db_stress.cc 53.3 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
#include <sys/types.h>
#include <stdio.h>
#include <stdlib.h>
#include "db/db_impl.h"
#include "db/version_set.h"
#include "db/db_statistics.h"
29
#include "rocksdb/cache.h"
30
#include "utilities/utility_db.h"
31 32 33
#include "rocksdb/env.h"
#include "rocksdb/write_batch.h"
#include "rocksdb/statistics.h"
34
#include "port/port.h"
35
#include "util/coding.h"
36 37 38 39 40
#include "util/crc32c.h"
#include "util/histogram.h"
#include "util/mutexlock.h"
#include "util/random.h"
#include "util/testutil.h"
41 42
#include "util/logging.h"
#include "utilities/ttl/db_ttl.h"
43
#include "hdfs/env_hdfs.h"
D
Deon Nicholas 已提交
44
#include "utilities/merge_operators.h"
45 46 47 48 49 50 51

static const long KB = 1024;

// Seed for PRNG
static uint32_t FLAGS_seed = 2341234;

// Max number of key/values to place in database
X
Xing Jin 已提交
52 53 54 55 56 57 58 59 60
static long FLAGS_max_key = 1 * KB * KB * KB;

// If set, the test uses MultiGet(), MultiPut() and MultiDelete() 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:
// (a) No need to acquire mutexes during writes (less cache flushes
//     in multi-core leading to speed up)
61
// (b) No long validation at the end (more speed up)
X
Xing Jin 已提交
62
// (c) Test snapshot and atomicity of batch writes
63 64
static bool FLAGS_test_batches_snapshots = false;

65 66 67
// Number of concurrent threads to run.
static int FLAGS_threads = 32;

68 69
// Opens the db with this ttl value if this is not -1
// Carefully specify a large value such that verifications on deleted
X
Xing Jin 已提交
70
// values don't fail
71 72
static int FLAGS_ttl = -1;

73 74 75 76 77 78 79 80
// Size of each value will be this number times rand_int(1,3) bytes
static int FLAGS_value_size_mult = 8;

static bool FLAGS_verify_before_write = false;

// Print histogram of operation timings
static bool FLAGS_histogram = false;

81 82 83
// Destroys the database dir before start if this is true
static bool FLAGS_destroy_db_initially = true;

84 85 86 87 88 89
static bool FLAGS_verbose = false;

// Number of bytes to buffer in memtable before compacting
// (initialized to default value by "main")
static int FLAGS_write_buffer_size = 0;

A
Abhishek Kona 已提交
90
// The number of in-memory memtables.
91 92 93 94
// Each memtable is of size FLAGS_write_buffer_size.
// This is initialized to default value of 2 in "main" function.
static int FLAGS_max_write_buffer_number = 0;

95 96 97 98 99 100 101 102 103 104
// 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.
static int FLAGS_min_write_buffer_number_to_merge = 0;

105 106 107 108 109
// The maximum number of concurrent background compactions
// that can occur in parallel.
// This is initialized to default value of 1 in "main" function.
static int FLAGS_max_background_compactions = 0;

110
// This is initialized to default value of false
111
static rocksdb::CompactionStyle FLAGS_compaction_style = rocksdb::kCompactionStyleLevel;
112

113 114 115 116 117 118 119 120 121 122 123 124
// The ratio of file sizes that trigger compaction in universal style
static unsigned int FLAGS_universal_size_ratio = 0;

// The minimum number of files to compact in universal style compaction
static unsigned int FLAGS_universal_min_merge_width = 0;

// The max number of files to compact in universal style compaction
static unsigned int FLAGS_universal_max_merge_width = 0;

// The max size amplification for universal style compaction
static unsigned int FLAGS_universal_max_size_amplification_percent = 0;

125 126 127 128 129 130
// Number of bytes to use as a cache of uncompressed data.
static long FLAGS_cache_size = 2 * KB * KB * KB;

// Number of bytes in a block.
static int FLAGS_block_size = 4 * KB;

131 132 133
// Number of times database reopens
static int FLAGS_reopen = 10;

134 135 136 137 138 139 140 141
// Maximum number of files to keep open at the same time (use default if == 0)
static int FLAGS_open_files = 0;

// Bloom filter bits per key.
// Negative means use default settings.
static int FLAGS_bloom_bits = 10;

// Use the db with the following name.
142
static const char* FLAGS_db = nullptr;
143 144 145 146

// Verify checksum for every block read from storage
static bool FLAGS_verify_checksum = false;

147
// Allow reads to occur via mmap-ing files
148
static bool FLAGS_use_mmap_reads = rocksdb::EnvOptions().use_mmap_reads;
149

150
// Database statistics
151
static std::shared_ptr<rocksdb::Statistics> dbstats;
152 153 154 155 156 157 158 159 160 161

// Sync all writes to disk
static bool FLAGS_sync = false;

// If true, do not wait until data is synced to disk.
static bool FLAGS_disable_data_sync = false;

// If true, issue fsync instead of fdatasync
static bool FLAGS_use_fsync = false;

162 163
// If non-zero, kill at various points in source code with probability 1/this
static int FLAGS_kill_random_test = 0;
164
extern int rocksdb_kill_odds;
165

166 167 168
// If true, do not write WAL for write.
static bool FLAGS_disable_wal = false;

X
Xing Jin 已提交
169
// Target level-1 file size for compaction
170 171
static int FLAGS_target_file_size_base = 64 * KB;

X
Xing Jin 已提交
172
// A multiplier to compute targe level-N file size (N >= 2)
173 174
static int FLAGS_target_file_size_multiplier = 1;

X
Xing Jin 已提交
175
// Max bytes for level-1
176
static uint64_t FLAGS_max_bytes_for_level_base = 256 * KB;
177

X
Xing Jin 已提交
178
// A multiplier to compute max bytes for level-N (N >= 2)
179 180 181 182 183 184 185 186
static int FLAGS_max_bytes_for_level_multiplier = 2;

// Number of files in level-0 that will trigger put stop.
static int FLAGS_level0_stop_writes_trigger = 12;

// Number of files in level-0 that will slow down writes.
static int FLAGS_level0_slowdown_writes_trigger = 8;

A
amayank 已提交
187 188 189
// Ratio of reads to total workload (expressed as a percentage)
static unsigned int FLAGS_readpercent = 10;

190
// Ratio of prefix iterators to total workload (expressed as a percentage)
191
static unsigned int FLAGS_prefixpercent = 20;
192 193

// Ratio of deletes to total workload (expressed as a percentage)
194
static unsigned int FLAGS_writepercent = 45;
195

A
amayank 已提交
196
// Ratio of deletes to total workload (expressed as a percentage)
197
static unsigned int FLAGS_delpercent = 15;
198

199 200 201 202 203 204
// Ratio of iterations to total workload (expressed as a percentage)
static unsigned int FLAGS_iterpercent = 10;

// Number of iterations per MultiIterate run
static uint32_t FLAGS_num_iterations = 10;

205 206 207
// Option to disable compation triggered by read.
static int FLAGS_disable_seek_compaction = false;

208 209 210 211 212
// Option to delete obsolete files periodically
// Default: 0 which means that obsolete files are
// deleted after every compaction run.
 static uint64_t FLAGS_delete_obsolete_files_period_micros = 0;

213
// Algorithm to use to compress the database
214 215
static enum rocksdb::CompressionType FLAGS_compression_type =
    rocksdb::kSnappyCompression;
216 217

// posix or hdfs environment
218
static rocksdb::Env* FLAGS_env = rocksdb::Env::Default();
219 220 221 222 223 224 225

// Number of operations per thread.
static uint32_t FLAGS_ops_per_thread = 600000;

// Log2 of number of keys per lock
static uint32_t FLAGS_log2_keys_per_lock = 2; // implies 2^2 keys per lock

226 227 228
// Percentage of times we want to purge redundant keys in memory before flushing
static uint32_t FLAGS_purge_redundant_percent = 50;

229
// On true, deletes use KeyMayExist to drop the delete if key not present
230
static bool FLAGS_filter_deletes = false;
231

232 233 234
// Level0 compaction start trigger
static int FLAGS_level0_file_num_compaction_trigger = 0;

J
Jim Paton 已提交
235 236 237 238 239 240 241 242 243 244 245 246
enum RepFactory {
  kSkipList,
  kPrefixHash,
  kUnsorted,
  kVectorRep
};

static enum RepFactory FLAGS_rep_factory;

// Control the prefix size for PrefixHashRep
static bool FLAGS_prefix_size = 0;

D
Deon Nicholas 已提交
247 248 249
// On true, replaces all writes with a Merge that behaves like a Put
static bool FLAGS_use_merge_put = false;

250
namespace rocksdb {
251

252 253 254 255 256 257 258 259 260 261 262 263 264
// 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;
}

265 266 267 268 269 270 271 272 273
class StressTest;
namespace {

class Stats {
 private:
  double start_;
  double finish_;
  double seconds_;
  long done_;
274 275
  long gets_;
  long prefixes_;
276
  long writes_;
A
amayank 已提交
277
  long deletes_;
278
  long iterator_size_sums_;
279
  long founds_;
280
  long iterations_;
281
  long errors_;
282 283 284
  int next_report_;
  size_t bytes_;
  double last_op_finish_;
285
  HistogramImpl hist_;
286 287 288 289 290 291 292 293

 public:
  Stats() { }

  void Start() {
    next_report_ = 100;
    hist_.Clear();
    done_ = 0;
294 295
    gets_ = 0;
    prefixes_ = 0;
296
    writes_ = 0;
A
amayank 已提交
297
    deletes_ = 0;
298
    iterator_size_sums_ = 0;
299
    founds_ = 0;
300
    iterations_ = 0;
301
    errors_ = 0;
302 303 304 305 306 307 308 309 310 311
    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_;
312 313
    gets_ += other.gets_;
    prefixes_ += other.prefixes_;
314
    writes_ += other.writes_;
A
amayank 已提交
315
    deletes_ += other.deletes_;
316
    iterator_size_sums_ += other.iterator_size_sums_;
317
    founds_ += other.founds_;
318
    iterations_ += other.iterations_;
319
    errors_ += other.errors_;
320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336
    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) {
337
        fprintf(stdout, "long op: %.1f micros%30s\r", micros, "");
338 339 340 341 342 343 344 345 346 347 348 349 350
      }
      last_op_finish_ = now;
    }

    done_++;
    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;
351
      fprintf(stdout, "... finished %ld ops%30s\r", done_, "");
352 353 354
    }
  }

355 356 357 358 359
  void AddBytesForWrites(int nwrites, size_t nbytes) {
    writes_ += nwrites;
    bytes_ += nbytes;
  }

360 361 362
  void AddGets(int ngets, int nfounds) {
    founds_ += nfounds;
    gets_ += ngets;
363 364
  }

365 366 367 368 369
  void AddPrefixes(int nprefixes, int count) {
    prefixes_ += nprefixes;
    iterator_size_sums_ += count;
  }

370 371 372 373
  void AddIterations(int n) {
    iterations_ += n;
  }

374 375 376 377
  void AddDeletes(int n) {
    deletes_ += n;
  }

378 379
  void AddErrors(int n) {
    errors_ += n;
A
amayank 已提交
380 381
  }

382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398
  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_);
399
    fprintf(stdout, "%-12s: Wrote %ld times\n", "", writes_);
A
amayank 已提交
400
    fprintf(stdout, "%-12s: Deleted %ld times\n", "", deletes_);
X
Xing Jin 已提交
401 402
    fprintf(stdout, "%-12s: %ld read and %ld found the key\n", "",
            gets_, founds_);
403 404 405
    fprintf(stdout, "%-12s: Prefix scanned %ld times\n", "", prefixes_);
    fprintf(stdout, "%-12s: Iterator size sum is %ld\n", "",
            iterator_size_sums_);
406
    fprintf(stdout, "%-12s: Iterated %ld times\n", "", iterations_);
407
    fprintf(stdout, "%-12s: Got errors %ld times\n", "", errors_);
408 409 410 411 412 413 414 415 416 417 418 419 420

    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:
  static const uint32_t SENTINEL = 0xffffffff;

421
  explicit SharedState(StressTest* stress_test) :
422 423 424 425 426 427 428
      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),
429
      vote_reopen_(0),
430 431 432 433
      num_done_(0),
      start_(false),
      start_verify_(false),
      stress_test_(stress_test) {
434 435 436 437 438 439
    if (FLAGS_test_batches_snapshots) {
      key_locks_ = nullptr;
      values_ = nullptr;
      fprintf(stdout, "No lock creation because test_batches_snapshots set\n");
      return;
    }
440 441 442 443
    values_ = new uint32_t[max_key_];
    for (long i = 0; i < max_key_; i++) {
      values_[i] = SENTINEL;
    }
444

445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481
    long num_locks = (max_key_ >> log2_keys_per_lock_);
    if (max_key_ & ((1 << log2_keys_per_lock_) - 1)) {
      num_locks ++;
    }
    fprintf(stdout, "Creating %ld locks\n", num_locks);
    key_locks_ = new port::Mutex[num_locks];
  }

  ~SharedState() {
    delete[] values_;
    delete[] key_locks_;
  }

  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
482
  void IncOperated() {
483 484 485 486 487 488 489
    num_populated_++;
  }

  void IncDone() {
    num_done_++;
  }

490 491 492 493
  void IncVotedReopen() {
    vote_reopen_ = (vote_reopen_ + 1) % num_threads_;
  }

494 495 496 497
  bool AllInitialized() const {
    return num_initialized_ >= num_threads_;
  }

A
amayank 已提交
498
  bool AllOperated() const {
499 500 501 502 503 504 505
    return num_populated_ >= num_threads_;
  }

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

506 507 508 509
  bool AllVotedReopen() {
    return (vote_reopen_ == 0);
  }

510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537
  void SetStart() {
    start_ = true;
  }

  void SetStartVerify() {
    start_verify_ = true;
  }

  bool Started() const {
    return start_;
  }

  bool VerifyStarted() const {
    return start_verify_;
  }

  port::Mutex* GetMutexForKey(long key) {
    return &key_locks_[key >> log2_keys_per_lock_];
  }

  void Put(long key, uint32_t value_base) {
    values_[key] = value_base;
  }

  uint32_t Get(long key) const {
    return values_[key];
  }

A
amayank 已提交
538 539 540 541
  void Delete(long key) const {
    values_[key] = SENTINEL;
  }

542 543 544 545 546 547 548 549 550 551 552 553 554
  uint32_t GetSeed() const {
    return seed_;
  }

 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_;
555
  long vote_reopen_;
556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587
  long num_done_;
  bool start_;
  bool start_verify_;
  StressTest* stress_test_;

  uint32_t *values_;
  port::Mutex *key_locks_;

};

// 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)),
        filter_policy_(FLAGS_bloom_bits >= 0
                       ? NewBloomFilterPolicy(FLAGS_bloom_bits)
588
                       : nullptr),
589 590 591
        prefix_extractor_(NewFixedPrefixTransform(
                          FLAGS_test_batches_snapshots ?
                          sizeof(long) : sizeof(long)-1)),
592
        db_(nullptr),
593
        num_times_reopened_(0) {
594 595 596 597 598 599 600
    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-")) {
          FLAGS_env->DeleteFile(std::string(FLAGS_db) + "/" + files[i]);
        }
601
      }
602
      DestroyDB(FLAGS_db, Options());
603 604 605 606 607 608
    }
  }

  ~StressTest() {
    delete db_;
    delete filter_policy_;
609
    delete prefix_extractor_;
610 611 612 613 614 615 616 617 618 619 620 621 622 623
  }

  void Run() {
    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]);
    }
    // Each thread goes through the following states:
A
amayank 已提交
624 625
    // initializing -> wait for others to init -> read/populate/depopulate
    // wait for others to operate -> verify -> done
626 627 628 629 630 631 632

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

633 634 635
      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
636

637 638
      shared.SetStart();
      shared.GetCondVar()->SignalAll();
A
amayank 已提交
639
      while (!shared.AllOperated()) {
640 641 642
        shared.GetCondVar()->Wait();
      }

643
      now = FLAGS_env->NowMicros();
644 645 646 647 648 649 650
      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());
      }
651

652 653 654 655 656 657 658
      shared.SetStartVerify();
      shared.GetCondVar()->SignalAll();
      while (!shared.AllDone()) {
        shared.GetCondVar()->Wait();
      }
    }

659
    for (unsigned int i = 1; i < n; i++) {
660 661 662 663
      threads[0]->stats.Merge(threads[i]->stats);
    }
    threads[0]->stats.Report("Stress Test");

664
    for (unsigned int i = 0; i < n; i++) {
665
      delete threads[i];
666
      threads[i] = nullptr;
667
    }
668
    double now = FLAGS_env->NowMicros();
669 670 671 672
    if (!FLAGS_test_batches_snapshots) {
      fprintf(stdout, "%s Verification successful\n",
              FLAGS_env->TimeToString((uint64_t) now/1000000).c_str());
    }
673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691
    PrintStatistics();
  }

 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
692
    thread->shared->GetStressTest()->OperateDb(thread);
693 694 695

    {
      MutexLock l(shared->GetMutex());
A
amayank 已提交
696 697
      shared->IncOperated();
      if (shared->AllOperated()) {
698 699 700 701 702 703 704
        shared->GetCondVar()->SignalAll();
      }
      while (!shared->VerifyStarted()) {
        shared->GetCondVar()->Wait();
      }
    }

705
    if (!FLAGS_test_batches_snapshots) {
M
Mayank Agarwal 已提交
706
      thread->shared->GetStressTest()->VerifyDb(thread);
707
    }
708 709 710 711 712 713 714 715 716 717 718

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

  }

719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735
  // 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.
  Status MultiPut(ThreadState* thread,
                  const WriteOptions& writeoptions,
                  const Slice& key, const Slice& value, size_t sz) {
    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];
D
Deon Nicholas 已提交
736 737 738 739 740
      if (FLAGS_use_merge_put) {
        batch.Merge(keys[i], value_slices[i]);
      } else {
        batch.Put(keys[i], value_slices[i]);
      }
741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805
    }

    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.
  Status MultiDelete(ThreadState* thread,
                     const WriteOptions& writeoptions,
                     const Slice& key) {
    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();
      batch.Delete(keys[i]);
    }

    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.
  Status MultiGet(ThreadState* thread,
                  const ReadOptions& readoptions,
                  const Slice& key, std::string* value) {
    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];
      s = db_->Get(readoptionscopy, key_slices[i], value);
      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] = "";
806
        thread->stats.AddGets(1, 0);
807 808 809 810 811 812
      } else {
        values[i] = *value;

        char expected_prefix = (keys[i])[0];
        char actual_prefix = (values[i])[0];
        if (actual_prefix != expected_prefix) {
813
          fprintf(stderr, "error expected prefix = %c actual = %c\n",
814 815 816
                  expected_prefix, actual_prefix);
        }
        (values[i])[0] = ' '; // blank out the differing character
817
        thread->stats.AddGets(1, 1);
818 819 820 821 822 823 824
      }
    }
    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]) {
825
        fprintf(stderr, "error : inconsistent values for key %s: %s, %s\n",
826 827 828 829 830 831 832 833 834 835
                key.ToString().c_str(), values[0].c_str(),
                values[i].c_str());
      // we continue after error rather than exiting so that we can
      // find more errors if any
      }
    }

    return s;
  }

836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911
  // Given a prefix P, this does prefix scans for "0"+P, "1"+P,..."9"+P
  // in the same snapshot.  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.
  // ASSUMES that MultiPut was used to put (K, V)
  Status MultiPrefixScan(ThreadState* thread,
                         const ReadOptions& readoptions,
                         const Slice& prefix) {
    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++) {
      prefixes[i] += prefix.ToString();
      prefix_slices[i] = prefixes[i];
      readoptionscopy[i] = readoptions;
      readoptionscopy[i].prefix = &prefix_slices[i];
      readoptionscopy[i].snapshot = snapshot;
      iters[i] = db_->NewIterator(readoptionscopy[i]);
      iters[i]->SeekToFirst();
    }

    int count = 0;
    while (iters[0]->Valid()) {
      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
        assert(iters[i]->Valid());
        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]) {
          fprintf(stderr, "error : inconsistent values for prefix %s: %s, %s\n",
                  prefix.ToString().c_str(), values[0].c_str(),
                  values[i].c_str());
          // 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
      assert(!iters[i]->Valid());
      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;
  }

912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942
  // Given a key K, this creates an iterator which scans to K and then
  // does a random sequence of Next/Prev operations.
  Status MultiIterate(ThreadState* thread,
                      const ReadOptions& readoptions,
                      const Slice& key) {
    Status s;
    const Snapshot* snapshot = db_->GetSnapshot();
    ReadOptions readoptionscopy = readoptions;
    readoptionscopy.snapshot = snapshot;
    unique_ptr<Iterator> iter(db_->NewIterator(readoptionscopy));

    iter->Seek(key);
    for (long i = 0; i < FLAGS_num_iterations && iter->Valid(); i++) {
      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
943
  void OperateDb(ThreadState* thread) {
944 945
    ReadOptions read_opts(FLAGS_verify_checksum, true);
    WriteOptions write_opts;
A
amayank 已提交
946
    char value[100];
947 948 949 950 951 952
    long max_key = thread->shared->GetMaxKey();
    std::string from_db;
    if (FLAGS_sync) {
      write_opts.sync = true;
    }
    write_opts.disableWAL = FLAGS_disable_wal;
953 954 955
    const int prefixBound = (int)FLAGS_readpercent + (int)FLAGS_prefixpercent;
    const int writeBound = prefixBound + (int)FLAGS_writepercent;
    const int delBound = writeBound + (int)FLAGS_delpercent;
956 957

    thread->stats.Start();
A
amayank 已提交
958
    for (long i = 0; i < FLAGS_ops_per_thread; i++) {
959 960
      if(i != 0 && (i % (FLAGS_ops_per_thread / (FLAGS_reopen + 1))) == 0) {
        {
961
          thread->stats.FinishedSingleOp();
962 963 964 965 966 967 968 969 970
          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();
          }
971 972
          // Commenting this out as we don't want to reset stats on each open.
          // thread->stats.Start();
973 974
        }
      }
975

976
      long rand_key = thread->rand.Next() % max_key;
977 978 979 980 981
      std::string keystr = Key(rand_key);
      Slice key = keystr;
      int prob_op = thread->rand.Uniform(100);

      if (prob_op >= 0 && prob_op < (int)FLAGS_readpercent) {
982
        // OPERATION read
983 984 985 986 987 988 989 990 991 992 993 994 995 996 997
        if (!FLAGS_test_batches_snapshots) {
          Status s = db_->Get(read_opts, key, &from_db);
          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 {
          MultiGet(thread, read_opts, key, &from_db);
        }
998 999
      } else if ((int)FLAGS_readpercent <= prob_op && prob_op < prefixBound) {
        // OPERATION prefix scan
1000 1001 1002 1003
        // keys are longs (e.g., 8 bytes), so we let prefixes be
        // everything except the last byte.  So there will be 2^8=256
        // keys per prefix.
        Slice prefix = Slice(key.data(), key.size() - 1);
1004
        if (!FLAGS_test_batches_snapshots) {
1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018
          read_opts.prefix = &prefix;
          Iterator* iter = db_->NewIterator(read_opts);
          int count = 0;
          for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
            assert(iter->key().starts_with(prefix));
            count++;
          }
          assert(count <= 256);
          if (iter->status().ok()) {
            thread->stats.AddPrefixes(1, count);
          } else {
            thread->stats.AddErrors(1);
          }
          delete iter;
1019
        } else {
1020
          MultiPrefixScan(thread, read_opts, prefix);
A
amayank 已提交
1021
        }
1022 1023
      } else if (prefixBound <= prob_op && prob_op < writeBound) {
        // OPERATION write
1024 1025 1026
        uint32_t value_base = thread->rand.Next();
        size_t sz = GenerateValue(value_base, value, sizeof(value));
        Slice v(value, sz);
1027
        if (!FLAGS_test_batches_snapshots) {
1028 1029
          MutexLock l(thread->shared->GetMutexForKey(rand_key));
          if (FLAGS_verify_before_write) {
1030 1031 1032 1033 1034 1035 1036 1037 1038
            std::string keystr = Key(rand_key);
            Slice k = keystr;
            Status s = db_->Get(read_opts, k, &from_db);
            VerifyValue(rand_key,
                        read_opts,
                        *(thread->shared),
                        from_db,
                        s,
                        true);
1039 1040
          }
          thread->shared->Put(rand_key, value_base);
D
Deon Nicholas 已提交
1041 1042 1043 1044 1045
          if (FLAGS_use_merge_put) {
            db_->Merge(write_opts, key, v);
          } else {
            db_->Put(write_opts, key, v);
          }
1046 1047 1048
          thread->stats.AddBytesForWrites(1, sz);
        } else {
          MultiPut(thread, write_opts, key, v, sz);
1049 1050
        }
        PrintKeyValue(rand_key, value, sz);
1051 1052
      } else if (writeBound <= prob_op && prob_op < delBound) {
        // OPERATION delete
1053 1054 1055 1056 1057 1058 1059 1060
        if (!FLAGS_test_batches_snapshots) {
          MutexLock l(thread->shared->GetMutexForKey(rand_key));
          thread->shared->Delete(rand_key);
          db_->Delete(write_opts, key);
          thread->stats.AddDeletes(1);
        } else {
          MultiDelete(thread, write_opts, key);
        }
1061 1062 1063
      } else {
        // OPERATION iterate
        MultiIterate(thread, read_opts, key);
1064
      }
1065 1066
      thread->stats.FinishedSingleOp();
    }
1067

1068 1069 1070
    thread->stats.Stop();
  }

M
Mayank Agarwal 已提交
1071
  void VerifyDb(ThreadState* thread) const {
1072
    ReadOptions options(FLAGS_verify_checksum, true);
M
Mayank Agarwal 已提交
1073 1074
    const SharedState& shared = *(thread->shared);
    static const long max_key = shared.GetMaxKey();
1075
    static const long keys_per_thread = max_key / shared.GetNumThreads();
M
Mayank Agarwal 已提交
1076
    long start = keys_per_thread * thread->tid;
1077
    long end = start + keys_per_thread;
M
Mayank Agarwal 已提交
1078
    if (thread->tid == shared.GetNumThreads() - 1) {
1079 1080
      end = max_key;
    }
M
Mayank Agarwal 已提交
1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101
    if (!thread->rand.OneIn(2)) {
      // Use iterator to verify this range
      unique_ptr<Iterator> iter(db_->NewIterator(options));
      iter->Seek(Key(start));
      for (long i = start; i < end; i++) {
        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) {
            VerificationAbort("An out of range key was found", i);
          }
        } else {
          // The iterator found no value for the key in question, so do not
          // move to the next item in the iterator
1102 1103
          s = Status::NotFound(Slice());
        }
M
Mayank Agarwal 已提交
1104 1105 1106 1107
        VerifyValue(i, options, shared, from_db, s, true);
        if (from_db.length()) {
          PrintKeyValue(i, from_db.data(), from_db.length());
        }
1108
      }
M
Mayank Agarwal 已提交
1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120
    }
    else {
      // Use Get to verify this range
      for (long i = start; i < end; i++) {
        std::string from_db;
        std::string keystr = Key(i);
        Slice k = keystr;
        Status s = db_->Get(options, k, &from_db);
        VerifyValue(i, options, shared, from_db, s, true);
        if (from_db.length()) {
          PrintKeyValue(i, from_db.data(), from_db.length());
        }
1121 1122 1123 1124
      }
    }
  }

1125
  void VerificationAbort(std::string msg, long key) const {
A
amayank 已提交
1126
    fprintf(stderr, "Verification failed for key %ld: %s\n",
1127 1128 1129 1130
            key, msg.c_str());
    exit(1);
  }

1131 1132 1133 1134 1135 1136 1137
  void VerifyValue(long key,
                   const ReadOptions &opts,
                   const SharedState &shared,
                   const std::string &value_from_db,
                   Status s,
                   bool strict=false) const {
    // compare value_from_db with the value in the shared state
A
amayank 已提交
1138
    char value[100];
1139 1140 1141 1142 1143
    uint32_t value_base = shared.Get(key);
    if (value_base == SharedState::SENTINEL && !strict) {
      return;
    }

1144
    if (s.ok()) {
1145 1146 1147
      if (value_base == SharedState::SENTINEL) {
        VerificationAbort("Unexpected value found", key);
      }
A
amayank 已提交
1148
      size_t sz = GenerateValue(value_base, value, sizeof(value));
1149
      if (value_from_db.length() != sz) {
1150 1151
        VerificationAbort("Length of value read is not equal", key);
      }
1152
      if (memcmp(value_from_db.data(), value, sz) != 0) {
1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163
        VerificationAbort("Contents of value read don't match", key);
      }
    } else {
      if (value_base != SharedState::SENTINEL) {
        VerificationAbort("Value not found", key);
      }
    }
  }

  static void PrintKeyValue(uint32_t key, const char *value, size_t sz) {
    if (!FLAGS_verbose) return;
1164
    fprintf(stdout, "%u ==> (%u) ", key, (unsigned int)sz);
1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177
    for (size_t i=0; i<sz; i++) {
      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);
    }
1178
    v[value_sz] = '\0';
1179 1180 1181 1182 1183 1184 1185
    return value_sz; // the size of the value set.
  }

  void PrintEnv() const {
    fprintf(stdout, "LevelDB version     : %d.%d\n",
            kMajorVersion, kMinorVersion);
    fprintf(stdout, "Number of threads   : %d\n", FLAGS_threads);
1186
    fprintf(stdout, "Ops per thread      : %d\n", FLAGS_ops_per_thread);
1187 1188 1189 1190 1191
    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());
A
amayank 已提交
1192
    fprintf(stdout, "Read percentage     : %d\n", FLAGS_readpercent);
1193 1194 1195
    fprintf(stdout, "Prefix percentage   : %d\n", FLAGS_prefixpercent);
    fprintf(stdout, "Write percentage    : %d\n", FLAGS_writepercent);
    fprintf(stdout, "Delete percentage   : %d\n", FLAGS_delpercent);
1196
    fprintf(stdout, "Iterate percentage  : %d\n", FLAGS_iterpercent);
1197
    fprintf(stdout, "Write-buffer-size   : %d\n", FLAGS_write_buffer_size);
1198
    fprintf(stdout, "Iterations          : %d\n", FLAGS_num_iterations);
1199
    fprintf(stdout, "Max key             : %ld\n", FLAGS_max_key);
X
Xing Jin 已提交
1200 1201
    fprintf(stdout, "Ratio #ops/#keys    : %f\n",
            (1.0 * FLAGS_ops_per_thread * FLAGS_threads)/FLAGS_max_key);
1202
    fprintf(stdout, "Num times DB reopens: %d\n", FLAGS_reopen);
1203 1204
    fprintf(stdout, "Batches/snapshots   : %d\n",
            FLAGS_test_batches_snapshots);
X
Xing Jin 已提交
1205
    fprintf(stdout, "Purge redundant %%   : %d\n",
1206
            FLAGS_purge_redundant_percent);
1207
    fprintf(stdout, "Deletes use filter  : %d\n",
1208
            FLAGS_filter_deletes);
1209
    fprintf(stdout, "Num keys per lock   : %d\n",
1210 1211
            1 << FLAGS_log2_keys_per_lock);

1212
    const char* compression = "";
1213
    switch (FLAGS_compression_type) {
1214
      case rocksdb::kNoCompression:
1215
        compression = "none";
1216
        break;
1217
      case rocksdb::kSnappyCompression:
1218
        compression = "snappy";
1219
        break;
1220
      case rocksdb::kZlibCompression:
1221
        compression = "zlib";
1222
        break;
1223
      case rocksdb::kBZip2Compression:
1224
        compression = "bzip2";
1225 1226 1227 1228
        break;
    }

    fprintf(stdout, "Compression         : %s\n", compression);
J
Jim Paton 已提交
1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247

    const char* memtablerep = "";
    switch (FLAGS_rep_factory) {
      case kSkipList:
        memtablerep = "skip_list";
        break;
      case kPrefixHash:
        memtablerep = "prefix_hash";
        break;
      case kUnsorted:
        memtablerep = "unsorted";
        break;
      case kVectorRep:
        memtablerep = "vector";
        break;
    }

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

1248 1249 1250 1251
    fprintf(stdout, "------------------------------------------------\n");
  }

  void Open() {
1252
    assert(db_ == nullptr);
1253 1254 1255
    Options options;
    options.block_cache = cache_;
    options.write_buffer_size = FLAGS_write_buffer_size;
1256
    options.max_write_buffer_number = FLAGS_max_write_buffer_number;
1257 1258
    options.min_write_buffer_number_to_merge =
      FLAGS_min_write_buffer_number_to_merge;
1259
    options.max_background_compactions = FLAGS_max_background_compactions;
1260
    options.compaction_style = FLAGS_compaction_style;
1261 1262
    options.block_size = FLAGS_block_size;
    options.filter_policy = filter_policy_;
1263
    options.prefix_extractor = prefix_extractor_;
1264 1265 1266 1267 1268
    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;
1269
    options.allow_mmap_reads = FLAGS_use_mmap_reads;
1270
    rocksdb_kill_odds = FLAGS_kill_random_test;
1271 1272 1273 1274 1275 1276 1277 1278
    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 =
        FLAGS_max_bytes_for_level_multiplier;
    options.level0_stop_writes_trigger = FLAGS_level0_stop_writes_trigger;
    options.level0_slowdown_writes_trigger =
      FLAGS_level0_slowdown_writes_trigger;
1279 1280
    options.level0_file_num_compaction_trigger =
      FLAGS_level0_file_num_compaction_trigger;
1281 1282 1283
    options.compression = FLAGS_compression_type;
    options.create_if_missing = true;
    options.disable_seek_compaction = FLAGS_disable_seek_compaction;
1284 1285
    options.delete_obsolete_files_period_micros =
      FLAGS_delete_obsolete_files_period_micros;
A
Abhishek Kona 已提交
1286
    options.max_manifest_file_size = 1024;
1287
    options.filter_deletes = FLAGS_filter_deletes;
J
Jim Paton 已提交
1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312
    if ((FLAGS_prefix_size == 0) == (FLAGS_rep_factory == kPrefixHash)) {
      fprintf(stderr,
            "prefix_size should be non-zero iff memtablerep == prefix_hash\n");
      exit(1);
    }
    switch (FLAGS_rep_factory) {
      case kPrefixHash:
        options.memtable_factory.reset(
          new PrefixHashRepFactory(NewFixedPrefixTransform(FLAGS_prefix_size))
        );
        break;
      case kUnsorted:
        options.memtable_factory.reset(
          new UnsortedRepFactory()
        );
        break;
      case kSkipList:
        // no need to do anything
        break;
      case kVectorRep:
        options.memtable_factory.reset(
          new VectorRepFactory()
        );
        break;
    }
1313 1314 1315 1316
    static Random purge_percent(1000); // no benefit from non-determinism here
    if (purge_percent.Uniform(100) < FLAGS_purge_redundant_percent - 1) {
      options.purge_redundant_kvs_while_flush = false;
    }
1317

D
Deon Nicholas 已提交
1318
    if (FLAGS_use_merge_put) {
1319
      options.merge_operator = MergeOperators::CreatePutOperator();
D
Deon Nicholas 已提交
1320 1321
    }

1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339
    // set universal style compaction configurations, if applicable
    if (FLAGS_universal_size_ratio != 0) {
      options.compaction_options_universal.size_ratio =
        FLAGS_universal_size_ratio;
    }
    if (FLAGS_universal_min_merge_width != 0) {
      options.compaction_options_universal.min_merge_width =
        FLAGS_universal_min_merge_width;
    }
    if (FLAGS_universal_max_merge_width != 0) {
      options.compaction_options_universal.max_merge_width =
        FLAGS_universal_max_merge_width;
    }
    if (FLAGS_universal_max_size_amplification_percent != 0) {
      options.compaction_options_universal.max_size_amplification_percent =
        FLAGS_universal_max_size_amplification_percent;
    }

1340 1341
    fprintf(stdout, "DB path: [%s]\n", FLAGS_db);

1342 1343 1344 1345
    Status s;
    if (FLAGS_ttl == -1) {
      s = DB::Open(options, FLAGS_db, &db_);
    } else {
1346 1347
      s = UtilityDB::OpenTtlDB(options, FLAGS_db, &sdb_, FLAGS_ttl);
      db_ = sdb_;
1348
    }
1349 1350 1351 1352 1353 1354
    if (!s.ok()) {
      fprintf(stderr, "open error: %s\n", s.ToString().c_str());
      exit(1);
    }
  }

1355
  void Reopen() {
1356 1357
    // do not close the db. Just delete the lock file. This
    // simulates a crash-recovery kind of situation.
1358 1359 1360 1361 1362
    if (FLAGS_ttl != -1) {
      ((DBWithTTL*) db_)->TEST_Destroy_DBWithTtl();
    } else {
      ((DBImpl*) db_)->TEST_Destroy_DBImpl();
    }
1363
    db_ = nullptr;
1364 1365 1366 1367 1368 1369

    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_);
1370 1371 1372
    Open();
  }

1373 1374
  void PrintStatistics() {
    if (dbstats) {
1375
      fprintf(stdout, "STATISTICS:\n%s\n", dbstats->ToString().c_str());
1376 1377 1378 1379
    }
  }

 private:
1380
  shared_ptr<Cache> cache_;
1381
  const FilterPolicy* filter_policy_;
1382
  const SliceTransform* prefix_extractor_;
1383
  DB* db_;
1384
  StackableDB* sdb_;
1385
  int num_times_reopened_;
1386 1387
};

1388
}  // namespace rocksdb
1389 1390

int main(int argc, char** argv) {
1391 1392
  FLAGS_write_buffer_size = rocksdb::Options().write_buffer_size;
  FLAGS_max_write_buffer_number = rocksdb::Options().max_write_buffer_number;
1393
  FLAGS_min_write_buffer_number_to_merge =
1394 1395
    rocksdb::Options().min_write_buffer_number_to_merge;
  FLAGS_open_files = rocksdb::Options().max_open_files;
1396
  FLAGS_max_background_compactions =
1397
    rocksdb::Options().max_background_compactions;
1398
  FLAGS_compaction_style =
1399
    rocksdb::Options().compaction_style;
1400
  FLAGS_level0_file_num_compaction_trigger =
1401
    rocksdb::Options().level0_file_num_compaction_trigger;
1402
  FLAGS_level0_slowdown_writes_trigger =
1403
    rocksdb::Options().level0_slowdown_writes_trigger;
1404
  FLAGS_level0_stop_writes_trigger =
1405
    rocksdb::Options().level0_stop_writes_trigger;
1406
  // Compression test code above refers to FLAGS_block_size
1407
  FLAGS_block_size = rocksdb::Options().block_size;
1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430
  std::string default_db_path;

  for (int i = 1; i < argc; i++) {
    int n;
    uint32_t u;
    long l;
    char junk;
    char hdfsname[2048];

    if (sscanf(argv[i], "--seed=%uf%c", &u, &junk) == 1) {
      FLAGS_seed = u;
    } else if (sscanf(argv[i], "--max_key=%ld%c", &l, &junk) == 1) {
      FLAGS_max_key = l;
    } else if (sscanf(argv[i], "--log2_keys_per_lock=%u%c", &u, &junk) == 1) {
      FLAGS_log2_keys_per_lock = u;
    } else if (sscanf(argv[i], "--ops_per_thread=%u%c", &u, &junk) == 1) {
      FLAGS_ops_per_thread = u;
    } else if (sscanf(argv[i], "--verbose=%d%c", &n, &junk) == 1 &&
               (n == 0 || n == 1)) {
      FLAGS_verbose = n;
    } else if (sscanf(argv[i], "--histogram=%d%c", &n, &junk) == 1 &&
               (n == 0 || n == 1)) {
      FLAGS_histogram = n;
1431 1432 1433
    } else if (sscanf(argv[i], "--destroy_db_initially=%d%c", &n, &junk) == 1 &&
               (n == 0 || n == 1)) {
      FLAGS_destroy_db_initially = n;
1434 1435 1436
    } else if (sscanf(argv[i], "--verify_before_write=%d%c", &n, &junk) == 1 &&
               (n == 0 || n == 1)) {
      FLAGS_verify_before_write = n;
1437 1438 1439
    } else if (sscanf(argv[i], "--test_batches_snapshots=%d%c", &n, &junk) == 1
               && (n == 0 || n == 1)) {
      FLAGS_test_batches_snapshots = n;
1440 1441
    } else if (sscanf(argv[i], "--threads=%d%c", &n, &junk) == 1) {
      FLAGS_threads = n;
1442 1443
    } else if (sscanf(argv[i], "--ttl=%d%c", &n, &junk) == 1) {
      FLAGS_ttl = n;
1444 1445 1446 1447
    } else if (sscanf(argv[i], "--value_size_mult=%d%c", &n, &junk) == 1) {
      FLAGS_value_size_mult = n;
    } else if (sscanf(argv[i], "--write_buffer_size=%d%c", &n, &junk) == 1) {
      FLAGS_write_buffer_size = n;
1448 1449
    } else if (sscanf(argv[i], "--max_write_buffer_number=%d%c", &n, &junk) == 1) {
      FLAGS_max_write_buffer_number = n;
1450 1451 1452
    } else if (sscanf(argv[i], "--min_write_buffer_number_to_merge=%d%c",
               &n, &junk) == 1) {
      FLAGS_min_write_buffer_number_to_merge = n;
1453 1454
    } else if (sscanf(argv[i], "--max_background_compactions=%d%c", &n, &junk) == 1) {
      FLAGS_max_background_compactions = n;
1455
    } else if (sscanf(argv[i], "--compaction_style=%d%c", &n, &junk) == 1) {
1456
      FLAGS_compaction_style = (rocksdb::CompactionStyle)n;
1457 1458 1459 1460
    } else if (sscanf(argv[i], "--cache_size=%ld%c", &l, &junk) == 1) {
      FLAGS_cache_size = l;
    } else if (sscanf(argv[i], "--block_size=%d%c", &n, &junk) == 1) {
      FLAGS_block_size = n;
1461 1462
    } else if (sscanf(argv[i], "--reopen=%d%c", &n, &junk) == 1 && n >= 0) {
      FLAGS_reopen = n;
1463 1464 1465 1466 1467 1468 1469 1470 1471
    } else if (sscanf(argv[i], "--bloom_bits=%d%c", &n, &junk) == 1) {
      FLAGS_bloom_bits = n;
    } else if (sscanf(argv[i], "--open_files=%d%c", &n, &junk) == 1) {
      FLAGS_open_files = n;
    } else if (strncmp(argv[i], "--db=", 5) == 0) {
      FLAGS_db = argv[i] + 5;
    } else if (sscanf(argv[i], "--verify_checksum=%d%c", &n, &junk) == 1 &&
               (n == 0 || n == 1)) {
      FLAGS_verify_checksum = n;
1472 1473 1474
    } else if (sscanf(argv[i], "--mmap_read=%d%c", &n, &junk) == 1 &&
               (n == 0 || n == 1)) {
      FLAGS_use_mmap_reads = n;
1475 1476 1477
    } else if (sscanf(argv[i], "--statistics=%d%c", &n, &junk) == 1 &&
               (n == 0 || n == 1)) {
      if (n == 1) {
1478
        dbstats = rocksdb::CreateDBStatistics();
1479 1480 1481 1482
      }
    } else if (sscanf(argv[i], "--sync=%d%c", &n, &junk) == 1 &&
               (n == 0 || n == 1)) {
      FLAGS_sync = n;
A
amayank 已提交
1483 1484 1485
    } else if (sscanf(argv[i], "--readpercent=%d%c", &n, &junk) == 1 &&
               (n >= 0 && n <= 100)) {
      FLAGS_readpercent = n;
1486 1487 1488 1489 1490 1491
    } else if (sscanf(argv[i], "--prefixpercent=%d%c", &n, &junk) == 1 &&
               (n >= 0 && n <= 100)) {
      FLAGS_prefixpercent = n;
    } else if (sscanf(argv[i], "--writepercent=%d%c", &n, &junk) == 1 &&
               (n >= 0 && n <= 100)) {
      FLAGS_writepercent = n;
1492 1493 1494
    } else if (sscanf(argv[i], "--iterpercent=%d%c", &n, &junk) == 1 &&
               (n >= 0 && n <= 100)) {
      FLAGS_iterpercent = n;
A
amayank 已提交
1495 1496 1497
    } else if (sscanf(argv[i], "--delpercent=%d%c", &n, &junk) == 1 &&
               (n >= 0 && n <= 100)) {
      FLAGS_delpercent = n;
1498 1499
    } else if (sscanf(argv[i], "--numiterations=%u%c", &u, &junk) == 1) {
      FLAGS_num_iterations = u;
1500 1501 1502 1503 1504 1505
    } else if (sscanf(argv[i], "--disable_data_sync=%d%c", &n, &junk) == 1 &&
        (n == 0 || n == 1)) {
      FLAGS_disable_data_sync = n;
    } else if (sscanf(argv[i], "--use_fsync=%d%c", &n, &junk) == 1 &&
        (n == 0 || n == 1)) {
      FLAGS_use_fsync = n;
1506 1507 1508
    } else if (sscanf(argv[i], "--kill_random_test=%d%c", &n, &junk) == 1 &&
               (n >= 0)) {
      FLAGS_kill_random_test = n;
1509 1510 1511 1512
    } else if (sscanf(argv[i], "--disable_wal=%d%c", &n, &junk) == 1 &&
        (n == 0 || n == 1)) {
      FLAGS_disable_wal = n;
    } else if (sscanf(argv[i], "--hdfs=%s", hdfsname) == 1) {
1513
      FLAGS_env  = new rocksdb::HdfsEnv(hdfsname);
1514 1515 1516 1517 1518 1519 1520
    } else if (sscanf(argv[i], "--target_file_size_base=%d%c",
        &n, &junk) == 1) {
      FLAGS_target_file_size_base = n;
    } else if ( sscanf(argv[i], "--target_file_size_multiplier=%d%c",
        &n, &junk) == 1) {
      FLAGS_target_file_size_multiplier = n;
    } else if (
1521 1522
        sscanf(argv[i], "--max_bytes_for_level_base=%ld%c", &l, &junk) == 1) {
      FLAGS_max_bytes_for_level_base = l;
1523 1524 1525 1526 1527 1528 1529 1530 1531
    } else if (sscanf(argv[i], "--max_bytes_for_level_multiplier=%d%c",
        &n, &junk) == 1) {
      FLAGS_max_bytes_for_level_multiplier = n;
    } else if (sscanf(argv[i],"--level0_stop_writes_trigger=%d%c",
        &n, &junk) == 1) {
      FLAGS_level0_stop_writes_trigger = n;
    } else if (sscanf(argv[i],"--level0_slowdown_writes_trigger=%d%c",
        &n, &junk) == 1) {
      FLAGS_level0_slowdown_writes_trigger = n;
1532 1533 1534
    } else if (sscanf(argv[i],"--level0_file_num_compaction_trigger=%d%c",
        &n, &junk) == 1) {
      FLAGS_level0_file_num_compaction_trigger = n;
1535 1536 1537
    } else if (strncmp(argv[i], "--compression_type=", 19) == 0) {
      const char* ctype = argv[i] + 19;
      if (!strcasecmp(ctype, "none"))
1538
        FLAGS_compression_type = rocksdb::kNoCompression;
1539
      else if (!strcasecmp(ctype, "snappy"))
1540
        FLAGS_compression_type = rocksdb::kSnappyCompression;
1541
      else if (!strcasecmp(ctype, "zlib"))
1542
        FLAGS_compression_type = rocksdb::kZlibCompression;
1543
      else if (!strcasecmp(ctype, "bzip2"))
1544
        FLAGS_compression_type = rocksdb::kBZip2Compression;
1545 1546 1547
      else {
        fprintf(stdout, "Cannot parse %s\n", argv[i]);
      }
J
Jim Paton 已提交
1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560
    } else if (strncmp(argv[i], "--memtablerep=", 14) == 0) {
      const char* ctype = argv[i] + 14;
      if (!strcasecmp(ctype, "skip_list"))
        FLAGS_rep_factory = kSkipList;
      else if (!strcasecmp(ctype, "prefix_hash"))
        FLAGS_rep_factory = kPrefixHash;
      else if (!strcasecmp(ctype, "unsorted"))
        FLAGS_rep_factory = kUnsorted;
      else if (!strcasecmp(ctype, "vector"))
        FLAGS_rep_factory = kVectorRep;
      else {
        fprintf(stdout, "Cannot parse %s\n", argv[i]);
      }
1561 1562 1563
    } else if (sscanf(argv[i], "--disable_seek_compaction=%d%c", &n, &junk) == 1
        && (n == 0 || n == 1)) {
      FLAGS_disable_seek_compaction = n;
1564 1565 1566
    } else if (sscanf(argv[i], "--delete_obsolete_files_period_micros=%ld%c",
                      &l, &junk) == 1) {
      FLAGS_delete_obsolete_files_period_micros = n;
1567 1568 1569
    } else if (sscanf(argv[i], "--purge_redundant_percent=%d%c", &n, &junk) == 1
        && (n >= 0 && n <= 100)) {
      FLAGS_purge_redundant_percent = n;
1570
    } else if (sscanf(argv[i], "--filter_deletes=%d%c", &n, &junk)
1571
        == 1 && (n == 0 || n == 1)) {
1572
      FLAGS_filter_deletes = n;
J
Jim Paton 已提交
1573 1574 1575
    } else if (sscanf(argv[i], "--prefix_size=%d%c", &n, &junk) == 1 &&
               n >= 0 && n < 2000000000) {
      FLAGS_prefix_size = n;
D
Deon Nicholas 已提交
1576 1577 1578
    } else if (sscanf(argv[i], "--use_merge=%d%c", &n, &junk)
        == 1 && (n == 0 || n == 1)) {
      FLAGS_use_merge_put = n;
1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591
    } else if (sscanf(argv[i], "--universal_size_ratio=%d%c",
                      &n, &junk) == 1) {
      FLAGS_universal_size_ratio = n;
    } else if (sscanf(argv[i], "--universal_min_merge_width=%d%c",
                      &n, &junk) == 1) {
      FLAGS_universal_min_merge_width = n;
    } else if (sscanf(argv[i], "--universal_max_merge_width=%d%c",
                      &n, &junk) == 1) {
      FLAGS_universal_max_merge_width = n;
    } else if (sscanf(argv[i],
                      "--universal_max_size_amplification_percent=%d%c",
                      &n, &junk) == 1) {
      FLAGS_universal_max_size_amplification_percent = n;
1592 1593 1594 1595 1596 1597
    } else {
      fprintf(stderr, "Invalid flag '%s'\n", argv[i]);
      exit(1);
    }
  }

1598 1599 1600 1601
  // The number of background threads should be at least as much the
  // max number of concurrent compactions.
  FLAGS_env->SetBackgroundThreads(FLAGS_max_background_compactions);

1602
  if ((FLAGS_readpercent + FLAGS_prefixpercent +
1603 1604 1605
       FLAGS_writepercent + FLAGS_delpercent + FLAGS_iterpercent) != 100) {
      fprintf(stderr,
              "Error: Read+Prefix+Write+Delete+Iterate percents != 100!\n");
A
amayank 已提交
1606 1607
      exit(1);
  }
1608 1609 1610 1611
  if (FLAGS_disable_wal == 1 && FLAGS_reopen > 0) {
      fprintf(stderr, "Error: Db cannot reopen safely with disable_wal set!\n");
      exit(1);
  }
1612 1613 1614 1615 1616 1617 1618
  if ((unsigned)FLAGS_reopen >= FLAGS_ops_per_thread) {
      fprintf(stderr, "Error: #DB-reopens should be < ops_per_thread\n"
        "Provided reopens = %d and ops_per_thread = %u\n", FLAGS_reopen,
        FLAGS_ops_per_thread);
      exit(1);
  }

1619
  // Choose a location for the test database if none given with --db=<path>
1620
  if (FLAGS_db == nullptr) {
1621
      rocksdb::Env::Default()->GetTestDirectory(&default_db_path);
1622 1623 1624 1625
      default_db_path += "/dbstress";
      FLAGS_db = default_db_path.c_str();
  }

1626
  rocksdb::StressTest stress;
1627 1628 1629
  stress.Run();
  return 0;
}