db_bench.cc 61.1 KB
Newer Older
J
jorlow@chromium.org 已提交
1 2 3 4
// 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.

5
#include <cstddef>
J
jorlow@chromium.org 已提交
6 7 8 9 10
#include <sys/types.h>
#include <stdio.h>
#include <stdlib.h>
#include "db/db_impl.h"
#include "db/version_set.h"
11
#include "db/db_statistics.h"
12 13 14 15
#include "leveldb/cache.h"
#include "leveldb/db.h"
#include "leveldb/env.h"
#include "leveldb/write_batch.h"
16
#include "leveldb/statistics.h"
J
jorlow@chromium.org 已提交
17 18
#include "port/port.h"
#include "util/crc32c.h"
J
jorlow@chromium.org 已提交
19
#include "util/histogram.h"
20
#include "util/mutexlock.h"
J
jorlow@chromium.org 已提交
21 22
#include "util/random.h"
#include "util/testutil.h"
23
#include "hdfs/env_hdfs.h"
J
jorlow@chromium.org 已提交
24 25 26

// Comma-separated list of operations to run in the specified order
//   Actual benchmarks:
27 28 29 30 31
//      fillseq       -- write N values in sequential key order in async mode
//      fillrandom    -- write N values in random key order in async mode
//      overwrite     -- overwrite N values in random key order in async mode
//      fillsync      -- write N/100 values in random key order in sync mode
//      fill100K      -- write N/1000 100K values in random order in async mode
S
Sanjay Ghemawat 已提交
32 33
//      deleteseq     -- delete N keys in sequential order
//      deleterandom  -- delete N keys in random order
34 35 36
//      readseq       -- read N times sequentially
//      readreverse   -- read N times in reverse order
//      readrandom    -- read N times in random order
S
Sanjay Ghemawat 已提交
37
//      readmissing   -- read N missing keys in random order
38
//      readhot       -- read N times in random order from 1% section of DB
39 40 41
//      readwhilewriting -- 1 writer, N threads doing random reads
//      readrandomwriterandom - N threads doing random-read, random-write
//      updaterandom  -- N threads doing read-modify-write for random keys
S
Sanjay Ghemawat 已提交
42
//      seekrandom    -- N random seeks
J
jorlow@chromium.org 已提交
43
//      crc32c        -- repeated crc32c of 4K of data
44
//      acquireload   -- load N*1000 times
J
jorlow@chromium.org 已提交
45 46
//   Meta operations:
//      compact     -- Compact the entire DB
47
//      stats       -- Print DB stats
48
//      levelstats  -- Print the number of files and bytes per level
S
Sanjay Ghemawat 已提交
49
//      sstables    -- Print sstable info
J
jorlow@chromium.org 已提交
50 51
//      heapprofile -- Dump a heap profile (if supported by this port)
static const char* FLAGS_benchmarks =
52
    "fillseq,"
J
jorlow@chromium.org 已提交
53
    "fillsync,"
54 55
    "fillrandom,"
    "overwrite,"
J
jorlow@chromium.org 已提交
56 57
    "readrandom,"
    "readrandom,"  // Extra run to allow previous compactions to quiesce
J
jorlow@chromium.org 已提交
58
    "readseq,"
J
jorlow@chromium.org 已提交
59
    "readreverse,"
J
jorlow@chromium.org 已提交
60
    "compact,"
J
jorlow@chromium.org 已提交
61
    "readrandom,"
J
jorlow@chromium.org 已提交
62
    "readseq,"
J
jorlow@chromium.org 已提交
63
    "readreverse,"
64
    "readwhilewriting,"
65
    "readrandomwriterandom," // mix reads and writes based on FLAGS_readwritepercent
M
Mark Callaghan 已提交
66
    "updaterandom," // read-modify-write for random keys
67
    "randomwithverify," // random reads and writes with some verification
J
jorlow@chromium.org 已提交
68 69
    "fill100K,"
    "crc32c,"
70 71
    "snappycomp,"
    "snappyuncomp,"
72
    "acquireload,"
J
jorlow@chromium.org 已提交
73
    ;
74 75
// the maximum size of key in bytes
static const int MAX_KEY_SIZE = 128;
J
jorlow@chromium.org 已提交
76
// Number of key/values to place in database
77
static long FLAGS_num = 1000000;
J
jorlow@chromium.org 已提交
78

79 80 81 82 83
// Number of distinct keys to use. Used in RandomWithVerify to read/write
// on fewer keys so that gets are more likely to find the key and puts
// are more likely to update the same key
static long FLAGS_numdistinct = 1000;

84
// Number of read operations to do.  If negative, do FLAGS_num reads.
85
static long FLAGS_reads = -1;
86

87 88 89 90 91 92
// When ==1 reads use ::Get, when >1 reads use an iterator
static long FLAGS_read_range = 1;

// Seed base for random number generators. When 0 it is deterministic.
static long FLAGS_seed = 0;

93 94 95
// Number of concurrent threads to run.
static int FLAGS_threads = 1;

M
Mark Callaghan 已提交
96 97 98 99
// Time in seconds for the random-ops tests to run. When 0 then
// FLAGS_num & FLAGS_reads determine the test duration
static int FLAGS_duration = 0;

J
jorlow@chromium.org 已提交
100 101 102
// Size of each value
static int FLAGS_value_size = 100;

103 104 105
//size of each key
static int FLAGS_key_size = 16;

J
jorlow@chromium.org 已提交
106 107
// Arrange to generate values that shrink to this fraction of
// their original size after compression
108
static double FLAGS_compression_ratio = 0.5;
J
jorlow@chromium.org 已提交
109 110 111 112 113

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

// Number of bytes to buffer in memtable before compacting
114 115 116
// (initialized to default value by "main")
static int FLAGS_write_buffer_size = 0;

A
Abhishek Kona 已提交
117
// The number of in-memory memtables.
118 119 120 121 122 123 124 125 126
// 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;

127 128
// Number of bytes to use as a cache of uncompressed data.
// Negative means use default settings.
D
Dhruba Borthakur 已提交
129
static long FLAGS_cache_size = -1;
J
jorlow@chromium.org 已提交
130

131 132 133
// Number of bytes in a block.
static int FLAGS_block_size = 0;

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

S
Sanjay Ghemawat 已提交
137 138 139 140
// Bloom filter bits per key.
// Negative means use default settings.
static int FLAGS_bloom_bits = -1;

141 142 143 144 145
// If true, do not destroy the existing database.  If you set this
// flag and also specify a benchmark that wants a fresh database, that
// benchmark will fail.
static bool FLAGS_use_existing_db = false;

146
// Use the db with the following name.
147
static const char* FLAGS_db = nullptr;
148

149 150 151 152 153
// Number of shards for the block cache is 2 ** FLAGS_cache_numshardbits.
// Negative means use default settings. This is applied only
// if FLAGS_cache_size is non-negative.
static int FLAGS_cache_numshardbits = -1;

154 155 156
// Verify checksum for every block read from storage
static bool FLAGS_verify_checksum = false;

157 158
// Database statistics
static bool FLAGS_statistics = false;
159
static class leveldb::DBStatistics* dbstats = nullptr;
160

161 162 163
// Number of write operations to do.  If negative, do FLAGS_num reads.
static long FLAGS_writes = -1;

H
heyongqiang 已提交
164 165
// These default values might change if the hardcoded

166 167 168
// Sync all writes to disk
static bool FLAGS_sync = false;

H
heyongqiang 已提交
169 170 171
// If true, do not wait until data is synced to disk.
static bool FLAGS_disable_data_sync = false;

172 173 174
// If true, issue fsync instead of fdatasync
static bool FLAGS_use_fsync = false;

H
heyongqiang 已提交
175 176 177
// If true, do not write WAL for write.
static bool FLAGS_disable_wal = false;

M
Mark Callaghan 已提交
178 179 180 181 182 183 184
// If true, create a snapshot per query when randomread benchmark is used
static bool FLAGS_use_snapshot = false;

// If true, call GetApproximateSizes per query when FLAGS_read_range is > 1
// and randomread benchmark is used
static bool FLAGS_get_approx = false;

185
// The total number of levels
186
static int FLAGS_num_levels = 7;
187

H
heyongqiang 已提交
188 189 190 191 192 193
// Target level-0 file size for compaction
static int FLAGS_target_file_size_base = 2 * 1048576;

// A multiplier to compute targe level-N file size
static int FLAGS_target_file_size_multiplier = 1;

194
// Max bytes for level-1
195
static uint64_t FLAGS_max_bytes_for_level_base = 10 * 1048576;
H
heyongqiang 已提交
196 197 198 199

// A multiplier to compute max bytes for level-N
static int FLAGS_max_bytes_for_level_multiplier = 10;

H
heyongqiang 已提交
200 201 202
// Number of files in level-0 that will trigger put stop.
static int FLAGS_level0_stop_writes_trigger = 12;

203 204
// Number of files in level-0 that will slow down writes.
static int FLAGS_level0_slowdown_writes_trigger = 8;
H
heyongqiang 已提交
205

206 207 208
// Number of files in level-0 when compactions start
static int FLAGS_level0_file_num_compaction_trigger = 4;

209 210 211 212 213
// Ratio of reads to writes (expressed as a percentage)
// for the ReadRandomWriteRandom workload. The default
// setting is 9 gets for every 1 put.
static int FLAGS_readwritepercent = 90;

214 215 216 217
// This percent of deletes are done (used in RandomWithVerify only)
// Must be smaller than total writepercent (i.e 100 - FLAGS_readwritepercent)
static int FLAGS_deletepercent = 2;

218 219 220
// Option to disable compation triggered by read.
static int FLAGS_disable_seek_compaction = false;

221 222 223 224 225
// 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;

226 227 228 229
// Algorithm to use to compress the database
static enum leveldb::CompressionType FLAGS_compression_type =
    leveldb::kSnappyCompression;

230 231
// Allows compression for levels 0 and 1 to be disabled when
// other levels are compressed
232
static int FLAGS_min_level_to_compress = -1;
233

234 235
static int FLAGS_table_cache_numshardbits = 4;

236 237 238
// posix or hdfs environment
static leveldb::Env* FLAGS_env = leveldb::Env::Default();

239 240 241 242
// Stats are reported every N operations when this is greater
// than zero. When 0 the interval grows over time.
static int FLAGS_stats_interval = 0;

243 244 245 246 247 248 249 250 251
// Reports additional stats per interval when this is greater
// than 0.
static int FLAGS_stats_per_interval = 0;

// When not equal to 0 this make threads sleep at each stats
// reporting interval until the compaction score for all levels is
// less than or equal to this value.
static double FLAGS_rate_limit = 0;

252 253 254
// When FLAGS_rate_limit is set then this is the max time a put will be stalled.
static int FLAGS_rate_limit_delay_milliseconds = 1000;

255 256
// Control maximum bytes of overlaps in grandparent (i.e., level+2) before we
// stop building a single file in a level->level+1 compaction.
257
static int FLAGS_max_grandparent_overlap_factor = 10;
258

H
heyongqiang 已提交
259 260 261
// Run read only benchmarks.
static bool FLAGS_read_only = false;

262 263 264
// Do not auto trigger compactions
static bool FLAGS_disable_auto_compactions = false;

265 266 267 268
// Cap the size of data in levelK for a compaction run
// that compacts Levelk with LevelK+1
static int FLAGS_source_compaction_factor = 1;

269 270 271
// Set the TTL for the WAL Files.
static uint64_t FLAGS_WAL_ttl_seconds = 0;

272
extern bool useOsBuffer;
273
extern bool useFsReadAhead;
274
extern bool useMmapRead;
275
extern bool useMmapWrite;
276

J
jorlow@chromium.org 已提交
277 278
namespace leveldb {

279
// Helper for quickly generating random data.
J
jorlow@chromium.org 已提交
280 281 282
class RandomGenerator {
 private:
  std::string data_;
283
  unsigned int pos_;
J
jorlow@chromium.org 已提交
284 285 286 287 288 289 290 291

 public:
  RandomGenerator() {
    // We use a limited amount of data over and over again and ensure
    // that it is larger than the compression window (32KB), and also
    // large enough to serve all typical value sizes we want to write.
    Random rnd(301);
    std::string piece;
292
    while (data_.size() < (unsigned)std::max(1048576, FLAGS_value_size)) {
J
jorlow@chromium.org 已提交
293 294 295 296 297 298 299 300
      // Add a short fragment that is as compressible as specified
      // by FLAGS_compression_ratio.
      test::CompressibleString(&rnd, FLAGS_compression_ratio, 100, &piece);
      data_.append(piece);
    }
    pos_ = 0;
  }

301
  Slice Generate(unsigned int len) {
J
jorlow@chromium.org 已提交
302 303 304 305 306 307 308
    if (pos_ + len > data_.size()) {
      pos_ = 0;
      assert(len < data_.size());
    }
    pos_ += len;
    return Slice(data_.data() + pos_ - len, len);
  }
309
};
310
static Slice TrimSpace(Slice s) {
311
  unsigned int start = 0;
312 313 314
  while (start < s.size() && isspace(s[start])) {
    start++;
  }
315
  unsigned int limit = s.size();
316 317 318 319 320 321
  while (limit > start && isspace(s[limit-1])) {
    limit--;
  }
  return Slice(s.data() + start, limit - start);
}

322 323 324 325 326 327 328 329 330 331
static void AppendWithSpace(std::string* str, Slice msg) {
  if (msg.empty()) return;
  if (!str->empty()) {
    str->push_back(' ');
  }
  str->append(msg.data(), msg.size());
}

class Stats {
 private:
332
  int id_;
333 334 335
  double start_;
  double finish_;
  double seconds_;
336
  long done_;
337
  long last_report_done_;
338 339 340
  int next_report_;
  int64_t bytes_;
  double last_op_finish_;
341
  double last_report_finish_;
342
  HistogramImpl hist_;
343
  std::string message_;
344
  bool exclude_from_merge_;
345 346

 public:
347
  Stats() { Start(-1); }
348

349 350 351
  void Start(int id) {
    id_ = id;
    next_report_ = FLAGS_stats_interval ? FLAGS_stats_interval : 100;
352 353 354
    last_op_finish_ = start_;
    hist_.Clear();
    done_ = 0;
355
    last_report_done_ = 0;
356 357
    bytes_ = 0;
    seconds_ = 0;
358
    start_ = FLAGS_env->NowMicros();
359
    finish_ = start_;
360
    last_report_finish_ = start_;
361
    message_.clear();
362 363
    // When set, stats from this thread won't be merged with others.
    exclude_from_merge_ = false;
364 365 366
  }

  void Merge(const Stats& other) {
367 368 369
    if (other.exclude_from_merge_)
      return;

370 371 372 373 374 375 376 377 378 379 380 381
    hist_.Merge(other.hist_);
    done_ += other.done_;
    bytes_ += other.bytes_;
    seconds_ += other.seconds_;
    if (other.start_ < start_) start_ = other.start_;
    if (other.finish_ > finish_) finish_ = other.finish_;

    // Just keep the messages from one thread
    if (message_.empty()) message_ = other.message_;
  }

  void Stop() {
382
    finish_ = FLAGS_env->NowMicros();
383 384 385 386 387 388 389
    seconds_ = (finish_ - start_) * 1e-6;
  }

  void AddMessage(Slice msg) {
    AppendWithSpace(&message_, msg);
  }

390
  void SetId(int id) { id_ = id; }
391
  void SetExcludeFromMerge() { exclude_from_merge_ = true; }
392

M
Mark Callaghan 已提交
393
  void FinishedSingleOp(DB* db) {
394
    if (FLAGS_histogram) {
395
      double now = FLAGS_env->NowMicros();
396 397
      double micros = now - last_op_finish_;
      hist_.Add(micros);
398
      if (micros > 20000 && !FLAGS_stats_interval) {
399 400 401 402 403 404 405 406
        fprintf(stderr, "long op: %.1f micros%30s\r", micros, "");
        fflush(stderr);
      }
      last_op_finish_ = now;
    }

    done_++;
    if (done_ >= next_report_) {
407 408 409 410 411 412 413 414 415 416 417 418 419
      if (!FLAGS_stats_interval) {
        if      (next_report_ < 1000)   next_report_ += 100;
        else if (next_report_ < 5000)   next_report_ += 500;
        else if (next_report_ < 10000)  next_report_ += 1000;
        else if (next_report_ < 50000)  next_report_ += 5000;
        else if (next_report_ < 100000) next_report_ += 10000;
        else if (next_report_ < 500000) next_report_ += 50000;
        else                            next_report_ += 100000;
        fprintf(stderr, "... finished %ld ops%30s\r", done_, "");
        fflush(stderr);
      } else {
        double now = FLAGS_env->NowMicros();
        fprintf(stderr,
420
                "%s ... thread %d: (%ld,%ld) ops and (%.1f,%.1f) ops/second in (%.6f,%.6f) seconds\n",
421
                FLAGS_env->TimeToString((uint64_t) now/1000000).c_str(),
422
                id_,
M
Mark Callaghan 已提交
423
                done_ - last_report_done_, done_,
424
                (done_ - last_report_done_) /
M
Mark Callaghan 已提交
425 426 427 428
                ((now - last_report_finish_) / 1000000.0),
                done_ / ((now - start_) / 1000000.0),
                (now - last_report_finish_) / 1000000.0,
                (now - start_) / 1000000.0);
M
Mark Callaghan 已提交
429

430 431 432
        if (FLAGS_stats_per_interval) {
          std::string stats;
          if (db && db->GetProperty("leveldb.stats", &stats))
433
            fprintf(stderr, "%s\n", stats.c_str());
434
        }
M
Mark Callaghan 已提交
435

436 437 438 439 440
        fflush(stderr);
        next_report_ += FLAGS_stats_interval;
        last_report_finish_ = now;
        last_report_done_ = done_;
      }
441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463
    }
  }

  void AddBytes(int64_t n) {
    bytes_ += n;
  }

  void Report(const Slice& name) {
    // Pretend at least one op was done in case we are running a benchmark
    // that does not call FinishedSingleOp().
    if (done_ < 1) done_ = 1;

    std::string extra;
    if (bytes_ > 0) {
      // Rate is computed on actual elapsed time, not the sum of per-thread
      // elapsed times.
      double elapsed = (finish_ - start_) * 1e-6;
      char rate[100];
      snprintf(rate, sizeof(rate), "%6.1f MB/s",
               (bytes_ / 1048576.0) / elapsed);
      extra = rate;
    }
    AppendWithSpace(&extra, message_);
464 465
    double elapsed = (finish_ - start_) * 1e-6;
    double throughput = (double)done_/elapsed;
466

D
Dhruba Borthakur 已提交
467
    fprintf(stdout, "%-12s : %11.3f micros/op %ld ops/sec;%s%s\n",
468 469
            name.ToString().c_str(),
            seconds_ * 1e6 / done_,
D
Dhruba Borthakur 已提交
470
            (long)throughput,
471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491
            (extra.empty() ? "" : " "),
            extra.c_str());
    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.
struct SharedState {
  port::Mutex mu;
  port::CondVar cv;
  int total;

  // Each thread goes through the following states:
  //    (1) initializing
  //    (2) waiting for others to be initialized
  //    (3) running
  //    (4) done

492 493
  long num_initialized;
  long num_done;
494 495 496 497 498 499 500 501 502 503
  bool start;

  SharedState() : cv(&mu) { }
};

// Per-thread state for concurrent executions of the same benchmark.
struct ThreadState {
  int tid;             // 0..n-1 when running in n threads
  Random rand;         // Has different seeds for different threads
  Stats stats;
504
  SharedState* shared;
505

A
Abhishek Kona 已提交
506
  /* implicit */ ThreadState(int index)
507
      : tid(index),
508
        rand((FLAGS_seed ? FLAGS_seed : 1000) + index) {
509 510 511
  }
};

M
Mark Callaghan 已提交
512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542
class Duration {
 public:
  Duration(int max_seconds, long max_ops) {
    max_seconds_ = max_seconds;
    max_ops_= max_ops;
    ops_ = 0;
    start_at_ = FLAGS_env->NowMicros();
  }

  bool Done(int increment) {
    ops_ += increment;

    if (max_seconds_) {
      if (!(ops_ % 1000)) {
        double now = FLAGS_env->NowMicros();
        return ((now - start_at_) / 1000000.0) >= max_seconds_;
      } else {
        return false;
      }
    } else {
      return ops_ > max_ops_;
    }
  }

 private:
  int max_seconds_;
  long max_ops_;
  long ops_;
  double start_at_;
};

J
jorlow@chromium.org 已提交
543 544
class Benchmark {
 private:
545
  shared_ptr<Cache> cache_;
S
Sanjay Ghemawat 已提交
546
  const FilterPolicy* filter_policy_;
J
jorlow@chromium.org 已提交
547
  DB* db_;
548
  long num_;
549
  int value_size_;
550
  int key_size_;
551 552
  int entries_per_batch_;
  WriteOptions write_options_;
553
  long reads_;
554
  long writes_;
555
  long readwrites_;
J
jorlow@chromium.org 已提交
556
  int heap_counter_;
557
  char keyFormat_[100]; // this string will contain the format of key. e.g "%016d"
558 559
  void PrintHeader() {
    PrintEnvironment();
560
    fprintf(stdout, "Keys:       %d bytes each\n", FLAGS_key_size);
561 562 563
    fprintf(stdout, "Values:     %d bytes each (%d bytes after compression)\n",
            FLAGS_value_size,
            static_cast<int>(FLAGS_value_size * FLAGS_compression_ratio + 0.5));
564
    fprintf(stdout, "Entries:    %ld\n", num_);
565
    fprintf(stdout, "RawSize:    %.1f MB (estimated)\n",
566
            ((static_cast<int64_t>(FLAGS_key_size + FLAGS_value_size) * num_)
567
             / 1048576.0));
568
    fprintf(stdout, "FileSize:   %.1f MB (estimated)\n",
569
            (((FLAGS_key_size + FLAGS_value_size * FLAGS_compression_ratio) * num_)
570
             / 1048576.0));
571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586

    switch (FLAGS_compression_type) {
      case leveldb::kNoCompression:
        fprintf(stdout, "Compression: none\n");
        break;
      case leveldb::kSnappyCompression:
        fprintf(stdout, "Compression: snappy\n");
        break;
      case leveldb::kZlibCompression:
        fprintf(stdout, "Compression: zlib\n");
        break;
      case leveldb::kBZip2Compression:
        fprintf(stdout, "Compression: bzip2\n");
        break;
    }

587 588 589 590 591 592 593 594 595 596 597 598 599 600
    PrintWarnings();
    fprintf(stdout, "------------------------------------------------\n");
  }

  void PrintWarnings() {
#if defined(__GNUC__) && !defined(__OPTIMIZE__)
    fprintf(stdout,
            "WARNING: Optimization is disabled: benchmarks unnecessarily slow\n"
            );
#endif
#ifndef NDEBUG
    fprintf(stdout,
            "WARNING: Assertions are enabled; benchmarks unnecessarily slow\n");
#endif
601

602 603 604 605 606
    if (FLAGS_compression_type != leveldb::kNoCompression) {
      // The test string should not be too small.
      const int len = FLAGS_block_size;
      char* text = (char*) malloc(len+1);
      bool result = true;
607
      const char* name = nullptr;
608 609 610 611 612 613 614
      std::string compressed;

      memset(text, (int) 'y', len);
      text[len] = '\0';

      switch (FLAGS_compression_type) {
        case kSnappyCompression:
615 616
          result = port::Snappy_Compress(Options().compression_opts, text,
                                         strlen(text), &compressed);
617 618 619
          name = "Snappy";
          break;
        case kZlibCompression:
620 621
          result = port::Zlib_Compress(Options().compression_opts, text,
                                       strlen(text), &compressed);
622 623 624
          name = "Zlib";
          break;
        case kBZip2Compression:
625 626
          result = port::BZip2_Compress(Options().compression_opts, text,
                                        strlen(text), &compressed);
627 628
          name = "BZip2";
          break;
629 630 631
        case kNoCompression:
          assert(false); // cannot happen
          break;
632 633 634 635 636 637 638 639 640
      }

      if (!result) {
        fprintf(stdout, "WARNING: %s compression is not enabled\n", name);
      } else if (name && compressed.size() >= strlen(text)) {
        fprintf(stdout, "WARNING: %s compression is not effective\n", name);
      }

      free(text);
641
    }
642 643 644 645 646 647 648
  }

  void PrintEnvironment() {
    fprintf(stderr, "LevelDB:    version %d.%d\n",
            kMajorVersion, kMinorVersion);

#if defined(__linux)
649
    time_t now = time(nullptr);
650 651 652
    fprintf(stderr, "Date:       %s", ctime(&now));  // ctime() adds newline

    FILE* cpuinfo = fopen("/proc/cpuinfo", "r");
653
    if (cpuinfo != nullptr) {
654 655 656 657
      char line[1000];
      int num_cpus = 0;
      std::string cpu_type;
      std::string cache_size;
658
      while (fgets(line, sizeof(line), cpuinfo) != nullptr) {
659
        const char* sep = strchr(line, ':');
660
        if (sep == nullptr) {
661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678
          continue;
        }
        Slice key = TrimSpace(Slice(line, sep - 1 - line));
        Slice val = TrimSpace(Slice(sep + 1));
        if (key == "model name") {
          ++num_cpus;
          cpu_type = val.ToString();
        } else if (key == "cache size") {
          cache_size = val.ToString();
        }
      }
      fclose(cpuinfo);
      fprintf(stderr, "CPU:        %d * %s\n", num_cpus, cpu_type.c_str());
      fprintf(stderr, "CPUCache:   %s\n", cache_size.c_str());
    }
#endif
  }

679 680 681 682 683 684 685 686
  void PrintHistogram(Histograms histogram_type, std::string name) {
    HistogramData histogramData;
    dbstats->histogramData(histogram_type, &histogramData);
    fprintf(stdout, "%s statistics : \n", name.c_str());
    fprintf(stdout, "Median : %f\n",histogramData.median);
    fprintf(stdout, "99ile : %f\n", histogramData.percentile99);
  }

687 688
  void PrintStatistics() {
    if (FLAGS_statistics) {
A
Abhishek Kona 已提交
689
      fprintf(stdout, "File opened:%ld closed:%ld errors:%ld\n"
690 691 692
          "Block Cache Hit Count:%ld Block Cache Miss Count:%ld\n"
          "Bloom Filter Useful: %ld \n"
          "Compaction key_drop_newer_entry: %ld key_drop_obsolete: %ld "
693
          "Compaction key_drop_user: %ld\n",
694 695 696
          dbstats->getTickerCount(NO_FILE_OPENS),
          dbstats->getTickerCount(NO_FILE_CLOSES),
          dbstats->getTickerCount(NO_FILE_ERRORS),
A
Abhishek Kona 已提交
697
          dbstats->getTickerCount(BLOCK_CACHE_HIT),
698 699 700 701 702
          dbstats->getTickerCount(BLOCK_CACHE_MISS),
          dbstats->getTickerCount(BLOOM_FILTER_USEFUL),
          dbstats->getTickerCount(COMPACTION_KEY_DROP_NEWER_ENTRY),
          dbstats->getTickerCount(COMPACTION_KEY_DROP_OBSOLETE),
          dbstats->getTickerCount(COMPACTION_KEY_DROP_USER));
703 704
      PrintHistogram(DB_GET, "DB_GET");
      PrintHistogram(DB_WRITE, "DB_WRITE");
705 706 707
    }
  }

J
jorlow@chromium.org 已提交
708
 public:
709
  Benchmark()
710 711 712
  : cache_(FLAGS_cache_size >= 0 ?
           (FLAGS_cache_numshardbits >= 1 ?
            NewLRUCache(FLAGS_cache_size, FLAGS_cache_numshardbits) :
713
            NewLRUCache(FLAGS_cache_size)) : nullptr),
S
Sanjay Ghemawat 已提交
714 715
    filter_policy_(FLAGS_bloom_bits >= 0
                   ? NewBloomFilterPolicy(FLAGS_bloom_bits)
716 717
                   : nullptr),
    db_(nullptr),
718
    num_(FLAGS_num),
719
    value_size_(FLAGS_value_size),
720
    key_size_(FLAGS_key_size),
721
    entries_per_batch_(1),
722
    reads_(FLAGS_reads < 0 ? FLAGS_num : FLAGS_reads),
723
    writes_(FLAGS_writes < 0 ? FLAGS_num : FLAGS_writes),
724
    readwrites_((FLAGS_writes < 0  && FLAGS_reads < 0)? FLAGS_num :
725
                ((FLAGS_writes > FLAGS_reads) ? FLAGS_writes : FLAGS_reads)
726
               ),
727
    heap_counter_(0) {
J
jorlow@chromium.org 已提交
728
    std::vector<std::string> files;
729
    FLAGS_env->GetChildren(FLAGS_db, &files);
730
    for (unsigned int i = 0; i < files.size(); i++) {
J
jorlow@chromium.org 已提交
731
      if (Slice(files[i]).starts_with("heap-")) {
732
        FLAGS_env->DeleteFile(std::string(FLAGS_db) + "/" + files[i]);
J
jorlow@chromium.org 已提交
733 734
      }
    }
735
    if (!FLAGS_use_existing_db) {
736
      DestroyDB(FLAGS_db, Options());
737
    }
J
jorlow@chromium.org 已提交
738 739 740 741
  }

  ~Benchmark() {
    delete db_;
S
Sanjay Ghemawat 已提交
742
    delete filter_policy_;
J
jorlow@chromium.org 已提交
743
  }
744 745 746 747 748 749 750
//this function will construct string format for key. e.g "%016d"
void ConstructStrFormatForKey(char* str, int keySize)
{
  str[0] = '%';
  str[1] = '0';
  sprintf(str+2, "%dd", keySize);
}
J
jorlow@chromium.org 已提交
751

752 753 754 755 756 757
unique_ptr<char []> GenerateKeyFromInt(int v)
{
  unique_ptr<char []> keyInStr(new char[MAX_KEY_SIZE]);
  snprintf(keyInStr.get(), MAX_KEY_SIZE, keyFormat_, v);
  return keyInStr;
}
J
jorlow@chromium.org 已提交
758
  void Run() {
759 760
    PrintHeader();
    Open();
J
jorlow@chromium.org 已提交
761 762

    const char* benchmarks = FLAGS_benchmarks;
763
    while (benchmarks != nullptr) {
J
jorlow@chromium.org 已提交
764 765
      const char* sep = strchr(benchmarks, ',');
      Slice name;
766
      if (sep == nullptr) {
J
jorlow@chromium.org 已提交
767
        name = benchmarks;
768
        benchmarks = nullptr;
J
jorlow@chromium.org 已提交
769 770 771 772 773
      } else {
        name = Slice(benchmarks, sep - benchmarks);
        benchmarks = sep + 1;
      }

774 775
      // Reset parameters that may be overriddden bwlow
      num_ = FLAGS_num;
776
      reads_ = (FLAGS_reads < 0 ? FLAGS_num : FLAGS_reads);
777
      writes_ = (FLAGS_writes < 0 ? FLAGS_num : FLAGS_writes);
778
      value_size_ = FLAGS_value_size;
779 780
      key_size_ = FLAGS_key_size;
      ConstructStrFormatForKey(keyFormat_, key_size_);
781 782
      entries_per_batch_ = 1;
      write_options_ = WriteOptions();
783 784 785
      if (FLAGS_sync) {
        write_options_.sync = true;
      }
H
heyongqiang 已提交
786 787
      write_options_.disableWAL = FLAGS_disable_wal;

788
      void (Benchmark::*method)(ThreadState*) = nullptr;
789
      bool fresh_db = false;
790
      int num_threads = FLAGS_threads;
791 792

      if (name == Slice("fillseq")) {
793 794
        fresh_db = true;
        method = &Benchmark::WriteSeq;
795
      } else if (name == Slice("fillbatch")) {
796 797 798
        fresh_db = true;
        entries_per_batch_ = 1000;
        method = &Benchmark::WriteSeq;
799
      } else if (name == Slice("fillrandom")) {
800 801
        fresh_db = true;
        method = &Benchmark::WriteRandom;
802
      } else if (name == Slice("overwrite")) {
803 804
        fresh_db = false;
        method = &Benchmark::WriteRandom;
805
      } else if (name == Slice("fillsync")) {
806 807 808 809
        fresh_db = true;
        num_ /= 1000;
        write_options_.sync = true;
        method = &Benchmark::WriteRandom;
810
      } else if (name == Slice("fill100K")) {
811 812 813 814
        fresh_db = true;
        num_ /= 1000;
        value_size_ = 100 * 1000;
        method = &Benchmark::WriteRandom;
J
jorlow@chromium.org 已提交
815
      } else if (name == Slice("readseq")) {
816
        method = &Benchmark::ReadSequential;
J
jorlow@chromium.org 已提交
817
      } else if (name == Slice("readreverse")) {
818
        method = &Benchmark::ReadReverse;
J
jorlow@chromium.org 已提交
819
      } else if (name == Slice("readrandom")) {
820
        method = &Benchmark::ReadRandom;
S
Sanjay Ghemawat 已提交
821 822 823 824
      } else if (name == Slice("readmissing")) {
        method = &Benchmark::ReadMissing;
      } else if (name == Slice("seekrandom")) {
        method = &Benchmark::SeekRandom;
825
      } else if (name == Slice("readhot")) {
826
        method = &Benchmark::ReadHot;
827
      } else if (name == Slice("readrandomsmall")) {
828
        reads_ /= 1000;
829
        method = &Benchmark::ReadRandom;
S
Sanjay Ghemawat 已提交
830 831 832 833
      } else if (name == Slice("deleteseq")) {
        method = &Benchmark::DeleteSeq;
      } else if (name == Slice("deleterandom")) {
        method = &Benchmark::DeleteRandom;
834 835 836
      } else if (name == Slice("readwhilewriting")) {
        num_threads++;  // Add extra thread for writing
        method = &Benchmark::ReadWhileWriting;
837 838
      } else if (name == Slice("readrandomwriterandom")) {
        method = &Benchmark::ReadRandomWriteRandom;
M
Mark Callaghan 已提交
839 840
      } else if (name == Slice("updaterandom")) {
        method = &Benchmark::UpdateRandom;
841 842
      } else if (name == Slice("randomwithverify")) {
        method = &Benchmark::RandomWithVerify;
J
jorlow@chromium.org 已提交
843
      } else if (name == Slice("compact")) {
844
        method = &Benchmark::Compact;
J
jorlow@chromium.org 已提交
845
      } else if (name == Slice("crc32c")) {
846
        method = &Benchmark::Crc32c;
847
      } else if (name == Slice("acquireload")) {
848
        method = &Benchmark::AcquireLoad;
849
      } else if (name == Slice("snappycomp")) {
850
        method = &Benchmark::SnappyCompress;
851
      } else if (name == Slice("snappyuncomp")) {
852
        method = &Benchmark::SnappyUncompress;
J
jorlow@chromium.org 已提交
853 854
      } else if (name == Slice("heapprofile")) {
        HeapProfile();
855
      } else if (name == Slice("stats")) {
S
Sanjay Ghemawat 已提交
856
        PrintStats("leveldb.stats");
857 858
      } else if (name == Slice("levelstats")) {
        PrintStats("leveldb.levelstats");
S
Sanjay Ghemawat 已提交
859 860
      } else if (name == Slice("sstables")) {
        PrintStats("leveldb.sstables");
J
jorlow@chromium.org 已提交
861
      } else {
862 863 864 865
        if (name != Slice()) {  // No error message for empty name
          fprintf(stderr, "unknown benchmark '%s'\n", name.ToString().c_str());
        }
      }
866 867 868 869 870

      if (fresh_db) {
        if (FLAGS_use_existing_db) {
          fprintf(stdout, "%-12s : skipped (--use_existing_db is true)\n",
                  name.ToString().c_str());
871
          method = nullptr;
872 873
        } else {
          delete db_;
874
          db_ = nullptr;
875 876 877 878 879
          DestroyDB(FLAGS_db, Options());
          Open();
        }
      }

880
      if (method != nullptr) {
881
        RunBenchmark(num_threads, name, method);
J
jorlow@chromium.org 已提交
882 883
      }
    }
884
    PrintStatistics();
J
jorlow@chromium.org 已提交
885 886
  }

887
 private:
888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909
  struct ThreadArg {
    Benchmark* bm;
    SharedState* shared;
    ThreadState* thread;
    void (Benchmark::*method)(ThreadState*);
  };

  static void ThreadBody(void* v) {
    ThreadArg* arg = reinterpret_cast<ThreadArg*>(v);
    SharedState* shared = arg->shared;
    ThreadState* thread = arg->thread;
    {
      MutexLock l(&shared->mu);
      shared->num_initialized++;
      if (shared->num_initialized >= shared->total) {
        shared->cv.SignalAll();
      }
      while (!shared->start) {
        shared->cv.Wait();
      }
    }

910
    thread->stats.Start(thread->tid);
911 912 913 914 915 916 917 918 919 920 921 922
    (arg->bm->*(arg->method))(thread);
    thread->stats.Stop();

    {
      MutexLock l(&shared->mu);
      shared->num_done++;
      if (shared->num_done >= shared->total) {
        shared->cv.SignalAll();
      }
    }
  }

923 924
  void RunBenchmark(int n, Slice name,
                    void (Benchmark::*method)(ThreadState*)) {
925 926 927 928 929 930 931 932 933 934 935 936
    SharedState shared;
    shared.total = n;
    shared.num_initialized = 0;
    shared.num_done = 0;
    shared.start = false;

    ThreadArg* arg = new ThreadArg[n];
    for (int i = 0; i < n; i++) {
      arg[i].bm = this;
      arg[i].method = method;
      arg[i].shared = &shared;
      arg[i].thread = new ThreadState(i);
937
      arg[i].thread->shared = &shared;
938
      FLAGS_env->StartThread(ThreadBody, &arg[i]);
939 940 941 942 943 944 945 946 947 948 949 950 951 952
    }

    shared.mu.Lock();
    while (shared.num_initialized < n) {
      shared.cv.Wait();
    }

    shared.start = true;
    shared.cv.SignalAll();
    while (shared.num_done < n) {
      shared.cv.Wait();
    }
    shared.mu.Unlock();

953 954 955 956
    // Stats for some threads can be excluded.
    Stats merge_stats;
    for (int i = 0; i < n; i++) {
      merge_stats.Merge(arg[i].thread->stats);
957
    }
958
    merge_stats.Report(name);
959 960 961 962 963 964 965 966

    for (int i = 0; i < n; i++) {
      delete arg[i].thread;
    }
    delete[] arg;
  }

  void Crc32c(ThreadState* thread) {
J
jorlow@chromium.org 已提交
967
    // Checksum about 500MB of data total
968 969
    const int size = 4096;
    const char* label = "(4K per op)";
J
jorlow@chromium.org 已提交
970
    std::string data(size, 'x');
J
jorlow@chromium.org 已提交
971 972 973 974
    int64_t bytes = 0;
    uint32_t crc = 0;
    while (bytes < 500 * 1048576) {
      crc = crc32c::Value(data.data(), size);
975
      thread->stats.FinishedSingleOp(nullptr);
J
jorlow@chromium.org 已提交
976 977 978 979 980
      bytes += size;
    }
    // Print so result is not dead
    fprintf(stderr, "... crc=0x%x\r", static_cast<unsigned int>(crc));

981 982
    thread->stats.AddBytes(bytes);
    thread->stats.AddMessage(label);
J
jorlow@chromium.org 已提交
983 984
  }

985
  void AcquireLoad(ThreadState* thread) {
986 987 988
    int dummy;
    port::AtomicPointer ap(&dummy);
    int count = 0;
989
    void *ptr = nullptr;
990
    thread->stats.AddMessage("(each op is 1000 loads)");
991 992 993 994 995
    while (count < 100000) {
      for (int i = 0; i < 1000; i++) {
        ptr = ap.Acquire_Load();
      }
      count++;
996
      thread->stats.FinishedSingleOp(nullptr);
997
    }
998
    if (ptr == nullptr) exit(1); // Disable unused variable warning.
999 1000
  }

1001 1002 1003
  void SnappyCompress(ThreadState* thread) {
    RandomGenerator gen;
    Slice input = gen.Generate(Options().block_size);
1004 1005 1006 1007 1008
    int64_t bytes = 0;
    int64_t produced = 0;
    bool ok = true;
    std::string compressed;
    while (ok && bytes < 1024 * 1048576) {  // Compress 1G
1009 1010
      ok = port::Snappy_Compress(Options().compression_opts, input.data(),
                                 input.size(), &compressed);
1011 1012
      produced += compressed.size();
      bytes += input.size();
1013
      thread->stats.FinishedSingleOp(nullptr);
1014 1015 1016
    }

    if (!ok) {
1017
      thread->stats.AddMessage("(snappy failure)");
1018 1019 1020 1021
    } else {
      char buf[100];
      snprintf(buf, sizeof(buf), "(output: %.1f%%)",
               (produced * 100.0) / bytes);
1022 1023
      thread->stats.AddMessage(buf);
      thread->stats.AddBytes(bytes);
1024 1025 1026
    }
  }

1027 1028 1029
  void SnappyUncompress(ThreadState* thread) {
    RandomGenerator gen;
    Slice input = gen.Generate(Options().block_size);
1030
    std::string compressed;
1031 1032
    bool ok = port::Snappy_Compress(Options().compression_opts, input.data(),
                                    input.size(), &compressed);
1033
    int64_t bytes = 0;
1034
    char* uncompressed = new char[input.size()];
1035 1036
    while (ok && bytes < 1024 * 1048576) {  // Compress 1G
      ok =  port::Snappy_Uncompress(compressed.data(), compressed.size(),
1037 1038
                                    uncompressed);
      bytes += input.size();
1039
      thread->stats.FinishedSingleOp(nullptr);
1040
    }
1041
    delete[] uncompressed;
1042 1043

    if (!ok) {
1044
      thread->stats.AddMessage("(snappy failure)");
1045
    } else {
1046
      thread->stats.AddBytes(bytes);
1047 1048 1049
    }
  }

1050
  void Open() {
1051
    assert(db_ == nullptr);
1052
    Options options;
1053
    options.create_if_missing = !FLAGS_use_existing_db;
1054
    options.block_cache = cache_;
1055
    if (cache_ == nullptr) {
1056 1057
      options.no_block_cache = true;
    }
1058
    options.write_buffer_size = FLAGS_write_buffer_size;
1059 1060
    options.max_write_buffer_number = FLAGS_max_write_buffer_number;
    options.max_background_compactions = FLAGS_max_background_compactions;
1061
    options.block_size = FLAGS_block_size;
S
Sanjay Ghemawat 已提交
1062
    options.filter_policy = filter_policy_;
1063 1064
    options.max_open_files = FLAGS_open_files;
    options.statistics = dbstats;
1065
    options.env = FLAGS_env;
H
heyongqiang 已提交
1066
    options.disableDataSync = FLAGS_disable_data_sync;
1067
    options.use_fsync = FLAGS_use_fsync;
1068
    options.num_levels = FLAGS_num_levels;
H
heyongqiang 已提交
1069 1070 1071 1072 1073
    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;
H
heyongqiang 已提交
1074
    options.level0_stop_writes_trigger = FLAGS_level0_stop_writes_trigger;
M
Mark Callaghan 已提交
1075
    options.level0_file_num_compaction_trigger =
1076
        FLAGS_level0_file_num_compaction_trigger;
H
heyongqiang 已提交
1077 1078
    options.level0_slowdown_writes_trigger =
      FLAGS_level0_slowdown_writes_trigger;
1079
    options.compression = FLAGS_compression_type;
1080
    options.WAL_ttl_seconds = FLAGS_WAL_ttl_seconds;
1081 1082
    if (FLAGS_min_level_to_compress >= 0) {
      assert(FLAGS_min_level_to_compress <= FLAGS_num_levels);
1083
      options.compression_per_level.resize(FLAGS_num_levels);
1084
      for (int i = 0; i < FLAGS_min_level_to_compress; i++) {
1085 1086
        options.compression_per_level[i] = kNoCompression;
      }
1087
      for (int i = FLAGS_min_level_to_compress;
1088 1089 1090 1091
           i < FLAGS_num_levels; i++) {
        options.compression_per_level[i] = FLAGS_compression_type;
      }
    }
1092
    options.disable_seek_compaction = FLAGS_disable_seek_compaction;
1093 1094
    options.delete_obsolete_files_period_micros =
      FLAGS_delete_obsolete_files_period_micros;
1095
    options.rate_limit = FLAGS_rate_limit;
1096
    options.rate_limit_delay_milliseconds = FLAGS_rate_limit_delay_milliseconds;
1097
    options.table_cache_numshardbits = FLAGS_table_cache_numshardbits;
A
Abhishek Kona 已提交
1098
    options.max_grandparent_overlap_factor =
1099
      FLAGS_max_grandparent_overlap_factor;
1100
    options.disable_auto_compactions = FLAGS_disable_auto_compactions;
1101
    options.source_compaction_factor = FLAGS_source_compaction_factor;
H
heyongqiang 已提交
1102 1103 1104 1105 1106 1107
    Status s;
    if(FLAGS_read_only) {
      s = DB::OpenForReadOnly(options, FLAGS_db, &db_);
    } else {
      s = DB::Open(options, FLAGS_db, &db_);
    }
1108 1109 1110 1111
    if (!s.ok()) {
      fprintf(stderr, "open error: %s\n", s.ToString().c_str());
      exit(1);
    }
1112
    if (FLAGS_min_level_to_compress >= 0) {
1113
      options.compression_per_level.clear();
1114
    }
1115 1116
  }

1117 1118 1119
  void WriteSeq(ThreadState* thread) {
    DoWrite(thread, true);
  }
1120

1121 1122 1123 1124 1125
  void WriteRandom(ThreadState* thread) {
    DoWrite(thread, false);
  }

  void DoWrite(ThreadState* thread, bool seq) {
M
Mark Callaghan 已提交
1126 1127
    Duration duration(seq ? 0 : FLAGS_duration, writes_);

1128
    if (num_ != FLAGS_num) {
1129
      char msg[100];
1130
      snprintf(msg, sizeof(msg), "(%ld ops)", num_);
1131
      thread->stats.AddMessage(msg);
1132 1133
    }

1134
    RandomGenerator gen;
J
jorlow@chromium.org 已提交
1135 1136
    WriteBatch batch;
    Status s;
1137
    int64_t bytes = 0;
M
Mark Callaghan 已提交
1138 1139
    int i = 0;
    while (!duration.Done(entries_per_batch_)) {
J
jorlow@chromium.org 已提交
1140
      batch.Clear();
1141 1142
      for (int j = 0; j < entries_per_batch_; j++) {
        const int k = seq ? i+j : (thread->rand.Next() % FLAGS_num);
1143 1144 1145
        unique_ptr<char []> key = GenerateKeyFromInt(k);
        batch.Put(key.get(), gen.Generate(value_size_));
        bytes += value_size_ + strlen(key.get());
M
Mark Callaghan 已提交
1146
        thread->stats.FinishedSingleOp(db_);
1147
      }
1148
      s = db_->Write(write_options_, &batch);
J
jorlow@chromium.org 已提交
1149 1150 1151 1152
      if (!s.ok()) {
        fprintf(stderr, "put error: %s\n", s.ToString().c_str());
        exit(1);
      }
M
Mark Callaghan 已提交
1153
      i += entries_per_batch_;
J
jorlow@chromium.org 已提交
1154
    }
1155
    thread->stats.AddBytes(bytes);
J
jorlow@chromium.org 已提交
1156 1157
  }

1158
  void ReadSequential(ThreadState* thread) {
1159
    Iterator* iter = db_->NewIterator(ReadOptions(FLAGS_verify_checksum, true));
1160
    long i = 0;
1161
    int64_t bytes = 0;
1162
    for (iter->SeekToFirst(); i < reads_ && iter->Valid(); iter->Next()) {
1163
      bytes += iter->key().size() + iter->value().size();
M
Mark Callaghan 已提交
1164
      thread->stats.FinishedSingleOp(db_);
1165 1166 1167
      ++i;
    }
    delete iter;
1168
    thread->stats.AddBytes(bytes);
1169 1170
  }

1171
  void ReadReverse(ThreadState* thread) {
1172
    Iterator* iter = db_->NewIterator(ReadOptions(FLAGS_verify_checksum, true));
1173
    long i = 0;
1174
    int64_t bytes = 0;
1175
    for (iter->SeekToLast(); i < reads_ && iter->Valid(); iter->Prev()) {
1176
      bytes += iter->key().size() + iter->value().size();
M
Mark Callaghan 已提交
1177
      thread->stats.FinishedSingleOp(db_);
1178 1179 1180
      ++i;
    }
    delete iter;
1181
    thread->stats.AddBytes(bytes);
1182 1183
  }

1184
  void ReadRandom(ThreadState* thread) {
1185
    ReadOptions options(FLAGS_verify_checksum, true);
1186
    Iterator* iter = db_->NewIterator(options);
M
Mark Callaghan 已提交
1187
    Duration duration(FLAGS_duration, reads_);
1188

1189
    std::string value;
1190
    long found = 0;
M
Mark Callaghan 已提交
1191
    while (!duration.Done(1)) {
1192
      const int k = thread->rand.Next() % FLAGS_num;
1193
      unique_ptr<char []> key = GenerateKeyFromInt(k);
M
Mark Callaghan 已提交
1194 1195 1196 1197 1198

      if (FLAGS_use_snapshot) {
        options.snapshot = db_->GetSnapshot();
      }

1199
      if (FLAGS_read_range < 2) {
1200
        if (db_->Get(options, key.get(), &value).ok()) {
1201 1202 1203
          found++;
        }
      } else {
1204
        Slice skey(key.get());
1205
        int count = 1;
M
Mark Callaghan 已提交
1206 1207 1208 1209 1210 1211 1212 1213 1214 1215

        if (FLAGS_get_approx) {
          unique_ptr<char []> key2 =
              GenerateKeyFromInt(k + (int) FLAGS_read_range);
          Slice skey2(key2.get());
          Range range(skey, skey2);
          uint64_t sizes;
          db_->GetApproximateSizes(&range, 1, &sizes);
        }

1216 1217 1218 1219 1220
        for (iter->Seek(skey);
             iter->Valid() && count <= FLAGS_read_range;
             ++count, iter->Next()) {
          found++;
        }
S
Sanjay Ghemawat 已提交
1221
      }
1222

M
Mark Callaghan 已提交
1223 1224 1225 1226 1227
      if (FLAGS_use_snapshot) {
        db_->ReleaseSnapshot(options.snapshot);
        options.snapshot = nullptr;
      }

M
Mark Callaghan 已提交
1228
      thread->stats.FinishedSingleOp(db_);
S
Sanjay Ghemawat 已提交
1229
    }
1230
    delete iter;
S
Sanjay Ghemawat 已提交
1231
    char msg[100];
1232
    snprintf(msg, sizeof(msg), "(%ld of %ld found)", found, num_);
S
Sanjay Ghemawat 已提交
1233 1234 1235 1236
    thread->stats.AddMessage(msg);
  }

  void ReadMissing(ThreadState* thread) {
M
Mark Callaghan 已提交
1237
    Duration duration(FLAGS_duration, reads_);
1238
    ReadOptions options(FLAGS_verify_checksum, true);
S
Sanjay Ghemawat 已提交
1239
    std::string value;
M
Mark Callaghan 已提交
1240
    while (!duration.Done(1)) {
S
Sanjay Ghemawat 已提交
1241
      const int k = thread->rand.Next() % FLAGS_num;
1242 1243
      unique_ptr<char []> key = GenerateKeyFromInt(k);
      db_->Get(options, key.get(), &value);
M
Mark Callaghan 已提交
1244
      thread->stats.FinishedSingleOp(db_);
J
jorlow@chromium.org 已提交
1245 1246 1247
    }
  }

1248
  void ReadHot(ThreadState* thread) {
M
Mark Callaghan 已提交
1249
    Duration duration(FLAGS_duration, reads_);
1250
    ReadOptions options(FLAGS_verify_checksum, true);
1251
    std::string value;
1252
    const long range = (FLAGS_num + 99) / 100;
M
Mark Callaghan 已提交
1253
    while (!duration.Done(1)) {
1254
      const int k = thread->rand.Next() % range;
1255 1256
      unique_ptr<char []> key = GenerateKeyFromInt(k);
      db_->Get(options, key.get(), &value);
M
Mark Callaghan 已提交
1257
      thread->stats.FinishedSingleOp(db_);
1258 1259 1260
    }
  }

S
Sanjay Ghemawat 已提交
1261
  void SeekRandom(ThreadState* thread) {
M
Mark Callaghan 已提交
1262
    Duration duration(FLAGS_duration, reads_);
1263
    ReadOptions options(FLAGS_verify_checksum, true);
S
Sanjay Ghemawat 已提交
1264
    std::string value;
1265
    long found = 0;
M
Mark Callaghan 已提交
1266
    while (!duration.Done(1)) {
S
Sanjay Ghemawat 已提交
1267 1268
      Iterator* iter = db_->NewIterator(options);
      const int k = thread->rand.Next() % FLAGS_num;
1269 1270 1271
      unique_ptr<char []> key = GenerateKeyFromInt(k);
      iter->Seek(key.get());
      if (iter->Valid() && iter->key() == key.get()) found++;
S
Sanjay Ghemawat 已提交
1272
      delete iter;
M
Mark Callaghan 已提交
1273
      thread->stats.FinishedSingleOp(db_);
S
Sanjay Ghemawat 已提交
1274 1275
    }
    char msg[100];
1276
    snprintf(msg, sizeof(msg), "(%ld of %ld found)", found, num_);
S
Sanjay Ghemawat 已提交
1277 1278 1279 1280 1281 1282 1283
    thread->stats.AddMessage(msg);
  }

  void DoDelete(ThreadState* thread, bool seq) {
    RandomGenerator gen;
    WriteBatch batch;
    Status s;
M
Mark Callaghan 已提交
1284 1285 1286
    Duration duration(seq ? 0 : FLAGS_duration, num_);
    long i = 0;
    while (!duration.Done(entries_per_batch_)) {
S
Sanjay Ghemawat 已提交
1287 1288 1289
      batch.Clear();
      for (int j = 0; j < entries_per_batch_; j++) {
        const int k = seq ? i+j : (thread->rand.Next() % FLAGS_num);
1290 1291
        unique_ptr<char []> key = GenerateKeyFromInt(k);
        batch.Delete(key.get());
M
Mark Callaghan 已提交
1292
        thread->stats.FinishedSingleOp(db_);
S
Sanjay Ghemawat 已提交
1293 1294 1295 1296 1297 1298
      }
      s = db_->Write(write_options_, &batch);
      if (!s.ok()) {
        fprintf(stderr, "del error: %s\n", s.ToString().c_str());
        exit(1);
      }
M
Mark Callaghan 已提交
1299
      ++i;
S
Sanjay Ghemawat 已提交
1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310
    }
  }

  void DeleteSeq(ThreadState* thread) {
    DoDelete(thread, true);
  }

  void DeleteRandom(ThreadState* thread) {
    DoDelete(thread, false);
  }

1311 1312 1313 1314 1315 1316
  void ReadWhileWriting(ThreadState* thread) {
    if (thread->tid > 0) {
      ReadRandom(thread);
    } else {
      // Special thread that keeps writing until other threads are done.
      RandomGenerator gen;
1317 1318 1319 1320

      // Don't merge stats from this thread with the readers.
      thread->stats.SetExcludeFromMerge();

1321 1322 1323 1324 1325 1326 1327 1328 1329 1330
      while (true) {
        {
          MutexLock l(&thread->shared->mu);
          if (thread->shared->num_done + 1 >= thread->shared->num_initialized) {
            // Other threads have finished
            break;
          }
        }

        const int k = thread->rand.Next() % FLAGS_num;
1331 1332
        unique_ptr<char []> key = GenerateKeyFromInt(k);
        Status s = db_->Put(write_options_, key.get(), gen.Generate(value_size_));
1333 1334 1335 1336
        if (!s.ok()) {
          fprintf(stderr, "put error: %s\n", s.ToString().c_str());
          exit(1);
        }
1337
        thread->stats.FinishedSingleOp(db_);
1338 1339 1340 1341
      }
    }
  }

1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438
  // Given a key K and value V, this puts (K+"0", V), (K+"1", V), (K+"2", V)
  // in DB atomically i.e in a single batch. Also refer MultiGet.
  Status MultiPut(const WriteOptions& writeoptions,
                  const Slice& key, const Slice& value) {
    std::string suffixes[3] = {"2", "1", "0"};
    std::string keys[3];

    WriteBatch batch;
    Status s;
    for (int i = 0; i < 3; i++) {
      keys[i] = key.ToString() + suffixes[i];
      batch.Put(keys[i], value);
    }

    s = db_->Write(writeoptions, &batch);
    return s;
  }


  // Given a key K, this deletes (K+"0", V), (K+"1", V), (K+"2", V)
  // in DB atomically i.e in a single batch. Also refer MultiGet.
  Status MultiDelete(const WriteOptions& writeoptions,
                  const Slice& key) {
    std::string suffixes[3] = {"1", "2", "0"};
    std::string keys[3];

    WriteBatch batch;
    Status s;
    for (int i = 0; i < 3; i++) {
      keys[i] = key.ToString() + suffixes[i];
      batch.Delete(keys[i]);
    }

    s = db_->Write(writeoptions, &batch);
    return s;
  }

  // Given a key K and value V, this gets values for K+"0", K+"1" and K+"2"
  // in the same snapshot, and verifies that all the values are identical.
  // ASSUMES that MultiPut was used to put (K, V) into the DB.
  Status MultiGet(const ReadOptions& readoptions,
                  const Slice& key, std::string* value) {
    std::string suffixes[3] = {"0", "1", "2"};
    std::string keys[3];
    Slice key_slices[3];
    std::string values[3];
    ReadOptions readoptionscopy = readoptions;
    readoptionscopy.snapshot = db_->GetSnapshot();
    Status s;
    for (int i = 0; i < 3; i++) {
      keys[i] = key.ToString() + suffixes[i];
      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] = "";
        // we continue after error rather than exiting so that we can
        // find more errors if any
      } else if (s.IsNotFound()) {
        values[i] = "";
      } else {
        values[i] = *value;
      }
    }
    db_->ReleaseSnapshot(readoptionscopy.snapshot);

    if ((values[0] != values[1]) || (values[1] != values[2])) {
      fprintf(stderr, "inconsistent values for key %s: %s, %s, %s\n",
              key.ToString().c_str(), values[0].c_str(), values[1].c_str(),
              values[2].c_str());
      // we continue after error rather than exiting so that we can
      // find more errors if any
    }

    return s;
  }

  // Differs from readrandomwriterandom in the following ways:
  // (a) Uses MultiGet/MultiPut to read/write key values. Refer to those funcs.
  // (b) Does deletes as well (per FLAGS_deletepercent)
  // (c) In order to achieve high % of 'found' during lookups, and to do
  //     multiple writes (including puts and deletes) it uses upto
  //     FLAGS_numdistinct distinct keys instead of FLAGS_num distinct keys.
  void RandomWithVerify(ThreadState* thread) {
    ReadOptions options(FLAGS_verify_checksum, true);
    RandomGenerator gen;
    std::string value;
    long found = 0;
    int get_weight = 0;
    int put_weight = 0;
    int delete_weight = 0;
    long gets_done = 0;
    long puts_done = 0;
    long deletes_done = 0;
    // the number of iterations is the larger of read_ or write_
    for (long i = 0; i < readwrites_; i++) {
      const int k = thread->rand.Next() % (FLAGS_numdistinct);
1439
      unique_ptr<char []> key = GenerateKeyFromInt(k);
1440 1441 1442 1443 1444 1445 1446 1447
      if (get_weight == 0 && put_weight == 0 && delete_weight == 0) {
        // one batch complated, reinitialize for next batch
        get_weight = FLAGS_readwritepercent;
        delete_weight = FLAGS_deletepercent;
        put_weight = 100 - get_weight - delete_weight;
      }
      if (get_weight > 0) {
        // do all the gets first
1448
        Status s = MultiGet(options, key.get(), &value);
1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460
        if (!s.ok() && !s.IsNotFound()) {
          fprintf(stderr, "get error: %s\n", s.ToString().c_str());
          // we continue after error rather than exiting so that we can
          // find more errors if any
        } else if (!s.IsNotFound()) {
          found++;
        }
        get_weight--;
        gets_done++;
      } else if (put_weight > 0) {
        // then do all the corresponding number of puts
        // for all the gets we have done earlier
1461
        Status s = MultiPut(write_options_, key.get(), gen.Generate(value_size_));
1462 1463 1464 1465 1466 1467 1468
        if (!s.ok()) {
          fprintf(stderr, "multiput error: %s\n", s.ToString().c_str());
          exit(1);
        }
        put_weight--;
        puts_done++;
      } else if (delete_weight > 0) {
1469
        Status s = MultiDelete(write_options_, key.get());
1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485
        if (!s.ok()) {
          fprintf(stderr, "multidelete error: %s\n", s.ToString().c_str());
          exit(1);
        }
        delete_weight--;
        deletes_done++;
      }

      thread->stats.FinishedSingleOp(db_);
    }
    char msg[100];
    snprintf(msg, sizeof(msg), "( get:%ld put:%ld del:%ld total:%ld found:%ld)",
             gets_done, puts_done, deletes_done, readwrites_, found);
    thread->stats.AddMessage(msg);
  }

1486 1487
  //
  // This is diffferent from ReadWhileWriting because it does not use
1488
  // an extra thread.
1489 1490 1491 1492 1493 1494 1495 1496 1497 1498
  //
  void ReadRandomWriteRandom(ThreadState* thread) {
    ReadOptions options(FLAGS_verify_checksum, true);
    RandomGenerator gen;
    std::string value;
    long found = 0;
    int get_weight = 0;
    int put_weight = 0;
    long reads_done = 0;
    long writes_done = 0;
M
Mark Callaghan 已提交
1499 1500
    Duration duration(FLAGS_duration, readwrites_);

1501
    // the number of iterations is the larger of read_ or write_
M
Mark Callaghan 已提交
1502
    while (!duration.Done(1)) {
1503
      const int k = thread->rand.Next() % FLAGS_num;
1504
      unique_ptr<char []> key = GenerateKeyFromInt(k);
1505 1506 1507 1508 1509 1510
      if (get_weight == 0 && put_weight == 0) {
        // one batch complated, reinitialize for next batch
        get_weight = FLAGS_readwritepercent;
        put_weight = 100 - get_weight;
      }
      if (get_weight > 0) {
M
Mark Callaghan 已提交
1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525

        if (FLAGS_use_snapshot) {
          options.snapshot = db_->GetSnapshot();
        }

        if (FLAGS_get_approx) {
          char key2[100];
          snprintf(key2, sizeof(key2), "%016d", k + 1);
          Slice skey2(key2);
          Slice skey(key2);
          Range range(skey, skey2);
          uint64_t sizes;
          db_->GetApproximateSizes(&range, 1, &sizes);
        }

1526
        // do all the gets first
1527
        Status s = db_->Get(options, key.get(), &value);
1528 1529 1530 1531 1532 1533 1534 1535
        if (!s.ok() && !s.IsNotFound()) {
          fprintf(stderr, "get error: %s\n", s.ToString().c_str());
          // we continue after error rather than exiting so that we can
          // find more errors if any
        } else if (!s.IsNotFound()) {
          found++;
        }

1536
        if (db_->Get(options, key.get(), &value).ok()) {
1537 1538 1539 1540
          found++;
        }
        get_weight--;
        reads_done++;
M
Mark Callaghan 已提交
1541 1542 1543 1544 1545

        if (FLAGS_use_snapshot) {
          db_->ReleaseSnapshot(options.snapshot);
        }

1546 1547 1548
      } else  if (put_weight > 0) {
        // then do all the corresponding number of puts
        // for all the gets we have done earlier
1549
        Status s = db_->Put(write_options_, key.get(), gen.Generate(value_size_));
1550 1551 1552 1553 1554 1555 1556
        if (!s.ok()) {
          fprintf(stderr, "put error: %s\n", s.ToString().c_str());
          exit(1);
        }
        put_weight--;
        writes_done++;
      }
M
Mark Callaghan 已提交
1557
      thread->stats.FinishedSingleOp(db_);
1558 1559
    }
    char msg[100];
1560 1561
    snprintf(msg, sizeof(msg), "( reads:%ld writes:%ld total:%ld found:%ld)",
             reads_done, writes_done, readwrites_, found);
1562 1563 1564
    thread->stats.AddMessage(msg);
  }

M
Mark Callaghan 已提交
1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613
  //
  // Read-modify-write for random keys
  //
  void UpdateRandom(ThreadState* thread) {
    ReadOptions options(FLAGS_verify_checksum, true);
    RandomGenerator gen;
    std::string value;
    long found = 0;
    Duration duration(FLAGS_duration, readwrites_);

    // the number of iterations is the larger of read_ or write_
    while (!duration.Done(1)) {
      const int k = thread->rand.Next() % FLAGS_num;
      unique_ptr<char []> key = GenerateKeyFromInt(k);

      if (FLAGS_use_snapshot) {
        options.snapshot = db_->GetSnapshot();
      }

      if (FLAGS_get_approx) {
        char key2[100];
        snprintf(key2, sizeof(key2), "%016d", k + 1);
        Slice skey2(key2);
        Slice skey(key2);
        Range range(skey, skey2);
        uint64_t sizes;
        db_->GetApproximateSizes(&range, 1, &sizes);
      }

      if (db_->Get(options, key.get(), &value).ok()) {
        found++;
      }

      if (FLAGS_use_snapshot) {
        db_->ReleaseSnapshot(options.snapshot);
      }

      Status s = db_->Put(write_options_, key.get(), gen.Generate(value_size_));
      if (!s.ok()) {
        fprintf(stderr, "put error: %s\n", s.ToString().c_str());
        exit(1);
      }
      thread->stats.FinishedSingleOp(db_);
    }
    char msg[100];
    snprintf(msg, sizeof(msg), "( updates:%ld found:%ld)", readwrites_, found);
    thread->stats.AddMessage(msg);
  }

1614
  void Compact(ThreadState* thread) {
1615
    db_->CompactRange(nullptr, nullptr);
J
jorlow@chromium.org 已提交
1616 1617
  }

S
Sanjay Ghemawat 已提交
1618
  void PrintStats(const char* key) {
1619
    std::string stats;
S
Sanjay Ghemawat 已提交
1620
    if (!db_->GetProperty(key, &stats)) {
1621
      stats = "(failed)";
1622
    }
1623
    fprintf(stdout, "\n%s\n", stats.c_str());
1624 1625
  }

J
jorlow@chromium.org 已提交
1626 1627 1628 1629 1630 1631
  static void WriteToFile(void* arg, const char* buf, int n) {
    reinterpret_cast<WritableFile*>(arg)->Append(Slice(buf, n));
  }

  void HeapProfile() {
    char fname[100];
1632
    snprintf(fname, sizeof(fname), "%s/heap-%04d", FLAGS_db, ++heap_counter_);
1633
    unique_ptr<WritableFile> file;
1634
    Status s = FLAGS_env->NewWritableFile(fname, &file);
J
jorlow@chromium.org 已提交
1635
    if (!s.ok()) {
1636
      fprintf(stderr, "%s\n", s.ToString().c_str());
J
jorlow@chromium.org 已提交
1637 1638
      return;
    }
1639
    bool ok = port::GetHeapProfile(WriteToFile, file.get());
J
jorlow@chromium.org 已提交
1640
    if (!ok) {
1641
      fprintf(stderr, "heap profiling not supported\n");
1642
      FLAGS_env->DeleteFile(fname);
J
jorlow@chromium.org 已提交
1643 1644 1645 1646
    }
  }
};

H
Hans Wennborg 已提交
1647
}  // namespace leveldb
J
jorlow@chromium.org 已提交
1648 1649

int main(int argc, char** argv) {
1650
  FLAGS_write_buffer_size = leveldb::Options().write_buffer_size;
1651
  FLAGS_max_write_buffer_number = leveldb::Options().max_write_buffer_number;
1652
  FLAGS_open_files = leveldb::Options().max_open_files;
A
Abhishek Kona 已提交
1653
  FLAGS_max_background_compactions =
1654
    leveldb::Options().max_background_compactions;
1655 1656
  // Compression test code above refers to FLAGS_block_size
  FLAGS_block_size = leveldb::Options().block_size;
H
heyongqiang 已提交
1657
  std::string default_db_path;
1658

J
jorlow@chromium.org 已提交
1659 1660 1661
  for (int i = 1; i < argc; i++) {
    double d;
    int n;
1662
    long l;
J
jorlow@chromium.org 已提交
1663
    char junk;
1664
    char hdfsname[2048];
J
jorlow@chromium.org 已提交
1665 1666 1667 1668 1669 1670 1671
    if (leveldb::Slice(argv[i]).starts_with("--benchmarks=")) {
      FLAGS_benchmarks = argv[i] + strlen("--benchmarks=");
    } else if (sscanf(argv[i], "--compression_ratio=%lf%c", &d, &junk) == 1) {
      FLAGS_compression_ratio = d;
    } else if (sscanf(argv[i], "--histogram=%d%c", &n, &junk) == 1 &&
               (n == 0 || n == 1)) {
      FLAGS_histogram = n;
1672 1673 1674
    } else if (sscanf(argv[i], "--use_existing_db=%d%c", &n, &junk) == 1 &&
               (n == 0 || n == 1)) {
      FLAGS_use_existing_db = n;
1675 1676
    } else if (sscanf(argv[i], "--num=%ld%c", &l, &junk) == 1) {
      FLAGS_num = l;
1677 1678
    } else if (sscanf(argv[i], "--numdistinct=%ld%c", &l, &junk) == 1) {
      FLAGS_numdistinct = l;
1679 1680
    } else if (sscanf(argv[i], "--reads=%d%c", &n, &junk) == 1) {
      FLAGS_reads = n;
1681 1682
    } else if (sscanf(argv[i], "--read_range=%d%c", &n, &junk) == 1) {
      FLAGS_read_range = n;
M
Mark Callaghan 已提交
1683 1684
    } else if (sscanf(argv[i], "--duration=%d%c", &n, &junk) == 1) {
      FLAGS_duration = n;
1685 1686
    } else if (sscanf(argv[i], "--seed=%ld%c", &l, &junk) == 1) {
      FLAGS_seed = l;
1687 1688
    } else if (sscanf(argv[i], "--threads=%d%c", &n, &junk) == 1) {
      FLAGS_threads = n;
J
jorlow@chromium.org 已提交
1689 1690
    } else if (sscanf(argv[i], "--value_size=%d%c", &n, &junk) == 1) {
      FLAGS_value_size = n;
1691 1692 1693 1694 1695 1696 1697
    } else if (sscanf(argv[i], "--key_size=%d%c", &n, &junk) == 1) {
      if (MAX_KEY_SIZE < n) {
         fprintf(stderr, "key_size should not be larger than %d\n", MAX_KEY_SIZE);
         exit(1);
      } else {
        FLAGS_key_size = n;
      }
J
jorlow@chromium.org 已提交
1698 1699
    } else if (sscanf(argv[i], "--write_buffer_size=%d%c", &n, &junk) == 1) {
      FLAGS_write_buffer_size = n;
1700 1701 1702 1703
    } 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;
1704 1705
    } else if (sscanf(argv[i], "--cache_size=%ld%c", &l, &junk) == 1) {
      FLAGS_cache_size = l;
1706 1707
    } else if (sscanf(argv[i], "--block_size=%d%c", &n, &junk) == 1) {
      FLAGS_block_size = n;
1708
    } else if (sscanf(argv[i], "--cache_numshardbits=%d%c", &n, &junk) == 1) {
1709 1710 1711 1712 1713 1714
      if (n < 20) {
        FLAGS_cache_numshardbits = n;
      } else {
        fprintf(stderr, "The cache cannot be sharded into 2**%d pieces\n", n);
        exit(1);
      }
1715
    } else if (sscanf(argv[i], "--table_cache_numshardbits=%d%c",
A
Abhishek Kona 已提交
1716
          &n, &junk) == 1) {
1717 1718 1719 1720 1721
      if (n <= 0 || n > 20) {
        fprintf(stderr, "The cache cannot be sharded into 2**%d pieces\n", n);
        exit(1);
      }
      FLAGS_table_cache_numshardbits = n;
S
Sanjay Ghemawat 已提交
1722 1723
    } else if (sscanf(argv[i], "--bloom_bits=%d%c", &n, &junk) == 1) {
      FLAGS_bloom_bits = n;
1724 1725
    } else if (sscanf(argv[i], "--open_files=%d%c", &n, &junk) == 1) {
      FLAGS_open_files = n;
1726 1727
    } else if (strncmp(argv[i], "--db=", 5) == 0) {
      FLAGS_db = argv[i] + 5;
1728 1729 1730
    } else if (sscanf(argv[i], "--verify_checksum=%d%c", &n, &junk) == 1 &&
               (n == 0 || n == 1)) {
      FLAGS_verify_checksum = n;
1731 1732 1733
    } else if (sscanf(argv[i], "--bufferedio=%d%c", &n, &junk) == 1 &&
               (n == 0 || n == 1)) {
      useOsBuffer = n;
1734 1735 1736
    } else if (sscanf(argv[i], "--mmap_read=%d%c", &n, &junk) == 1 &&
               (n == 0 || n == 1)) {
      useMmapRead = n;
1737 1738 1739
    } else if (sscanf(argv[i], "--mmap_write=%d%c", &n, &junk) == 1 &&
               (n == 0 || n == 1)) {
      useMmapWrite = n;
1740
    } else if (sscanf(argv[i], "--readahead=%d%c", &n, &junk) == 1 &&
1741 1742
               (n == 0 || n == 1)) {
      useFsReadAhead = n;
1743 1744 1745 1746 1747 1748
    } else if (sscanf(argv[i], "--statistics=%d%c", &n, &junk) == 1 &&
               (n == 0 || n == 1)) {
      if (n == 1) {
        dbstats = new leveldb::DBStatistics();
        FLAGS_statistics = true;
      }
1749 1750 1751 1752 1753
    } else if (sscanf(argv[i], "--writes=%d%c", &n, &junk) == 1) {
      FLAGS_writes = n;
    } else if (sscanf(argv[i], "--sync=%d%c", &n, &junk) == 1 &&
               (n == 0 || n == 1)) {
      FLAGS_sync = n;
1754
    } else if (sscanf(argv[i], "--readwritepercent=%d%c", &n, &junk) == 1 &&
1755
               n > 0 && n < 100) {
1756
      FLAGS_readwritepercent = n;
1757 1758 1759
    } else if (sscanf(argv[i], "--deletepercent=%d%c", &n, &junk) == 1 &&
               n > 0 && n < 100) {
      FLAGS_deletepercent = n;
H
heyongqiang 已提交
1760 1761 1762
    } else if (sscanf(argv[i], "--disable_data_sync=%d%c", &n, &junk) == 1 &&
        (n == 0 || n == 1)) {
      FLAGS_disable_data_sync = n;
1763 1764 1765
    } else if (sscanf(argv[i], "--use_fsync=%d%c", &n, &junk) == 1 &&
        (n == 0 || n == 1)) {
      FLAGS_use_fsync = n;
H
heyongqiang 已提交
1766
    } else if (sscanf(argv[i], "--disable_wal=%d%c", &n, &junk) == 1 &&
1767
        (n == 0 || n == 1)) {
H
heyongqiang 已提交
1768
      FLAGS_disable_wal = n;
M
Mark Callaghan 已提交
1769 1770 1771 1772 1773 1774
    } else if (sscanf(argv[i], "--use_snapshot=%d%c", &n, &junk) == 1 &&
        (n == 0 || n == 1)) {
      FLAGS_use_snapshot = n;
    } else if (sscanf(argv[i], "--get_approx=%d%c", &n, &junk) == 1 &&
        (n == 0 || n == 1)) {
      FLAGS_get_approx = n;
1775
    } else if (sscanf(argv[i], "--hdfs=%s", hdfsname) == 1) {
1776
      FLAGS_env  = new leveldb::HdfsEnv(hdfsname);
1777 1778 1779
    } else if (sscanf(argv[i], "--num_levels=%d%c",
        &n, &junk) == 1) {
      FLAGS_num_levels = n;
H
heyongqiang 已提交
1780 1781 1782 1783 1784 1785 1786
    } 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 (
1787 1788
        sscanf(argv[i], "--max_bytes_for_level_base=%ld%c", &l, &junk) == 1) {
      FLAGS_max_bytes_for_level_base = l;
H
heyongqiang 已提交
1789 1790 1791
    } else if (sscanf(argv[i], "--max_bytes_for_level_multiplier=%d%c",
        &n, &junk) == 1) {
      FLAGS_max_bytes_for_level_multiplier = n;
H
heyongqiang 已提交
1792 1793 1794 1795 1796 1797
    } 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;
1798 1799 1800
    } else if (sscanf(argv[i],"--level0_file_num_compaction_trigger=%d%c",
        &n, &junk) == 1) {
      FLAGS_level0_file_num_compaction_trigger = n;
1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813
    } 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]);
      }
1814 1815
    } else if (sscanf(argv[i], "--min_level_to_compress=%d%c", &n, &junk) == 1
        && n >= 0) {
1816
      FLAGS_min_level_to_compress = n;
1817 1818 1819
    } else if (sscanf(argv[i], "--disable_seek_compaction=%d%c", &n, &junk) == 1
        && (n == 0 || n == 1)) {
      FLAGS_disable_seek_compaction = n;
1820 1821
    } else if (sscanf(argv[i], "--delete_obsolete_files_period_micros=%ld%c",
                      &l, &junk) == 1) {
1822
      FLAGS_delete_obsolete_files_period_micros = l;
1823 1824 1825
    } else if (sscanf(argv[i], "--stats_interval=%d%c", &n, &junk) == 1 &&
               n >= 0 && n < 2000000000) {
      FLAGS_stats_interval = n;
1826 1827 1828 1829 1830 1831
    } else if (sscanf(argv[i], "--stats_per_interval=%d%c", &n, &junk) == 1
        && (n == 0 || n == 1)) {
      FLAGS_stats_per_interval = n;
    } else if (sscanf(argv[i], "--rate_limit=%lf%c", &d, &junk) == 1 &&
               d > 1.0) {
      FLAGS_rate_limit = d;
1832 1833 1834 1835
    } else if (sscanf(argv[i],
               "--rate_limit_delay_milliseconds=%d%c", &n, &junk) == 1
        && n > 0) {
      FLAGS_rate_limit_delay_milliseconds = n;
H
heyongqiang 已提交
1836 1837 1838
    } else if (sscanf(argv[i], "--readonly=%d%c", &n, &junk) == 1 &&
        (n == 0 || n ==1 )) {
      FLAGS_read_only = n;
1839 1840 1841
    } else if (sscanf(argv[i], "--max_grandparent_overlap_factor=%d%c",
               &n, &junk) == 1) {
      FLAGS_max_grandparent_overlap_factor = n;
A
Abhishek Kona 已提交
1842
    } else if (sscanf(argv[i], "--disable_auto_compactions=%d%c",
1843 1844
               &n, &junk) == 1 && (n == 0 || n ==1)) {
      FLAGS_disable_auto_compactions = n;
A
Abhishek Kona 已提交
1845
    } else if (sscanf(argv[i], "--source_compaction_factor=%d%c",
1846 1847
               &n, &junk) == 1 && n > 0) {
      FLAGS_source_compaction_factor = n;
1848 1849
    } else if (sscanf(argv[i], "--wal_ttl=%d%c", &n, &junk) == 1) {
      FLAGS_WAL_ttl_seconds = static_cast<uint64_t>(n);
1850
    } else {
J
jorlow@chromium.org 已提交
1851 1852 1853 1854 1855
      fprintf(stderr, "Invalid flag '%s'\n", argv[i]);
      exit(1);
    }
  }

1856 1857 1858 1859
  // The number of background threads should be at least as much the
  // max number of concurrent compactions.
  FLAGS_env->SetBackgroundThreads(FLAGS_max_background_compactions);

H
heyongqiang 已提交
1860
  // Choose a location for the test database if none given with --db=<path>
1861
  if (FLAGS_db == nullptr) {
H
heyongqiang 已提交
1862 1863 1864 1865 1866
      leveldb::Env::Default()->GetTestDirectory(&default_db_path);
      default_db_path += "/dbbench";
      FLAGS_db = default_db_path.c_str();
  }

J
jorlow@chromium.org 已提交
1867 1868 1869 1870
  leveldb::Benchmark benchmark;
  benchmark.Run();
  return 0;
}