor Canadi 已提交
673 674
  port::Mutex* GetMutexForKey(int cf, long key) {
    return &key_locks_[cf][key >> log2_keys_per_lock_];
675 676
  }

I
Igor Canadi 已提交
677 678 679 680
  void LockColumnFamily(int cf) {
    for (auto& mutex : key_locks_[cf]) {
      mutex.Lock();
    }
681 682
  }

I
Igor Canadi 已提交
683 684 685 686
  void UnlockColumnFamily(int cf) {
    for (auto& mutex : key_locks_[cf]) {
      mutex.Unlock();
    }
687 688
  }

I
Igor Canadi 已提交
689 690
  void ClearColumnFamily(int cf) {
    std::fill(values_[cf].begin(), values_[cf].end(), SENTINEL);
A
amayank 已提交
691 692
  }

I
Igor Canadi 已提交
693 694
  void Put(int cf, long key, uint32_t value_base) {
    values_[cf][key] = value_base;
695 696
  }

I
Igor Canadi 已提交
697 698 699 700 701 702
  uint32_t Get(int cf, long key) const { return values_[cf][key]; }

  void Delete(int cf, long key) { values_[cf][key] = SENTINEL; }

  uint32_t GetSeed() const { return seed_; }

703 704 705 706 707 708 709 710
  void SetShouldStopBgThread() { should_stop_bg_thread_ = true; }

  bool ShoudStopBgThread() { return should_stop_bg_thread_; }

  void SetBgThreadFinish() { bg_thread_finished_ = true; }

  bool BgThreadFinished() const { return bg_thread_finished_; }

711 712 713 714 715 716 717 718 719
 private:
  port::Mutex mu_;
  port::CondVar cv_;
  const uint32_t seed_;
  const long max_key_;
  const uint32_t log2_keys_per_lock_;
  const int num_threads_;
  long num_initialized_;
  long num_populated_;
720
  long vote_reopen_;
721 722 723
  long num_done_;
  bool start_;
  bool start_verify_;
724 725
  bool should_stop_bg_thread_;
  bool bg_thread_finished_;
726
  StressTest* stress_test_;
I
Igor Canadi 已提交
727
  std::atomic<bool> verification_failure_;
728

I
Igor Canadi 已提交
729 730
  std::vector<std::vector<uint32_t>> values_;
  std::vector<std::vector<port::Mutex>> key_locks_;
731 732
};

733 734
const uint32_t SharedState::SENTINEL = 0xffffffff;

735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754
// Per-thread state for concurrent executions of the same benchmark.
struct ThreadState {
  uint32_t tid; // 0..n-1
  Random rand;  // Has different seeds for different threads
  SharedState* shared;
  Stats stats;

  ThreadState(uint32_t index, SharedState *shared)
      : tid(index),
        rand(1000 + index + shared->GetSeed()),
        shared(shared) {
  }
};

}  // namespace

class StressTest {
 public:
  StressTest()
      : cache_(NewLRUCache(FLAGS_cache_size)),
I
Igor Canadi 已提交
755 756 757
        compressed_cache_(FLAGS_compressed_cache_size >= 0
                              ? NewLRUCache(FLAGS_compressed_cache_size)
                              : nullptr),
758
        filter_policy_(FLAGS_bloom_bits >= 0
I
Igor Canadi 已提交
759 760
                           ? NewBloomFilterPolicy(FLAGS_bloom_bits)
                           : nullptr),
761
        db_(nullptr),
I
Igor Canadi 已提交
762
        new_column_family_name_(0),
763
        num_times_reopened_(0) {
764 765 766 767 768
    if (FLAGS_destroy_db_initially) {
      std::vector<std::string> files;
      FLAGS_env->GetChildren(FLAGS_db, &files);
      for (unsigned int i = 0; i < files.size(); i++) {
        if (Slice(files[i]).starts_with("heap-")) {
769
          FLAGS_env->DeleteFile(FLAGS_db + "/" + files[i]);
770
        }
771
      }
772
      DestroyDB(FLAGS_db, Options());
773 774 775 776
    }
  }

  ~StressTest() {
I
Igor Canadi 已提交
777 778 779 780
    for (auto cf : column_families_) {
      delete cf;
    }
    column_families_.clear();
781 782 783
    delete db_;
  }

I
Igor Canadi 已提交
784
  bool Run() {
785 786 787 788 789 790 791 792 793 794
    PrintEnv();
    Open();
    SharedState shared(this);
    uint32_t n = shared.GetNumThreads();

    std::vector<ThreadState*> threads(n);
    for (uint32_t i = 0; i < n; i++) {
      threads[i] = new ThreadState(i, &shared);
      FLAGS_env->StartThread(ThreadBody, threads[i]);
    }
795 796 797 798 799
    ThreadState bg_thread(0, &shared);
    if (FLAGS_compaction_thread_pool_adjust_interval > 0) {
      FLAGS_env->StartThread(PoolSizeChangeThread, &bg_thread);
    }

800
    // Each thread goes through the following states:
A
amayank 已提交
801 802
    // initializing -> wait for others to init -> read/populate/depopulate
    // wait for others to operate -> verify -> done
803 804 805 806 807 808 809

    {
      MutexLock l(shared.GetMutex());
      while (!shared.AllInitialized()) {
        shared.GetCondVar()->Wait();
      }

810 811 812
      double now = FLAGS_env->NowMicros();
      fprintf(stdout, "%s Starting database operations\n",
              FLAGS_env->TimeToString((uint64_t) now/1000000).c_str());
A
amayank 已提交
813

814 815
      shared.SetStart();
      shared.GetCondVar()->SignalAll();
A
amayank 已提交
816
      while (!shared.AllOperated()) {
817 818 819
        shared.GetCondVar()->Wait();
      }

820
      now = FLAGS_env->NowMicros();
821 822 823 824 825 826 827
      if (FLAGS_test_batches_snapshots) {
        fprintf(stdout, "%s Limited verification already done during gets\n",
                FLAGS_env->TimeToString((uint64_t) now/1000000).c_str());
      } else {
        fprintf(stdout, "%s Starting verification\n",
                FLAGS_env->TimeToString((uint64_t) now/1000000).c_str());
      }
828

829 830 831 832 833 834 835
      shared.SetStartVerify();
      shared.GetCondVar()->SignalAll();
      while (!shared.AllDone()) {
        shared.GetCondVar()->Wait();
      }
    }

836
    for (unsigned int i = 1; i < n; i++) {
837 838 839 840
      threads[0]->stats.Merge(threads[i]->stats);
    }
    threads[0]->stats.Report("Stress Test");

841
    for (unsigned int i = 0; i < n; i++) {
842
      delete threads[i];
843
      threads[i] = nullptr;
844
    }
845
    double now = FLAGS_env->NowMicros();
846 847 848 849
    if (!FLAGS_test_batches_snapshots) {
      fprintf(stdout, "%s Verification successful\n",
              FLAGS_env->TimeToString((uint64_t) now/1000000).c_str());
    }
850
    PrintStatistics();
I
Igor Canadi 已提交
851

852 853 854 855 856 857 858 859
    if (FLAGS_compaction_thread_pool_adjust_interval > 0) {
      MutexLock l(shared.GetMutex());
      shared.SetShouldStopBgThread();
      while (!shared.BgThreadFinished()) {
        shared.GetCondVar()->Wait();
      }
    }

I
Igor Canadi 已提交
860 861 862 863 864
    if (shared.HasVerificationFailedYet()) {
      printf("Verification failed :(\n");
      return false;
    }
    return true;
865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882
  }

 private:

  static void ThreadBody(void* v) {
    ThreadState* thread = reinterpret_cast<ThreadState*>(v);
    SharedState* shared = thread->shared;

    {
      MutexLock l(shared->GetMutex());
      shared->IncInitialized();
      if (shared->AllInitialized()) {
        shared->GetCondVar()->SignalAll();
      }
      while (!shared->Started()) {
        shared->GetCondVar()->Wait();
      }
    }
A
amayank 已提交
883
    thread->shared->GetStressTest()->OperateDb(thread);
884 885 886

    {
      MutexLock l(shared->GetMutex());
A
amayank 已提交
887 888
      shared->IncOperated();
      if (shared->AllOperated()) {
889 890 891 892 893 894 895
        shared->GetCondVar()->SignalAll();
      }
      while (!shared->VerifyStarted()) {
        shared->GetCondVar()->Wait();
      }
    }

896
    if (!FLAGS_test_batches_snapshots) {
M
Mayank Agarwal 已提交
897
      thread->shared->GetStressTest()->VerifyDb(thread);
898
    }
899 900 901 902 903 904 905 906 907 908 909

    {
      MutexLock l(shared->GetMutex());
      shared->IncDone();
      if (shared->AllDone()) {
        shared->GetCondVar()->SignalAll();
      }
    }

  }

910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941
  static void PoolSizeChangeThread(void* v) {
    assert(FLAGS_compaction_thread_pool_adjust_interval > 0);
    ThreadState* thread = reinterpret_cast<ThreadState*>(v);
    SharedState* shared = thread->shared;

    while (true) {
      {
        MutexLock l(shared->GetMutex());
        if (shared->ShoudStopBgThread()) {
          shared->SetBgThreadFinish();
          shared->GetCondVar()->SignalAll();
          return;
        }
      }

      auto thread_pool_size_base = FLAGS_max_background_compactions;
      auto thread_pool_size_var = FLAGS_compaction_thread_pool_varations;
      int new_thread_pool_size =
          thread_pool_size_base - thread_pool_size_var +
          thread->rand.Next() % (thread_pool_size_var * 2 + 1);
      if (new_thread_pool_size < 1) {
        new_thread_pool_size = 1;
      }
      FLAGS_env->SetBackgroundThreads(new_thread_pool_size);
      // Sleep up to 3 seconds
      FLAGS_env->SleepForMicroseconds(
          thread->rand.Next() % FLAGS_compaction_thread_pool_adjust_interval *
              1000 +
          1);
    }
  }

942 943 944
  // Given a key K and value V, this puts ("0"+K, "0"+V), ("1"+K, "1"+V), ...
  // ("9"+K, "9"+V) in DB atomically i.e in a single batch.
  // Also refer MultiGet.
I
Igor Canadi 已提交
945 946 947
  Status MultiPut(ThreadState* thread, const WriteOptions& writeoptions,
                  ColumnFamilyHandle* column_family, const Slice& key,
                  const Slice& value, size_t sz) {
948 949 950 951 952 953 954 955 956 957 958
    std::string keys[10] = {"9", "8", "7", "6", "5",
                            "4", "3", "2", "1", "0"};
    std::string values[10] = {"9", "8", "7", "6", "5",
                              "4", "3", "2", "1", "0"};
    Slice value_slices[10];
    WriteBatch batch;
    Status s;
    for (int i = 0; i < 10; i++) {
      keys[i] += key.ToString();
      values[i] += value.ToString();
      value_slices[i] = values[i];
959
      if (FLAGS_use_merge) {
I
Igor Canadi 已提交
960
        batch.Merge(column_family, keys[i], value_slices[i]);
D
Deon Nicholas 已提交
961
      } else {
I
Igor Canadi 已提交
962
        batch.Put(column_family, keys[i], value_slices[i]);
D
Deon Nicholas 已提交
963
      }
964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979
    }

    s = db_->Write(writeoptions, &batch);
    if (!s.ok()) {
      fprintf(stderr, "multiput error: %s\n", s.ToString().c_str());
      thread->stats.AddErrors(1);
    } else {
      // we did 10 writes each of size sz + 1
      thread->stats.AddBytesForWrites(10, (sz + 1) * 10);
    }

    return s;
  }

  // Given a key K, this deletes ("0"+K), ("1"+K),... ("9"+K)
  // in DB atomically i.e in a single batch. Also refer MultiGet.
I
Igor Canadi 已提交
980 981
  Status MultiDelete(ThreadState* thread, const WriteOptions& writeoptions,
                     ColumnFamilyHandle* column_family, const Slice& key) {
982 983 984 985 986 987 988
    std::string keys[10] = {"9", "7", "5", "3", "1",
                            "8", "6", "4", "2", "0"};

    WriteBatch batch;
    Status s;
    for (int i = 0; i < 10; i++) {
      keys[i] += key.ToString();
I
Igor Canadi 已提交
989
      batch.Delete(column_family, keys[i]);
990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006
    }

    s = db_->Write(writeoptions, &batch);
    if (!s.ok()) {
      fprintf(stderr, "multidelete error: %s\n", s.ToString().c_str());
      thread->stats.AddErrors(1);
    } else {
      thread->stats.AddDeletes(10);
    }

    return s;
  }

  // Given a key K, this gets values for "0"+K, "1"+K,..."9"+K
  // in the same snapshot, and verifies that all the values are of the form
  // "0"+V, "1"+V,..."9"+V.
  // ASSUMES that MultiPut was used to put (K, V) into the DB.
I
Igor Canadi 已提交
1007 1008 1009
  Status MultiGet(ThreadState* thread, const ReadOptions& readoptions,
                  ColumnFamilyHandle* column_family, const Slice& key,
                  std::string* value) {
1010 1011 1012 1013 1014 1015 1016 1017 1018
    std::string keys[10] = {"0", "1", "2", "3", "4", "5", "6", "7", "8", "9"};
    Slice key_slices[10];
    std::string values[10];
    ReadOptions readoptionscopy = readoptions;
    readoptionscopy.snapshot = db_->GetSnapshot();
    Status s;
    for (int i = 0; i < 10; i++) {
      keys[i] += key.ToString();
      key_slices[i] = keys[i];
I
Igor Canadi 已提交
1019
      s = db_->Get(readoptionscopy, column_family, key_slices[i], value);
1020 1021 1022 1023 1024 1025 1026 1027
      if (!s.ok() && !s.IsNotFound()) {
        fprintf(stderr, "get error: %s\n", s.ToString().c_str());
        values[i] = "";
        thread->stats.AddErrors(1);
        // we continue after error rather than exiting so that we can
        // find more errors if any
      } else if (s.IsNotFound()) {
        values[i] = "";
1028
        thread->stats.AddGets(1, 0);
1029 1030 1031 1032 1033 1034
      } else {
        values[i] = *value;

        char expected_prefix = (keys[i])[0];
        char actual_prefix = (values[i])[0];
        if (actual_prefix != expected_prefix) {
1035
          fprintf(stderr, "error expected prefix = %c actual = %c\n",
1036 1037 1038
                  expected_prefix, actual_prefix);
        }
        (values[i])[0] = ' '; // blank out the differing character
1039
        thread->stats.AddGets(1, 1);
1040 1041 1042 1043 1044 1045 1046
      }
    }
    db_->ReleaseSnapshot(readoptionscopy.snapshot);

    // Now that we retrieved all values, check that they all match
    for (int i = 1; i < 10; i++) {
      if (values[i] != values[0]) {
1047
        fprintf(stderr, "error : inconsistent values for key %s: %s, %s\n",
1048 1049
                key.ToString(true).c_str(), StringToHex(values[0]).c_str(),
                StringToHex(values[i]).c_str());
1050 1051 1052 1053 1054 1055 1056 1057
      // we continue after error rather than exiting so that we can
      // find more errors if any
      }
    }

    return s;
  }

L
Lei Jin 已提交
1058 1059 1060 1061 1062
  // Given a key, this does prefix scans for "0"+P, "1"+P,..."9"+P
  // in the same snapshot where P is the first FLAGS_prefix_size - 1 bytes
  // of the key. Each of these 10 scans returns a series of values;
  // each series should be the same length, and it is verified for each
  // index i that all the i'th values are of the form "0"+V, "1"+V,..."9"+V.
1063
  // ASSUMES that MultiPut was used to put (K, V)
I
Igor Canadi 已提交
1064 1065
  Status MultiPrefixScan(ThreadState* thread, const ReadOptions& readoptions,
                         ColumnFamilyHandle* column_family,
L
Lei Jin 已提交
1066
                         const Slice& key) {
1067 1068 1069 1070 1071 1072 1073 1074
    std::string prefixes[10] = {"0", "1", "2", "3", "4",
                                "5", "6", "7", "8", "9"};
    Slice prefix_slices[10];
    ReadOptions readoptionscopy[10];
    const Snapshot* snapshot = db_->GetSnapshot();
    Iterator* iters[10];
    Status s = Status::OK();
    for (int i = 0; i < 10; i++) {
L
Lei Jin 已提交
1075 1076 1077
      prefixes[i] += key.ToString();
      prefixes[i].resize(FLAGS_prefix_size);
      prefix_slices[i] = Slice(prefixes[i]);
1078 1079
      readoptionscopy[i] = readoptions;
      readoptionscopy[i].snapshot = snapshot;
I
Igor Canadi 已提交
1080
      iters[i] = db_->NewIterator(readoptionscopy[i], column_family);
I
Igor Canadi 已提交
1081
      iters[i]->Seek(prefix_slices[i]);
1082 1083 1084
    }

    int count = 0;
1085
    while (iters[0]->Valid() && iters[0]->key().starts_with(prefix_slices[0])) {
1086 1087 1088 1089 1090
      count++;
      std::string values[10];
      // get list of all values for this iteration
      for (int i = 0; i < 10; i++) {
        // no iterator should finish before the first one
1091 1092
        assert(iters[i]->Valid() &&
               iters[i]->key().starts_with(prefix_slices[i]));
1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106
        values[i] = iters[i]->value().ToString();

        char expected_first = (prefixes[i])[0];
        char actual_first = (values[i])[0];

        if (actual_first != expected_first) {
          fprintf(stderr, "error expected first = %c actual = %c\n",
                  expected_first, actual_first);
        }
        (values[i])[0] = ' '; // blank out the differing character
      }
      // make sure all values are equivalent
      for (int i = 0; i < 10; i++) {
        if (values[i] != values[0]) {
1107 1108 1109
          fprintf(stderr, "error : %d, inconsistent values for prefix %s: %s, %s\n",
                  i, prefixes[i].c_str(), StringToHex(values[0]).c_str(),
                  StringToHex(values[i]).c_str());
1110 1111 1112 1113 1114 1115 1116 1117 1118 1119
          // we continue after error rather than exiting so that we can
          // find more errors if any
        }
        iters[i]->Next();
      }
    }

    // cleanup iterators and snapshot
    for (int i = 0; i < 10; i++) {
      // if the first iterator finished, they should have all finished
1120 1121
      assert(!iters[i]->Valid() ||
             !iters[i]->key().starts_with(prefix_slices[i]));
1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135
      assert(iters[i]->status().ok());
      delete iters[i];
    }
    db_->ReleaseSnapshot(snapshot);

    if (s.ok()) {
      thread->stats.AddPrefixes(1, count);
    } else {
      thread->stats.AddErrors(1);
    }

    return s;
  }

1136 1137
  // Given a key K, this creates an iterator which scans to K and then
  // does a random sequence of Next/Prev operations.
I
Igor Canadi 已提交
1138 1139
  Status MultiIterate(ThreadState* thread, const ReadOptions& readoptions,
                      ColumnFamilyHandle* column_family, const Slice& key) {
1140 1141 1142 1143
    Status s;
    const Snapshot* snapshot = db_->GetSnapshot();
    ReadOptions readoptionscopy = readoptions;
    readoptionscopy.snapshot = snapshot;
I
Igor Canadi 已提交
1144
    unique_ptr<Iterator> iter(db_->NewIterator(readoptionscopy, column_family));
1145 1146

    iter->Seek(key);
1147
    for (uint64_t i = 0; i < FLAGS_num_iterations && iter->Valid(); i++) {
1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165
      if (thread->rand.OneIn(2)) {
        iter->Next();
      } else {
        iter->Prev();
      }
    }

    if (s.ok()) {
      thread->stats.AddIterations(1);
    } else {
      thread->stats.AddErrors(1);
    }

    db_->ReleaseSnapshot(snapshot);

    return s;
  }

A
amayank 已提交
1166
  void OperateDb(ThreadState* thread) {
1167 1168
    ReadOptions read_opts(FLAGS_verify_checksum, true);
    WriteOptions write_opts;
A
amayank 已提交
1169
    char value[100];
1170 1171 1172 1173 1174 1175
    long max_key = thread->shared->GetMaxKey();
    std::string from_db;
    if (FLAGS_sync) {
      write_opts.sync = true;
    }
    write_opts.disableWAL = FLAGS_disable_wal;
1176 1177 1178
    const int prefixBound = (int)FLAGS_readpercent + (int)FLAGS_prefixpercent;
    const int writeBound = prefixBound + (int)FLAGS_writepercent;
    const int delBound = writeBound + (int)FLAGS_delpercent;
1179 1180

    thread->stats.Start();
1181
    for (uint64_t i = 0; i < FLAGS_ops_per_thread; i++) {
I
Igor Canadi 已提交
1182 1183 1184 1185
      if (thread->shared->HasVerificationFailedYet()) {
        break;
      }
      if (i != 0 && (i % (FLAGS_ops_per_thread / (FLAGS_reopen + 1))) == 0) {
1186
        {
1187
          thread->stats.FinishedSingleOp();
1188 1189 1190 1191 1192 1193 1194 1195 1196
          MutexLock l(thread->shared->GetMutex());
          thread->shared->IncVotedReopen();
          if (thread->shared->AllVotedReopen()) {
            thread->shared->GetStressTest()->Reopen();
            thread->shared->GetCondVar()->SignalAll();
          }
          else {
            thread->shared->GetCondVar()->Wait();
          }
1197 1198
          // Commenting this out as we don't want to reset stats on each open.
          // thread->stats.Start();
1199 1200
        }
      }
1201

I
Igor Canadi 已提交
1202
      if (!FLAGS_test_batches_snapshots &&
1203
          FLAGS_clear_column_family_one_in != 0 && FLAGS_column_families > 1) {
I
Igor Canadi 已提交
1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229
        if (thread->rand.OneIn(FLAGS_clear_column_family_one_in)) {
          // drop column family and then create it again (can't drop default)
          int cf = thread->rand.Next() % (FLAGS_column_families - 1) + 1;
          std::string new_name =
              std::to_string(new_column_family_name_.fetch_add(1));
          {
            MutexLock l(thread->shared->GetMutex());
            fprintf(
                stdout,
                "[CF %d] Dropping and recreating column family. new name: %s\n",
                cf, new_name.c_str());
          }
          thread->shared->LockColumnFamily(cf);
          Status s __attribute__((unused));
          s = db_->DropColumnFamily(column_families_[cf]);
          delete column_families_[cf];
          assert(s.ok());
          s = db_->CreateColumnFamily(ColumnFamilyOptions(options_), new_name,
                                      &column_families_[cf]);
          column_family_names_[cf] = new_name;
          thread->shared->ClearColumnFamily(cf);
          assert(s.ok());
          thread->shared->UnlockColumnFamily(cf);
        }
      }

1230
      long rand_key = thread->rand.Next() % max_key;
I
Igor Canadi 已提交
1231
      int rand_column_family = thread->rand.Next() % FLAGS_column_families;
1232 1233 1234
      std::string keystr = Key(rand_key);
      Slice key = keystr;
      int prob_op = thread->rand.Uniform(100);
I
Igor Canadi 已提交
1235 1236 1237 1238 1239 1240
      std::unique_ptr<MutexLock> l;
      if (!FLAGS_test_batches_snapshots) {
        l.reset(new MutexLock(
            thread->shared->GetMutexForKey(rand_column_family, rand_key)));
      }
      auto column_family = column_families_[rand_column_family];
1241 1242

      if (prob_op >= 0 && prob_op < (int)FLAGS_readpercent) {
1243
        // OPERATION read
1244
        if (!FLAGS_test_batches_snapshots) {
I
Igor Canadi 已提交
1245
          Status s = db_->Get(read_opts, column_family, key, &from_db);
1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256
          if (s.ok()) {
            // found case
            thread->stats.AddGets(1, 1);
          } else if (s.IsNotFound()) {
            // not found case
            thread->stats.AddGets(1, 0);
          } else {
            // errors case
            thread->stats.AddErrors(1);
          }
        } else {
I
Igor Canadi 已提交
1257
          MultiGet(thread, read_opts, column_family, key, &from_db);
1258
        }
1259 1260
      } else if ((int)FLAGS_readpercent <= prob_op && prob_op < prefixBound) {
        // OPERATION prefix scan
L
Lei Jin 已提交
1261 1262
        // keys are 8 bytes long, prefix size is FLAGS_prefix_size. There are
        // (8 - FLAGS_prefix_size) bytes besides the prefix. So there will
I
Igor Canadi 已提交
1263 1264
        // be 2 ^ ((8 - FLAGS_prefix_size) * 8) possible keys with the same
        // prefix
1265
        if (!FLAGS_test_batches_snapshots) {
L
Lei Jin 已提交
1266
          Slice prefix = Slice(key.data(), FLAGS_prefix_size);
I
Igor Canadi 已提交
1267
          Iterator* iter = db_->NewIterator(read_opts, column_family);
L
Lei Jin 已提交
1268
          int64_t count = 0;
I
Igor Canadi 已提交
1269 1270
          for (iter->Seek(prefix);
               iter->Valid() && iter->key().starts_with(prefix); iter->Next()) {
L
Lei Jin 已提交
1271
            ++count;
1272
          }
I
Igor Canadi 已提交
1273 1274
          assert(count <=
                 (static_cast<int64_t>(1) << ((8 - FLAGS_prefix_size) * 8)));
1275 1276 1277 1278 1279 1280
          if (iter->status().ok()) {
            thread->stats.AddPrefixes(1, count);
          } else {
            thread->stats.AddErrors(1);
          }
          delete iter;
1281
        } else {
1282
          MultiPrefixScan(thread, read_opts, column_family, key);
A
amayank 已提交
1283
        }
1284 1285
      } else if (prefixBound <= prob_op && prob_op < writeBound) {
        // OPERATION write
1286 1287 1288
        uint32_t value_base = thread->rand.Next();
        size_t sz = GenerateValue(value_base, value, sizeof(value));
        Slice v(value, sz);
1289
        if (!FLAGS_test_batches_snapshots) {
1290
          if (FLAGS_verify_before_write) {
I
Igor Canadi 已提交
1291 1292
            std::string keystr2 = Key(rand_key);
            Slice k = keystr2;
I
Igor Canadi 已提交
1293
            Status s = db_->Get(read_opts, column_family, k, &from_db);
I
Igor Canadi 已提交
1294 1295 1296 1297
            if (VerifyValue(rand_column_family, rand_key, read_opts,
                            thread->shared, from_db, s, true) == false) {
              break;
            }
1298
          }
I
Igor Canadi 已提交
1299
          thread->shared->Put(rand_column_family, rand_key, value_base);
1300
          if (FLAGS_use_merge) {
I
Igor Canadi 已提交
1301
            db_->Merge(write_opts, column_family, key, v);
D
Deon Nicholas 已提交
1302
          } else {
I
Igor Canadi 已提交
1303
            db_->Put(write_opts, column_family, key, v);
D
Deon Nicholas 已提交
1304
          }
1305 1306
          thread->stats.AddBytesForWrites(1, sz);
        } else {
I
Igor Canadi 已提交
1307
          MultiPut(thread, write_opts, column_family, key, v, sz);
1308
        }
I
Igor Canadi 已提交
1309
        PrintKeyValue(rand_column_family, rand_key, value, sz);
1310 1311
      } else if (writeBound <= prob_op && prob_op < delBound) {
        // OPERATION delete
1312
        if (!FLAGS_test_batches_snapshots) {
I
Igor Canadi 已提交
1313 1314
          thread->shared->Delete(rand_column_family, rand_key);
          db_->Delete(write_opts, column_family, key);
1315 1316
          thread->stats.AddDeletes(1);
        } else {
I
Igor Canadi 已提交
1317
          MultiDelete(thread, write_opts, column_family, key);
1318
        }
1319 1320
      } else {
        // OPERATION iterate
I
Igor Canadi 已提交
1321
        MultiIterate(thread, read_opts, column_family, key);
1322
      }
1323 1324
      thread->stats.FinishedSingleOp();
    }
1325

1326 1327 1328
    thread->stats.Stop();
  }

M
Mayank Agarwal 已提交
1329
  void VerifyDb(ThreadState* thread) const {
1330
    ReadOptions options(FLAGS_verify_checksum, true);
I
Igor Canadi 已提交
1331 1332 1333
    auto shared = thread->shared;
    static const long max_key = shared->GetMaxKey();
    static const long keys_per_thread = max_key / shared->GetNumThreads();
M
Mayank Agarwal 已提交
1334
    long start = keys_per_thread * thread->tid;
1335
    long end = start + keys_per_thread;
I
Igor Canadi 已提交
1336
    if (thread->tid == shared->GetNumThreads() - 1) {
1337 1338
      end = max_key;
    }
I
Igor Canadi 已提交
1339
    for (size_t cf = 0; cf < column_families_.size(); ++cf) {
I
Igor Canadi 已提交
1340 1341 1342
      if (thread->shared->HasVerificationFailedYet()) {
        break;
      }
I
Igor Canadi 已提交
1343 1344 1345 1346 1347 1348
      if (!thread->rand.OneIn(2)) {
        // Use iterator to verify this range
        unique_ptr<Iterator> iter(
            db_->NewIterator(options, column_families_[cf]));
        iter->Seek(Key(start));
        for (long i = start; i < end; i++) {
I
Igor Canadi 已提交
1349 1350 1351
          if (thread->shared->HasVerificationFailedYet()) {
            break;
          }
1352 1353 1354 1355 1356 1357
          // TODO(ljin): update "long" to uint64_t
          // Reseek when the prefix changes
          if (i % (static_cast<int64_t>(1) << 8 * (8 - FLAGS_prefix_size)) ==
              0) {
            iter->Seek(Key(i));
          }
I
Igor Canadi 已提交
1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368
          std::string from_db;
          std::string keystr = Key(i);
          Slice k = keystr;
          Status s = iter->status();
          if (iter->Valid()) {
            if (iter->key().compare(k) > 0) {
              s = Status::NotFound(Slice());
            } else if (iter->key().compare(k) == 0) {
              from_db = iter->value().ToString();
              iter->Next();
            } else if (iter->key().compare(k) < 0) {
I
Igor Canadi 已提交
1369
              VerificationAbort(shared, "An out of range key was found", cf, i);
I
Igor Canadi 已提交
1370 1371 1372 1373
            }
          } else {
            // The iterator found no value for the key in question, so do not
            // move to the next item in the iterator
M
Mayank Agarwal 已提交
1374 1375
            s = Status::NotFound(Slice());
          }
I
Igor Canadi 已提交
1376 1377 1378 1379
          VerifyValue(cf, i, options, shared, from_db, s, true);
          if (from_db.length()) {
            PrintKeyValue(cf, i, from_db.data(), from_db.length());
          }
M
Mayank Agarwal 已提交
1380
        }
I
Igor Canadi 已提交
1381 1382 1383
      } else {
        // Use Get to verify this range
        for (long i = start; i < end; i++) {
I
Igor Canadi 已提交
1384 1385 1386
          if (thread->shared->HasVerificationFailedYet()) {
            break;
          }
I
Igor Canadi 已提交
1387 1388 1389 1390
          std::string from_db;
          std::string keystr = Key(i);
          Slice k = keystr;
          Status s = db_->Get(options, column_families_[cf], k, &from_db);
1391
          VerifyValue(cf, i, options, shared, from_db, s, true);
I
Igor Canadi 已提交
1392 1393 1394
          if (from_db.length()) {
            PrintKeyValue(cf, i, from_db.data(), from_db.length());
          }
M
Mayank Agarwal 已提交
1395
        }
1396 1397 1398 1399
      }
    }
  }

I
Igor Canadi 已提交
1400 1401 1402 1403 1404
  void VerificationAbort(SharedState* shared, std::string msg, int cf,
                         long key) const {
    printf("Verification failed for column family %d key %ld: %s\n", cf, key,
           msg.c_str());
    shared->SetVerificationFailure();
1405 1406
  }

I
Igor Canadi 已提交
1407 1408
  bool VerifyValue(int cf, long key, const ReadOptions& opts,
                   SharedState* shared, const std::string& value_from_db,
I
Igor Canadi 已提交
1409
                   Status s, bool strict = false) const {
I
Igor Canadi 已提交
1410 1411 1412
    if (shared->HasVerificationFailedYet()) {
      return false;
    }
1413
    // compare value_from_db with the value in the shared state
A
amayank 已提交
1414
    char value[100];
I
Igor Canadi 已提交
1415
    uint32_t value_base = shared->Get(cf, key);
1416
    if (value_base == SharedState::SENTINEL && !strict) {
I
Igor Canadi 已提交
1417
      return true;
1418 1419
    }

1420
    if (s.ok()) {
1421
      if (value_base == SharedState::SENTINEL) {
I
Igor Canadi 已提交
1422 1423
        VerificationAbort(shared, "Unexpected value found", cf, key);
        return false;
1424
      }
A
amayank 已提交
1425
      size_t sz = GenerateValue(value_base, value, sizeof(value));
1426
      if (value_from_db.length() != sz) {
I
Igor Canadi 已提交
1427 1428
        VerificationAbort(shared, "Length of value read is not equal", cf, key);
        return false;
1429
      }
1430
      if (memcmp(value_from_db.data(), value, sz) != 0) {
I
Igor Canadi 已提交
1431 1432 1433
        VerificationAbort(shared, "Contents of value read don't match", cf,
                          key);
        return false;
1434 1435 1436
      }
    } else {
      if (value_base != SharedState::SENTINEL) {
I
Igor Canadi 已提交
1437 1438
        VerificationAbort(shared, "Value not found: " + s.ToString(), cf, key);
        return false;
1439 1440
      }
    }
I
Igor Canadi 已提交
1441
    return true;
1442 1443
  }

I
Igor Canadi 已提交
1444 1445 1446 1447 1448 1449 1450
  static void PrintKeyValue(int cf, uint32_t key, const char* value,
                            size_t sz) {
    if (!FLAGS_verbose) {
      return;
    }
    fprintf(stdout, "[CF %d] %u ==> (%u) ", cf, key, (unsigned int)sz);
    for (size_t i = 0; i < sz; i++) {
1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462
      fprintf(stdout, "%X", value[i]);
    }
    fprintf(stdout, "\n");
  }

  static size_t GenerateValue(uint32_t rand, char *v, size_t max_sz) {
    size_t value_sz = ((rand % 3) + 1) * FLAGS_value_size_mult;
    assert(value_sz <= max_sz && value_sz >= sizeof(uint32_t));
    *((uint32_t*)v) = rand;
    for (size_t i=sizeof(uint32_t); i < value_sz; i++) {
      v[i] = (char)(rand ^ i);
    }
1463
    v[value_sz] = '\0';
1464 1465 1466 1467
    return value_sz; // the size of the value set.
  }

  void PrintEnv() const {
I
Igor Canadi 已提交
1468 1469 1470 1471 1472 1473 1474
    fprintf(stdout, "RocksDB version     : %d.%d\n", kMajorVersion,
            kMinorVersion);
    fprintf(stdout, "Column families     : %d\n", FLAGS_column_families);
    if (!FLAGS_test_batches_snapshots) {
      fprintf(stdout, "Clear CFs one in    : %d\n",
              FLAGS_clear_column_family_one_in);
    }
1475
    fprintf(stdout, "Number of threads   : %d\n", FLAGS_threads);
K
kailiu 已提交
1476 1477 1478
    fprintf(stdout,
            "Ops per thread      : %lu\n",
            (unsigned long)FLAGS_ops_per_thread);
1479 1480 1481 1482 1483
    std::string ttl_state("unused");
    if (FLAGS_ttl > 0) {
      ttl_state = NumberToString(FLAGS_ttl);
    }
    fprintf(stdout, "Time to live(sec)   : %s\n", ttl_state.c_str());
K
Kai Liu 已提交
1484 1485 1486 1487 1488
    fprintf(stdout, "Read percentage     : %d%%\n", FLAGS_readpercent);
    fprintf(stdout, "Prefix percentage   : %d%%\n", FLAGS_prefixpercent);
    fprintf(stdout, "Write percentage    : %d%%\n", FLAGS_writepercent);
    fprintf(stdout, "Delete percentage   : %d%%\n", FLAGS_delpercent);
    fprintf(stdout, "Iterate percentage  : %d%%\n", FLAGS_iterpercent);
1489
    fprintf(stdout, "Write-buffer-size   : %d\n", FLAGS_write_buffer_size);
K
kailiu 已提交
1490 1491 1492 1493 1494 1495
    fprintf(stdout,
            "Iterations          : %lu\n",
            (unsigned long)FLAGS_num_iterations);
    fprintf(stdout,
            "Max key             : %lu\n",
            (unsigned long)FLAGS_max_key);
X
Xing Jin 已提交
1496 1497
    fprintf(stdout, "Ratio #ops/#keys    : %f\n",
            (1.0 * FLAGS_ops_per_thread * FLAGS_threads)/FLAGS_max_key);
1498
    fprintf(stdout, "Num times DB reopens: %d\n", FLAGS_reopen);
1499 1500
    fprintf(stdout, "Batches/snapshots   : %d\n",
            FLAGS_test_batches_snapshots);
X
Xing Jin 已提交
1501
    fprintf(stdout, "Purge redundant %%   : %d\n",
1502
            FLAGS_purge_redundant_percent);
1503
    fprintf(stdout, "Deletes use filter  : %d\n",
1504
            FLAGS_filter_deletes);
1505
    fprintf(stdout, "Num keys per lock   : %d\n",
1506 1507
            1 << FLAGS_log2_keys_per_lock);

1508
    const char* compression = "";
1509
    switch (FLAGS_compression_type_e) {
1510
      case rocksdb::kNoCompression:
1511
        compression = "none";
1512
        break;
1513
      case rocksdb::kSnappyCompression:
1514
        compression = "snappy";
1515
        break;
1516
      case rocksdb::kZlibCompression:
1517
        compression = "zlib";
1518
        break;
1519
      case rocksdb::kBZip2Compression:
1520
        compression = "bzip2";
1521
        break;
A
Albert Strasheim 已提交
1522 1523 1524 1525 1526 1527
      case rocksdb::kLZ4Compression:
        compression = "lz4";
      case rocksdb::kLZ4HCCompression:
        compression = "lz4hc";
        break;
      }
1528 1529

    fprintf(stdout, "Compression         : %s\n", compression);
J
Jim Paton 已提交
1530 1531 1532 1533 1534 1535

    const char* memtablerep = "";
    switch (FLAGS_rep_factory) {
      case kSkipList:
        memtablerep = "skip_list";
        break;
I
Igor Canadi 已提交
1536
      case kHashSkipList:
J
Jim Paton 已提交
1537 1538 1539 1540 1541 1542 1543 1544 1545
        memtablerep = "prefix_hash";
        break;
      case kVectorRep:
        memtablerep = "vector";
        break;
    }

    fprintf(stdout, "Memtablerep         : %s\n", memtablerep);

1546 1547 1548 1549
    fprintf(stdout, "------------------------------------------------\n");
  }

  void Open() {
1550
    assert(db_ == nullptr);
1551 1552 1553 1554 1555 1556 1557
    BlockBasedTableOptions block_based_options;
    block_based_options.block_cache = cache_;
    block_based_options.block_cache_compressed = compressed_cache_;
    block_based_options.block_size = FLAGS_block_size;
    block_based_options.filter_policy = filter_policy_;
    options_.table_factory.reset(
        NewBlockBasedTableFactory(block_based_options));
I
Igor Canadi 已提交
1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572
    options_.write_buffer_size = FLAGS_write_buffer_size;
    options_.max_write_buffer_number = FLAGS_max_write_buffer_number;
    options_.min_write_buffer_number_to_merge =
        FLAGS_min_write_buffer_number_to_merge;
    options_.max_background_compactions = FLAGS_max_background_compactions;
    options_.max_background_flushes = FLAGS_max_background_flushes;
    options_.compaction_style =
        static_cast<rocksdb::CompactionStyle>(FLAGS_compaction_style);
    options_.prefix_extractor.reset(NewFixedPrefixTransform(FLAGS_prefix_size));
    options_.max_open_files = FLAGS_open_files;
    options_.statistics = dbstats;
    options_.env = FLAGS_env;
    options_.disableDataSync = FLAGS_disable_data_sync;
    options_.use_fsync = FLAGS_use_fsync;
    options_.allow_mmap_reads = FLAGS_mmap_read;
1573
    rocksdb_kill_odds = FLAGS_kill_random_test;
I
Igor Canadi 已提交
1574 1575 1576 1577
    options_.target_file_size_base = FLAGS_target_file_size_base;
    options_.target_file_size_multiplier = FLAGS_target_file_size_multiplier;
    options_.max_bytes_for_level_base = FLAGS_max_bytes_for_level_base;
    options_.max_bytes_for_level_multiplier =
1578
        FLAGS_max_bytes_for_level_multiplier;
I
Igor Canadi 已提交
1579 1580 1581 1582 1583 1584 1585 1586 1587
    options_.level0_stop_writes_trigger = FLAGS_level0_stop_writes_trigger;
    options_.level0_slowdown_writes_trigger =
        FLAGS_level0_slowdown_writes_trigger;
    options_.level0_file_num_compaction_trigger =
        FLAGS_level0_file_num_compaction_trigger;
    options_.compression = FLAGS_compression_type_e;
    options_.create_if_missing = true;
    options_.max_manifest_file_size = 10 * 1024;
    options_.filter_deletes = FLAGS_filter_deletes;
I
Igor Canadi 已提交
1588
    if ((FLAGS_prefix_size == 0) == (FLAGS_rep_factory == kHashSkipList)) {
J
Jim Paton 已提交
1589 1590 1591 1592 1593
      fprintf(stderr,
            "prefix_size should be non-zero iff memtablerep == prefix_hash\n");
      exit(1);
    }
    switch (FLAGS_rep_factory) {
I
Igor Canadi 已提交
1594
      case kHashSkipList:
1595
        options_.memtable_factory.reset(NewHashSkipListRepFactory(10000));
J
Jim Paton 已提交
1596 1597 1598 1599 1600
        break;
      case kSkipList:
        // no need to do anything
        break;
      case kVectorRep:
I
Igor Canadi 已提交
1601
        options_.memtable_factory.reset(new VectorRepFactory());
J
Jim Paton 已提交
1602 1603
        break;
    }
1604
    static Random purge_percent(1000); // no benefit from non-determinism here
1605 1606
    if (static_cast<int32_t>(purge_percent.Uniform(100)) <
        FLAGS_purge_redundant_percent - 1) {
I
Igor Canadi 已提交
1607
      options_.purge_redundant_kvs_while_flush = false;
1608
    }
1609

1610
    if (FLAGS_use_merge) {
I
Igor Canadi 已提交
1611
      options_.merge_operator = MergeOperators::CreatePutOperator();
D
Deon Nicholas 已提交
1612 1613
    }

1614 1615
    // set universal style compaction configurations, if applicable
    if (FLAGS_universal_size_ratio != 0) {
I
Igor Canadi 已提交
1616 1617
      options_.compaction_options_universal.size_ratio =
          FLAGS_universal_size_ratio;
1618 1619
    }
    if (FLAGS_universal_min_merge_width != 0) {
I
Igor Canadi 已提交
1620 1621
      options_.compaction_options_universal.min_merge_width =
          FLAGS_universal_min_merge_width;
1622 1623
    }
    if (FLAGS_universal_max_merge_width != 0) {
I
Igor Canadi 已提交
1624 1625
      options_.compaction_options_universal.max_merge_width =
          FLAGS_universal_max_merge_width;
1626 1627
    }
    if (FLAGS_universal_max_size_amplification_percent != 0) {
I
Igor Canadi 已提交
1628 1629
      options_.compaction_options_universal.max_size_amplification_percent =
          FLAGS_universal_max_size_amplification_percent;
1630 1631
    }

1632
    fprintf(stdout, "DB path: [%s]\n", FLAGS_db.c_str());
1633

1634 1635
    Status s;
    if (FLAGS_ttl == -1) {
I
Igor Canadi 已提交
1636 1637 1638 1639 1640 1641 1642
      std::vector<std::string> existing_column_families;
      s = DB::ListColumnFamilies(DBOptions(options_), FLAGS_db,
                                 &existing_column_families);  // ignore errors
      if (!s.ok()) {
        // DB doesn't exist
        assert(existing_column_families.empty());
        assert(column_family_names_.empty());
1643
        column_family_names_.push_back(kDefaultColumnFamilyName);
I
Igor Canadi 已提交
1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670
      } else if (column_family_names_.empty()) {
        // this is the first call to the function Open()
        column_family_names_ = existing_column_families;
      } else {
        // this is a reopen. just assert that existing column_family_names are
        // equivalent to what we remember
        auto sorted_cfn = column_family_names_;
        sort(sorted_cfn.begin(), sorted_cfn.end());
        sort(existing_column_families.begin(), existing_column_families.end());
        if (sorted_cfn != existing_column_families) {
          fprintf(stderr,
                  "Expected column families differ from the existing:\n");
          printf("Expected: {");
          for (auto cf : sorted_cfn) {
            printf("%s ", cf.c_str());
          }
          printf("}\n");
          printf("Existing: {");
          for (auto cf : existing_column_families) {
            printf("%s ", cf.c_str());
          }
          printf("}\n");
        }
        assert(sorted_cfn == existing_column_families);
      }
      std::vector<ColumnFamilyDescriptor> cf_descriptors;
      for (auto name : column_family_names_) {
1671
        if (name != kDefaultColumnFamilyName) {
I
Igor Canadi 已提交
1672 1673 1674 1675 1676
          new_column_family_name_ =
              std::max(new_column_family_name_.load(), std::stoi(name) + 1);
        }
        cf_descriptors.emplace_back(name, ColumnFamilyOptions(options_));
      }
I
Igor Canadi 已提交
1677 1678 1679 1680 1681 1682 1683
      while (cf_descriptors.size() < (size_t)FLAGS_column_families) {
        std::string name = std::to_string(new_column_family_name_.load());
        new_column_family_name_++;
        cf_descriptors.emplace_back(name, ColumnFamilyOptions(options_));
        column_family_names_.push_back(name);
      }
      options_.create_missing_column_families = true;
I
Igor Canadi 已提交
1684 1685 1686 1687
      s = DB::Open(DBOptions(options_), FLAGS_db, cf_descriptors,
                   &column_families_, &db_);
      assert(!s.ok() || column_families_.size() ==
                            static_cast<size_t>(FLAGS_column_families));
1688
    } else {
I
Igor Canadi 已提交
1689 1690 1691
      DBWithTTL* db_with_ttl;
      s = DBWithTTL::Open(options_, FLAGS_db, &db_with_ttl, FLAGS_ttl);
      db_ = db_with_ttl;
1692
    }
1693 1694 1695 1696 1697 1698
    if (!s.ok()) {
      fprintf(stderr, "open error: %s\n", s.ToString().c_str());
      exit(1);
    }
  }

1699
  void Reopen() {
I
Igor Canadi 已提交
1700 1701
    for (auto cf : column_families_) {
      delete cf;
1702
    }
I
Igor Canadi 已提交
1703 1704
    column_families_.clear();
    delete db_;
1705
    db_ = nullptr;
1706 1707 1708 1709 1710 1711

    num_times_reopened_++;
    double now = FLAGS_env->NowMicros();
    fprintf(stdout, "%s Reopening database for the %dth time\n",
            FLAGS_env->TimeToString((uint64_t) now/1000000).c_str(),
            num_times_reopened_);
1712 1713 1714
    Open();
  }

1715 1716
  void PrintStatistics() {
    if (dbstats) {
1717
      fprintf(stdout, "STATISTICS:\n%s\n", dbstats->ToString().c_str());
1718 1719 1720 1721
    }
  }

 private:
1722 1723 1724
  std::shared_ptr<Cache> cache_;
  std::shared_ptr<Cache> compressed_cache_;
  std::shared_ptr<const FilterPolicy> filter_policy_;
1725
  DB* db_;
I
Igor Canadi 已提交
1726 1727 1728 1729
  Options options_;
  std::vector<ColumnFamilyHandle*> column_families_;
  std::vector<std::string> column_family_names_;
  std::atomic<int> new_column_family_name_;
1730
  int num_times_reopened_;
1731 1732
};

1733
}  // namespace rocksdb
1734 1735

int main(int argc, char** argv) {
1736 1737 1738
  SetUsageMessage(std::string("\nUSAGE:\n") + std::string(argv[0]) +
                  " [OPTIONS]...");
  ParseCommandLineFlags(&argc, &argv, true);
1739 1740 1741 1742 1743 1744 1745 1746

  if (FLAGS_statistics) {
    dbstats = rocksdb::CreateDBStatistics();
  }
  FLAGS_compression_type_e =
    StringToCompressionType(FLAGS_compression_type.c_str());
  if (!FLAGS_hdfs.empty()) {
    FLAGS_env  = new rocksdb::HdfsEnv(FLAGS_hdfs);
1747
  }
1748
  FLAGS_rep_factory = StringToRepFactory(FLAGS_memtablerep.c_str());
1749

1750 1751 1752 1753
  // The number of background threads should be at least as much the
  // max number of concurrent compactions.
  FLAGS_env->SetBackgroundThreads(FLAGS_max_background_compactions);

L
Lei Jin 已提交
1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765
  if (FLAGS_prefixpercent > 0 && FLAGS_prefix_size <= 0) {
    fprintf(stderr,
            "Error: prefixpercent is non-zero while prefix_size is "
            "not positive!\n");
    exit(1);
  }
  if (FLAGS_test_batches_snapshots && FLAGS_prefix_size <= 0) {
    fprintf(stderr,
            "Error: please specify prefix_size for "
            "test_batches_snapshots test!\n");
    exit(1);
  }
1766
  if ((FLAGS_readpercent + FLAGS_prefixpercent +
1767 1768 1769
       FLAGS_writepercent + FLAGS_delpercent + FLAGS_iterpercent) != 100) {
      fprintf(stderr,
              "Error: Read+Prefix+Write+Delete+Iterate percents != 100!\n");
A
amayank 已提交
1770 1771
      exit(1);
  }
1772 1773 1774 1775
  if (FLAGS_disable_wal == 1 && FLAGS_reopen > 0) {
      fprintf(stderr, "Error: Db cannot reopen safely with disable_wal set!\n");
      exit(1);
  }
1776
  if ((unsigned)FLAGS_reopen >= FLAGS_ops_per_thread) {
K
kailiu 已提交
1777 1778 1779 1780 1781
      fprintf(stderr,
              "Error: #DB-reopens should be < ops_per_thread\n"
              "Provided reopens = %d and ops_per_thread = %lu\n",
              FLAGS_reopen,
              (unsigned long)FLAGS_ops_per_thread);
1782 1783 1784
      exit(1);
  }

1785
  // Choose a location for the test database if none given with --db=<path>
1786 1787
  if (FLAGS_db.empty()) {
      std::string default_db_path;
1788
      rocksdb::Env::Default()->GetTestDirectory(&default_db_path);
1789
      default_db_path += "/dbstress";
1790
      FLAGS_db = default_db_path;
1791 1792
  }

1793
  rocksdb::StressTest stress;
I
Igor Canadi 已提交
1794 1795 1796 1797 1798
  if (stress.Run()) {
    return 0;
  } else {
    return 1;
  }
1799
}
1800 1801

#endif  // GFLAGS