db_stress.cc 44.9 KB
Newer Older
1 2 3
// 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.
4
//
X
Xing Jin 已提交
5 6 7 8 9 10 11 12 13 14 15 16
// 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
17

18 19 20 21 22 23 24
#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"
#include "leveldb/cache.h"
25
#include "utilities/utility_db.h"
26 27 28 29
#include "leveldb/env.h"
#include "leveldb/write_batch.h"
#include "leveldb/statistics.h"
#include "port/port.h"
30
#include "util/coding.h"
31 32 33 34 35
#include "util/crc32c.h"
#include "util/histogram.h"
#include "util/mutexlock.h"
#include "util/random.h"
#include "util/testutil.h"
36 37
#include "util/logging.h"
#include "utilities/ttl/db_ttl.h"
38
#include "hdfs/env_hdfs.h"
D
Deon Nicholas 已提交
39
#include "utilities/merge_operators.h"
40 41 42 43 44 45 46

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
47 48 49 50 51 52 53 54 55
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)
56
// (b) No long validation at the end (more speed up)
X
Xing Jin 已提交
57
// (c) Test snapshot and atomicity of batch writes
58 59
static bool FLAGS_test_batches_snapshots = false;

60 61 62
// Number of concurrent threads to run.
static int FLAGS_threads = 32;

63 64
// 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
65
// values don't fail
66 67
static int FLAGS_ttl = -1;

68 69 70 71 72 73 74 75
// 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;

76 77 78
// Destroys the database dir before start if this is true
static bool FLAGS_destroy_db_initially = true;

79 80 81 82 83 84
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
85
// The number of in-memory memtables.
86 87 88 89 90 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;

// 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;

95
// This is initialized to default value of false
96
static leveldb::CompactionStyle FLAGS_compaction_style = leveldb::kCompactionStyleLevel;
97

98 99 100 101 102 103
// 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;

104 105 106
// Number of times database reopens
static int FLAGS_reopen = 10;

107 108 109 110 111 112 113 114
// 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.
115
static const char* FLAGS_db = nullptr;
116 117 118 119

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

120 121 122
// Allow reads to occur via mmap-ing files
static bool FLAGS_use_mmap_reads = leveldb::EnvOptions().use_mmap_reads;

123
// Database statistics
A
Abhishek Kona 已提交
124
static std::shared_ptr<leveldb::Statistics> dbstats;
125 126 127 128 129 130 131 132 133 134

// 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;

135 136 137 138
// If non-zero, kill at various points in source code with probability 1/this
static int FLAGS_kill_random_test = 0;
extern int leveldb_kill_odds;

139 140 141
// If true, do not write WAL for write.
static bool FLAGS_disable_wal = false;

X
Xing Jin 已提交
142
// Target level-1 file size for compaction
143 144
static int FLAGS_target_file_size_base = 64 * KB;

X
Xing Jin 已提交
145
// A multiplier to compute targe level-N file size (N >= 2)
146 147
static int FLAGS_target_file_size_multiplier = 1;

X
Xing Jin 已提交
148
// Max bytes for level-1
149
static uint64_t FLAGS_max_bytes_for_level_base = 256 * KB;
150

X
Xing Jin 已提交
151
// A multiplier to compute max bytes for level-N (N >= 2)
152 153 154 155 156 157 158 159
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
160 161 162
// Ratio of reads to total workload (expressed as a percentage)
static unsigned int FLAGS_readpercent = 10;

163 164 165 166 167 168
// Ratio of prefix iterators to total workload (expressed as a percentage)
static unsigned int FLAGS_prefixpercent = 25;

// Ratio of deletes to total workload (expressed as a percentage)
static unsigned int FLAGS_writepercent = 50;

A
amayank 已提交
169
// Ratio of deletes to total workload (expressed as a percentage)
170
static unsigned int FLAGS_delpercent = 15;
171 172 173 174

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

175 176 177 178 179
// 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;

180 181 182 183 184 185 186 187 188 189 190 191 192
// Algorithm to use to compress the database
static enum leveldb::CompressionType FLAGS_compression_type =
    leveldb::kSnappyCompression;

// posix or hdfs environment
static leveldb::Env* FLAGS_env = leveldb::Env::Default();

// 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

193 194 195
// Percentage of times we want to purge redundant keys in memory before flushing
static uint32_t FLAGS_purge_redundant_percent = 50;

196
// On true, deletes use KeyMayExist to drop the delete if key not present
197
static bool FLAGS_filter_deletes = false;
198

199 200 201
// Level0 compaction start trigger
static int FLAGS_level0_file_num_compaction_trigger = 0;

D
Deon Nicholas 已提交
202 203 204
// On true, replaces all writes with a Merge that behaves like a Put
static bool FLAGS_use_merge_put = false;

205 206
namespace leveldb {

207 208 209 210 211 212 213 214 215 216 217 218 219
// 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;
}

220 221 222 223 224 225 226 227 228
class StressTest;
namespace {

class Stats {
 private:
  double start_;
  double finish_;
  double seconds_;
  long done_;
229 230
  long gets_;
  long prefixes_;
231
  long writes_;
A
amayank 已提交
232
  long deletes_;
233
  long iterator_size_sums_;
234 235
  long founds_;
  long errors_;
236 237 238
  int next_report_;
  size_t bytes_;
  double last_op_finish_;
239
  HistogramImpl hist_;
240 241 242 243 244 245 246 247

 public:
  Stats() { }

  void Start() {
    next_report_ = 100;
    hist_.Clear();
    done_ = 0;
248 249
    gets_ = 0;
    prefixes_ = 0;
250
    writes_ = 0;
A
amayank 已提交
251
    deletes_ = 0;
252
    iterator_size_sums_ = 0;
253 254
    founds_ = 0;
    errors_ = 0;
255 256 257 258 259 260 261 262 263 264
    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_;
265 266
    gets_ += other.gets_;
    prefixes_ += other.prefixes_;
267
    writes_ += other.writes_;
A
amayank 已提交
268
    deletes_ += other.deletes_;
269
    iterator_size_sums_ += other.iterator_size_sums_;
270 271
    founds_ += other.founds_;
    errors_ += other.errors_;
272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288
    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) {
289
        fprintf(stdout, "long op: %.1f micros%30s\r", micros, "");
290 291 292 293 294 295 296 297 298 299 300 301 302
      }
      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;
303
      fprintf(stdout, "... finished %ld ops%30s\r", done_, "");
304 305 306
    }
  }

307 308 309 310 311
  void AddBytesForWrites(int nwrites, size_t nbytes) {
    writes_ += nwrites;
    bytes_ += nbytes;
  }

312 313 314
  void AddGets(int ngets, int nfounds) {
    founds_ += nfounds;
    gets_ += ngets;
315 316
  }

317 318 319 320 321 322 323 324 325
  void AddPrefixes(int nprefixes, int count) {
    prefixes_ += nprefixes;
    iterator_size_sums_ += count;
  }

  void AddDeletes(int n) {
    deletes_ += n;
  }

326 327
  void AddErrors(int n) {
    errors_ += n;
A
amayank 已提交
328 329
  }

330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346
  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_);
347
    fprintf(stdout, "%-12s: Wrote %ld times\n", "", writes_);
A
amayank 已提交
348
    fprintf(stdout, "%-12s: Deleted %ld times\n", "", deletes_);
X
Xing Jin 已提交
349 350
    fprintf(stdout, "%-12s: %ld read and %ld found the key\n", "",
            gets_, founds_);
351 352 353
    fprintf(stdout, "%-12s: Prefix scanned %ld times\n", "", prefixes_);
    fprintf(stdout, "%-12s: Iterator size sum is %ld\n", "",
            iterator_size_sums_);
354
    fprintf(stdout, "%-12s: Got errors %ld times\n", "", errors_);
355 356 357 358 359 360 361 362 363 364 365 366 367

    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;

368
  explicit SharedState(StressTest* stress_test) :
369 370 371 372 373 374 375
      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),
376
      vote_reopen_(0),
377 378 379 380
      num_done_(0),
      start_(false),
      start_verify_(false),
      stress_test_(stress_test) {
381 382 383 384 385 386
    if (FLAGS_test_batches_snapshots) {
      key_locks_ = nullptr;
      values_ = nullptr;
      fprintf(stdout, "No lock creation because test_batches_snapshots set\n");
      return;
    }
387 388 389 390
    values_ = new uint32_t[max_key_];
    for (long i = 0; i < max_key_; i++) {
      values_[i] = SENTINEL;
    }
391

392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428
    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
429
  void IncOperated() {
430 431 432 433 434 435 436
    num_populated_++;
  }

  void IncDone() {
    num_done_++;
  }

437 438 439 440
  void IncVotedReopen() {
    vote_reopen_ = (vote_reopen_ + 1) % num_threads_;
  }

441 442 443 444
  bool AllInitialized() const {
    return num_initialized_ >= num_threads_;
  }

A
amayank 已提交
445
  bool AllOperated() const {
446 447 448 449 450 451 452
    return num_populated_ >= num_threads_;
  }

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

453 454 455 456
  bool AllVotedReopen() {
    return (vote_reopen_ == 0);
  }

457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484
  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
485 486 487 488
  void Delete(long key) const {
    values_[key] = SENTINEL;
  }

489 490 491 492 493 494 495 496 497 498 499 500 501
  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_;
502
  long vote_reopen_;
503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534
  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)
535
                       : nullptr),
536 537 538
        prefix_extractor_(NewFixedPrefixTransform(
                          FLAGS_test_batches_snapshots ?
                          sizeof(long) : sizeof(long)-1)),
539
        db_(nullptr),
540
        num_times_reopened_(0) {
541 542 543 544 545 546 547
    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]);
        }
548
      }
549
      DestroyDB(FLAGS_db, Options());
550 551 552 553 554 555
    }
  }

  ~StressTest() {
    delete db_;
    delete filter_policy_;
556
    delete prefix_extractor_;
557 558 559 560 561 562 563 564 565 566 567 568 569 570
  }

  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
571 572
    // initializing -> wait for others to init -> read/populate/depopulate
    // wait for others to operate -> verify -> done
573 574 575 576 577 578 579

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

580 581 582
      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
583

584 585
      shared.SetStart();
      shared.GetCondVar()->SignalAll();
A
amayank 已提交
586
      while (!shared.AllOperated()) {
587 588 589
        shared.GetCondVar()->Wait();
      }

590
      now = FLAGS_env->NowMicros();
591 592 593 594 595 596 597
      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());
      }
598

599 600 601 602 603 604 605
      shared.SetStartVerify();
      shared.GetCondVar()->SignalAll();
      while (!shared.AllDone()) {
        shared.GetCondVar()->Wait();
      }
    }

606
    for (unsigned int i = 1; i < n; i++) {
607 608 609 610
      threads[0]->stats.Merge(threads[i]->stats);
    }
    threads[0]->stats.Report("Stress Test");

611
    for (unsigned int i = 0; i < n; i++) {
612
      delete threads[i];
613
      threads[i] = nullptr;
614
    }
615
    double now = FLAGS_env->NowMicros();
616 617 618 619
    if (!FLAGS_test_batches_snapshots) {
      fprintf(stdout, "%s Verification successful\n",
              FLAGS_env->TimeToString((uint64_t) now/1000000).c_str());
    }
620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638
    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
639
    thread->shared->GetStressTest()->OperateDb(thread);
640 641 642

    {
      MutexLock l(shared->GetMutex());
A
amayank 已提交
643 644
      shared->IncOperated();
      if (shared->AllOperated()) {
645 646 647 648 649 650 651
        shared->GetCondVar()->SignalAll();
      }
      while (!shared->VerifyStarted()) {
        shared->GetCondVar()->Wait();
      }
    }

652 653 654 655
    if (!FLAGS_test_batches_snapshots) {
      thread->shared->GetStressTest()->VerifyDb(*(thread->shared),
                                                 thread->tid);
    }
656 657 658 659 660 661 662 663 664 665 666

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

  }

667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683
  // 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
684 685 686 687 688
      if (FLAGS_use_merge_put) {
        batch.Merge(keys[i], value_slices[i]);
      } else {
        batch.Put(keys[i], value_slices[i]);
      }
689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753
    }

    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] = "";
754
        thread->stats.AddGets(1, 0);
755 756 757 758 759 760
      } else {
        values[i] = *value;

        char expected_prefix = (keys[i])[0];
        char actual_prefix = (values[i])[0];
        if (actual_prefix != expected_prefix) {
761
          fprintf(stderr, "error expected prefix = %c actual = %c\n",
762 763 764
                  expected_prefix, actual_prefix);
        }
        (values[i])[0] = ' '; // blank out the differing character
765
        thread->stats.AddGets(1, 1);
766 767 768 769 770 771 772
      }
    }
    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]) {
773
        fprintf(stderr, "error : inconsistent values for key %s: %s, %s\n",
774 775 776 777 778 779 780 781 782 783
                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;
  }

784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859
  // 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;
  }

A
amayank 已提交
860
  void OperateDb(ThreadState* thread) {
861 862
    ReadOptions read_opts(FLAGS_verify_checksum, true);
    WriteOptions write_opts;
A
amayank 已提交
863
    char value[100];
864 865 866 867 868 869 870 871
    long max_key = thread->shared->GetMaxKey();
    std::string from_db;
    if (FLAGS_sync) {
      write_opts.sync = true;
    }
    write_opts.disableWAL = FLAGS_disable_wal;

    thread->stats.Start();
A
amayank 已提交
872
    for (long i = 0; i < FLAGS_ops_per_thread; i++) {
873 874
      if(i != 0 && (i % (FLAGS_ops_per_thread / (FLAGS_reopen + 1))) == 0) {
        {
875
          thread->stats.FinishedSingleOp();
876 877 878 879 880 881 882 883 884
          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();
          }
885 886
          // Commenting this out as we don't want to reset stats on each open.
          // thread->stats.Start();
887 888
        }
      }
889

890
      long rand_key = thread->rand.Next() % max_key;
891 892 893 894 895 896
      std::string keystr = Key(rand_key);
      Slice key = keystr;
      int prob_op = thread->rand.Uniform(100);

      // OPERATION read?
      if (prob_op >= 0 && prob_op < (int)FLAGS_readpercent) {
897 898 899 900 901 902 903 904 905 906 907 908 909 910 911
        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);
        }
912 913 914 915 916 917 918 919 920
      }
      prob_op -= FLAGS_readpercent;

      // OPERATION prefix scan?
      if (prob_op >= 0 && prob_op < (int)FLAGS_prefixpercent) {
        // 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);
921
        if (!FLAGS_test_batches_snapshots) {
922 923 924 925 926 927 928 929 930 931 932 933 934 935
          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;
936
        } else {
937
          MultiPrefixScan(thread, read_opts, prefix);
A
amayank 已提交
938
        }
939 940
      }
      prob_op -= FLAGS_prefixpercent;
941

942 943
      // OPERATION write?
      if (prob_op >= 0 && prob_op < (int)FLAGS_writepercent) {
944 945 946
        uint32_t value_base = thread->rand.Next();
        size_t sz = GenerateValue(value_base, value, sizeof(value));
        Slice v(value, sz);
947
        if (!FLAGS_test_batches_snapshots) {
948 949
          MutexLock l(thread->shared->GetMutexForKey(rand_key));
          if (FLAGS_verify_before_write) {
A
amayank 已提交
950
            VerifyValue(rand_key, read_opts, *(thread->shared), &from_db, true);
951 952
          }
          thread->shared->Put(rand_key, value_base);
D
Deon Nicholas 已提交
953 954 955 956 957
          if (FLAGS_use_merge_put) {
            db_->Merge(write_opts, key, v);
          } else {
            db_->Put(write_opts, key, v);
          }
958 959 960
          thread->stats.AddBytesForWrites(1, sz);
        } else {
          MultiPut(thread, write_opts, key, v, sz);
961 962 963
        }
        PrintKeyValue(rand_key, value, sz);
      }
964 965 966 967 968 969 970 971 972 973 974 975 976 977 978
      prob_op -= FLAGS_writepercent;

      // OPERATION delete?
      if (prob_op >= 0 && prob_op < (int)FLAGS_delpercent) {
        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);
        }
      }
      prob_op -= FLAGS_delpercent;

979 980 981 982 983 984 985 986 987 988 989
      thread->stats.FinishedSingleOp();
    }
    thread->stats.Stop();
  }

  void VerifyDb(const SharedState &shared, long start) const {
    ReadOptions options(FLAGS_verify_checksum, true);
    long max_key = shared.GetMaxKey();
    long step = shared.GetNumThreads();
    for (long i = start; i < max_key; i+= step) {
      std::string from_db;
A
amayank 已提交
990
      VerifyValue(i, options, shared, &from_db, true);
991 992 993 994 995 996
      if (from_db.length()) {
        PrintKeyValue(i, from_db.data(), from_db.length());
      }
    }
  }

997
  void VerificationAbort(std::string msg, long key) const {
A
amayank 已提交
998
    fprintf(stderr, "Verification failed for key %ld: %s\n",
999 1000 1001 1002
            key, msg.c_str());
    exit(1);
  }

1003 1004
  void VerifyValue(long key, const ReadOptions &opts, const SharedState &shared,
                   std::string *value_from_db, bool strict=false) const {
1005 1006
    std::string keystr = Key(key);
    Slice k = keystr;
A
amayank 已提交
1007
    char value[100];
1008 1009 1010 1011 1012 1013 1014 1015 1016
    uint32_t value_base = shared.Get(key);
    if (value_base == SharedState::SENTINEL && !strict) {
      return;
    }

    if (db_->Get(opts, k, value_from_db).ok()) {
      if (value_base == SharedState::SENTINEL) {
        VerificationAbort("Unexpected value found", key);
      }
A
amayank 已提交
1017
      size_t sz = GenerateValue(value_base, value, sizeof(value));
1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032
      if (value_from_db->length() != sz) {
        VerificationAbort("Length of value read is not equal", key);
      }
      if (memcmp(value_from_db->data(), value, sz) != 0) {
        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;
1033
    fprintf(stdout, "%u ==> (%u) ", key, (unsigned int)sz);
1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046
    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);
    }
1047
    v[value_sz] = '\0';
1048 1049 1050 1051 1052 1053 1054
    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);
1055
    fprintf(stdout, "Ops per thread      : %d\n", FLAGS_ops_per_thread);
1056 1057 1058 1059 1060
    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
1061
    fprintf(stdout, "Read percentage     : %d\n", FLAGS_readpercent);
1062 1063 1064
    fprintf(stdout, "Prefix percentage   : %d\n", FLAGS_prefixpercent);
    fprintf(stdout, "Write percentage    : %d\n", FLAGS_writepercent);
    fprintf(stdout, "Delete percentage   : %d\n", FLAGS_delpercent);
1065
    fprintf(stdout, "Write-buffer-size   : %d\n", FLAGS_write_buffer_size);
A
amayank 已提交
1066
    fprintf(stdout, "Delete percentage   : %d\n", FLAGS_delpercent);
1067
    fprintf(stdout, "Max key             : %ld\n", FLAGS_max_key);
X
Xing Jin 已提交
1068 1069
    fprintf(stdout, "Ratio #ops/#keys    : %f\n",
            (1.0 * FLAGS_ops_per_thread * FLAGS_threads)/FLAGS_max_key);
1070
    fprintf(stdout, "Num times DB reopens: %d\n", FLAGS_reopen);
1071 1072
    fprintf(stdout, "Batches/snapshots   : %d\n",
            FLAGS_test_batches_snapshots);
X
Xing Jin 已提交
1073
    fprintf(stdout, "Purge redundant %%   : %d\n",
1074
            FLAGS_purge_redundant_percent);
1075
    fprintf(stdout, "Deletes use filter  : %d\n",
1076
            FLAGS_filter_deletes);
1077
    fprintf(stdout, "Num keys per lock   : %d\n",
1078 1079
            1 << FLAGS_log2_keys_per_lock);

1080
    const char* compression = "";
1081 1082
    switch (FLAGS_compression_type) {
      case leveldb::kNoCompression:
1083
        compression = "none";
1084 1085
        break;
      case leveldb::kSnappyCompression:
1086
        compression = "snappy";
1087 1088
        break;
      case leveldb::kZlibCompression:
1089
        compression = "zlib";
1090 1091
        break;
      case leveldb::kBZip2Compression:
1092
        compression = "bzip2";
1093 1094 1095 1096 1097 1098 1099 1100
        break;
    }

    fprintf(stdout, "Compression         : %s\n", compression);
    fprintf(stdout, "------------------------------------------------\n");
  }

  void Open() {
1101
    assert(db_ == nullptr);
1102 1103 1104
    Options options;
    options.block_cache = cache_;
    options.write_buffer_size = FLAGS_write_buffer_size;
1105 1106
    options.max_write_buffer_number = FLAGS_max_write_buffer_number;
    options.max_background_compactions = FLAGS_max_background_compactions;
1107
    options.compaction_style = FLAGS_compaction_style;
1108 1109
    options.block_size = FLAGS_block_size;
    options.filter_policy = filter_policy_;
1110
    options.prefix_extractor = prefix_extractor_;
1111 1112 1113 1114 1115
    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;
1116
    options.allow_mmap_reads = FLAGS_use_mmap_reads;
1117
    leveldb_kill_odds = FLAGS_kill_random_test;
1118 1119 1120 1121 1122 1123 1124 1125
    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;
1126 1127
    options.level0_file_num_compaction_trigger =
      FLAGS_level0_file_num_compaction_trigger;
1128 1129 1130
    options.compression = FLAGS_compression_type;
    options.create_if_missing = true;
    options.disable_seek_compaction = FLAGS_disable_seek_compaction;
1131 1132
    options.delete_obsolete_files_period_micros =
      FLAGS_delete_obsolete_files_period_micros;
A
Abhishek Kona 已提交
1133
    options.max_manifest_file_size = 1024;
1134
    options.filter_deletes = FLAGS_filter_deletes;
1135 1136 1137 1138
    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;
    }
1139

D
Deon Nicholas 已提交
1140
    if (FLAGS_use_merge_put) {
1141
      options.merge_operator = MergeOperators::CreatePutOperator();
D
Deon Nicholas 已提交
1142 1143
    }

1144 1145
    fprintf(stdout, "DB path: [%s]\n", FLAGS_db);

1146 1147 1148 1149
    Status s;
    if (FLAGS_ttl == -1) {
      s = DB::Open(options, FLAGS_db, &db_);
    } else {
1150 1151
      s = UtilityDB::OpenTtlDB(options, FLAGS_db, &sdb_, FLAGS_ttl);
      db_ = sdb_;
1152
    }
1153 1154 1155 1156 1157 1158
    if (!s.ok()) {
      fprintf(stderr, "open error: %s\n", s.ToString().c_str());
      exit(1);
    }
  }

1159
  void Reopen() {
1160 1161
    // do not close the db. Just delete the lock file. This
    // simulates a crash-recovery kind of situation.
1162 1163 1164 1165 1166
    if (FLAGS_ttl != -1) {
      ((DBWithTTL*) db_)->TEST_Destroy_DBWithTtl();
    } else {
      ((DBImpl*) db_)->TEST_Destroy_DBImpl();
    }
1167
    db_ = nullptr;
1168 1169 1170 1171 1172 1173

    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_);
1174 1175 1176
    Open();
  }

1177 1178
  void PrintStatistics() {
    if (dbstats) {
1179
      fprintf(stdout, "STATISTICS:\n%s\n", dbstats->ToString().c_str());
1180 1181 1182 1183
    }
  }

 private:
1184
  shared_ptr<Cache> cache_;
1185
  const FilterPolicy* filter_policy_;
1186
  const SliceTransform* prefix_extractor_;
1187
  DB* db_;
1188
  StackableDB* sdb_;
1189
  int num_times_reopened_;
1190 1191 1192 1193 1194 1195
};

}  // namespace leveldb

int main(int argc, char** argv) {
  FLAGS_write_buffer_size = leveldb::Options().write_buffer_size;
1196
  FLAGS_max_write_buffer_number = leveldb::Options().max_write_buffer_number;
1197
  FLAGS_open_files = leveldb::Options().max_open_files;
1198 1199
  FLAGS_max_background_compactions =
    leveldb::Options().max_background_compactions;
1200 1201
  FLAGS_compaction_style =
    leveldb::Options().compaction_style;
1202 1203 1204 1205 1206 1207
  FLAGS_level0_file_num_compaction_trigger =
    leveldb::Options().level0_file_num_compaction_trigger;
  FLAGS_level0_slowdown_writes_trigger =
    leveldb::Options().level0_slowdown_writes_trigger;
  FLAGS_level0_stop_writes_trigger =
    leveldb::Options().level0_stop_writes_trigger;
1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232
  // Compression test code above refers to FLAGS_block_size
  FLAGS_block_size = leveldb::Options().block_size;
  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;
1233 1234 1235
    } else if (sscanf(argv[i], "--destroy_db_initially=%d%c", &n, &junk) == 1 &&
               (n == 0 || n == 1)) {
      FLAGS_destroy_db_initially = n;
1236 1237 1238
    } else if (sscanf(argv[i], "--verify_before_write=%d%c", &n, &junk) == 1 &&
               (n == 0 || n == 1)) {
      FLAGS_verify_before_write = n;
1239 1240 1241
    } else if (sscanf(argv[i], "--test_batches_snapshots=%d%c", &n, &junk) == 1
               && (n == 0 || n == 1)) {
      FLAGS_test_batches_snapshots = n;
1242 1243
    } else if (sscanf(argv[i], "--threads=%d%c", &n, &junk) == 1) {
      FLAGS_threads = n;
1244 1245
    } else if (sscanf(argv[i], "--ttl=%d%c", &n, &junk) == 1) {
      FLAGS_ttl = n;
1246 1247 1248 1249
    } 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;
1250 1251 1252 1253
    } else if (sscanf(argv[i], "--max_write_buffer_number=%d%c", &n, &junk) == 1) {
      FLAGS_max_write_buffer_number = n;
    } else if (sscanf(argv[i], "--max_background_compactions=%d%c", &n, &junk) == 1) {
      FLAGS_max_background_compactions = n;
1254 1255
    } else if (sscanf(argv[i], "--compaction_style=%d%c", &n, &junk) == 1) {
      FLAGS_compaction_style = (leveldb::CompactionStyle)n;
1256 1257 1258 1259
    } 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;
1260 1261
    } else if (sscanf(argv[i], "--reopen=%d%c", &n, &junk) == 1 && n >= 0) {
      FLAGS_reopen = n;
1262 1263 1264 1265 1266 1267 1268 1269 1270
    } 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;
1271 1272 1273
    } else if (sscanf(argv[i], "--mmap_read=%d%c", &n, &junk) == 1 &&
               (n == 0 || n == 1)) {
      FLAGS_use_mmap_reads = n;
1274 1275 1276
    } else if (sscanf(argv[i], "--statistics=%d%c", &n, &junk) == 1 &&
               (n == 0 || n == 1)) {
      if (n == 1) {
H
Haobo Xu 已提交
1277
        dbstats = leveldb::CreateDBStatistics();
1278 1279 1280 1281
      }
    } else if (sscanf(argv[i], "--sync=%d%c", &n, &junk) == 1 &&
               (n == 0 || n == 1)) {
      FLAGS_sync = n;
A
amayank 已提交
1282 1283 1284
    } else if (sscanf(argv[i], "--readpercent=%d%c", &n, &junk) == 1 &&
               (n >= 0 && n <= 100)) {
      FLAGS_readpercent = n;
1285 1286 1287 1288 1289 1290
    } 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;
A
amayank 已提交
1291 1292 1293
    } else if (sscanf(argv[i], "--delpercent=%d%c", &n, &junk) == 1 &&
               (n >= 0 && n <= 100)) {
      FLAGS_delpercent = n;
1294 1295 1296 1297 1298 1299
    } 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;
1300 1301 1302
    } else if (sscanf(argv[i], "--kill_random_test=%d%c", &n, &junk) == 1 &&
               (n >= 0)) {
      FLAGS_kill_random_test = n;
1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314
    } 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) {
      FLAGS_env  = new leveldb::HdfsEnv(hdfsname);
    } 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 (
1315 1316
        sscanf(argv[i], "--max_bytes_for_level_base=%ld%c", &l, &junk) == 1) {
      FLAGS_max_bytes_for_level_base = l;
1317 1318 1319 1320 1321 1322 1323 1324 1325
    } 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;
1326 1327 1328
    } else if (sscanf(argv[i],"--level0_file_num_compaction_trigger=%d%c",
        &n, &junk) == 1) {
      FLAGS_level0_file_num_compaction_trigger = n;
1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344
    } else if (strncmp(argv[i], "--compression_type=", 19) == 0) {
      const char* ctype = argv[i] + 19;
      if (!strcasecmp(ctype, "none"))
        FLAGS_compression_type = leveldb::kNoCompression;
      else if (!strcasecmp(ctype, "snappy"))
        FLAGS_compression_type = leveldb::kSnappyCompression;
      else if (!strcasecmp(ctype, "zlib"))
        FLAGS_compression_type = leveldb::kZlibCompression;
      else if (!strcasecmp(ctype, "bzip2"))
        FLAGS_compression_type = leveldb::kBZip2Compression;
      else {
        fprintf(stdout, "Cannot parse %s\n", argv[i]);
      }
    } else if (sscanf(argv[i], "--disable_seek_compaction=%d%c", &n, &junk) == 1
        && (n == 0 || n == 1)) {
      FLAGS_disable_seek_compaction = n;
1345 1346 1347
    } else if (sscanf(argv[i], "--delete_obsolete_files_period_micros=%ld%c",
                      &l, &junk) == 1) {
      FLAGS_delete_obsolete_files_period_micros = n;
1348 1349 1350
    } else if (sscanf(argv[i], "--purge_redundant_percent=%d%c", &n, &junk) == 1
        && (n >= 0 && n <= 100)) {
      FLAGS_purge_redundant_percent = n;
1351
    } else if (sscanf(argv[i], "--filter_deletes=%d%c", &n, &junk)
1352
        == 1 && (n == 0 || n == 1)) {
1353
      FLAGS_filter_deletes = n;
D
Deon Nicholas 已提交
1354 1355 1356
    } else if (sscanf(argv[i], "--use_merge=%d%c", &n, &junk)
        == 1 && (n == 0 || n == 1)) {
      FLAGS_use_merge_put = n;
1357 1358 1359 1360 1361 1362
    } else {
      fprintf(stderr, "Invalid flag '%s'\n", argv[i]);
      exit(1);
    }
  }

1363 1364 1365 1366
  // The number of background threads should be at least as much the
  // max number of concurrent compactions.
  FLAGS_env->SetBackgroundThreads(FLAGS_max_background_compactions);

1367 1368 1369
  if ((FLAGS_readpercent + FLAGS_prefixpercent +
       FLAGS_writepercent + FLAGS_delpercent) != 100) {
      fprintf(stderr, "Error: Read+Prefix+Write+Delete percents != 100!\n");
A
amayank 已提交
1370 1371
      exit(1);
  }
1372 1373 1374 1375
  if (FLAGS_disable_wal == 1 && FLAGS_reopen > 0) {
      fprintf(stderr, "Error: Db cannot reopen safely with disable_wal set!\n");
      exit(1);
  }
1376 1377 1378 1379 1380 1381 1382
  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);
  }

1383
  // Choose a location for the test database if none given with --db=<path>
1384
  if (FLAGS_db == nullptr) {
1385 1386 1387 1388 1389 1390 1391 1392 1393
      leveldb::Env::Default()->GetTestDirectory(&default_db_path);
      default_db_path += "/dbstress";
      FLAGS_db = default_db_path.c_str();
  }

  leveldb::StressTest stress;
  stress.Run();
  return 0;
}