db_bench.cc 62.2 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;

272 273 274 275 276 277 278 279 280 281 282 283 284 285
// Allow buffered io using OS buffers
static bool FLAGS_use_os_buffer;

// Allow filesystem to do read-aheads
static bool FLAGS_use_fsreadahead;

// Allow reads to occur via mmap-ing files
static bool FLAGS_use_mmap_reads;

// Allow writes to occur via mmap-ing files
static bool FLAGS_use_mmap_writes;

// Allow readaheads to occur for compactions
static bool FLAGS_use_readahead_compactions;
286

J
jorlow@chromium.org 已提交
287 288
namespace leveldb {

289
// Helper for quickly generating random data.
J
jorlow@chromium.org 已提交
290 291 292
class RandomGenerator {
 private:
  std::string data_;
293
  unsigned int pos_;
J
jorlow@chromium.org 已提交
294 295 296 297 298 299 300 301

 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;
302
    while (data_.size() < (unsigned)std::max(1048576, FLAGS_value_size)) {
J
jorlow@chromium.org 已提交
303 304 305 306 307 308 309 310
      // 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;
  }

311
  Slice Generate(unsigned int len) {
J
jorlow@chromium.org 已提交
312 313 314 315 316 317 318
    if (pos_ + len > data_.size()) {
      pos_ = 0;
      assert(len < data_.size());
    }
    pos_ += len;
    return Slice(data_.data() + pos_ - len, len);
  }
319
};
320
static Slice TrimSpace(Slice s) {
321
  unsigned int start = 0;
322 323 324
  while (start < s.size() && isspace(s[start])) {
    start++;
  }
325
  unsigned int limit = s.size();
326 327 328 329 330 331
  while (limit > start && isspace(s[limit-1])) {
    limit--;
  }
  return Slice(s.data() + start, limit - start);
}

332 333 334 335 336 337 338 339 340 341
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:
342
  int id_;
343 344 345
  double start_;
  double finish_;
  double seconds_;
346
  long done_;
347
  long last_report_done_;
348 349 350
  int next_report_;
  int64_t bytes_;
  double last_op_finish_;
351
  double last_report_finish_;
352
  HistogramImpl hist_;
353
  std::string message_;
354
  bool exclude_from_merge_;
355 356

 public:
357
  Stats() { Start(-1); }
358

359 360 361
  void Start(int id) {
    id_ = id;
    next_report_ = FLAGS_stats_interval ? FLAGS_stats_interval : 100;
362 363 364
    last_op_finish_ = start_;
    hist_.Clear();
    done_ = 0;
365
    last_report_done_ = 0;
366 367
    bytes_ = 0;
    seconds_ = 0;
368
    start_ = FLAGS_env->NowMicros();
369
    finish_ = start_;
370
    last_report_finish_ = start_;
371
    message_.clear();
372 373
    // When set, stats from this thread won't be merged with others.
    exclude_from_merge_ = false;
374 375 376
  }

  void Merge(const Stats& other) {
377 378 379
    if (other.exclude_from_merge_)
      return;

380 381 382 383 384 385 386 387 388 389 390 391
    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() {
392
    finish_ = FLAGS_env->NowMicros();
393 394 395 396 397 398 399
    seconds_ = (finish_ - start_) * 1e-6;
  }

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

400
  void SetId(int id) { id_ = id; }
401
  void SetExcludeFromMerge() { exclude_from_merge_ = true; }
402

M
Mark Callaghan 已提交
403
  void FinishedSingleOp(DB* db) {
404
    if (FLAGS_histogram) {
405
      double now = FLAGS_env->NowMicros();
406 407
      double micros = now - last_op_finish_;
      hist_.Add(micros);
408
      if (micros > 20000 && !FLAGS_stats_interval) {
409 410 411 412 413 414 415 416
        fprintf(stderr, "long op: %.1f micros%30s\r", micros, "");
        fflush(stderr);
      }
      last_op_finish_ = now;
    }

    done_++;
    if (done_ >= next_report_) {
417 418 419 420 421 422 423 424 425 426 427 428 429
      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,
430
                "%s ... thread %d: (%ld,%ld) ops and (%.1f,%.1f) ops/second in (%.6f,%.6f) seconds\n",
431
                FLAGS_env->TimeToString((uint64_t) now/1000000).c_str(),
432
                id_,
M
Mark Callaghan 已提交
433
                done_ - last_report_done_, done_,
434
                (done_ - last_report_done_) /
M
Mark Callaghan 已提交
435 436 437 438
                ((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
439

440 441 442
        if (FLAGS_stats_per_interval) {
          std::string stats;
          if (db && db->GetProperty("leveldb.stats", &stats))
443
            fprintf(stderr, "%s\n", stats.c_str());
444
        }
M
Mark Callaghan 已提交
445

446 447 448 449 450
        fflush(stderr);
        next_report_ += FLAGS_stats_interval;
        last_report_finish_ = now;
        last_report_done_ = done_;
      }
451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473
    }
  }

  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_);
474 475
    double elapsed = (finish_ - start_) * 1e-6;
    double throughput = (double)done_/elapsed;
476

D
Dhruba Borthakur 已提交
477
    fprintf(stdout, "%-12s : %11.3f micros/op %ld ops/sec;%s%s\n",
478 479
            name.ToString().c_str(),
            seconds_ * 1e6 / done_,
D
Dhruba Borthakur 已提交
480
            (long)throughput,
481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501
            (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

502 503
  long num_initialized;
  long num_done;
504 505 506 507 508 509 510 511 512 513
  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;
514
  SharedState* shared;
515

A
Abhishek Kona 已提交
516
  /* implicit */ ThreadState(int index)
517
      : tid(index),
518
        rand((FLAGS_seed ? FLAGS_seed : 1000) + index) {
519 520 521
  }
};

M
Mark Callaghan 已提交
522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552
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
553 554
class Benchmark {
 private:
555
  shared_ptr<Cache> cache_;
S
Sanjay Ghemawat 已提交
556
  const FilterPolicy* filter_policy_;
J
jorlow@chromium.org 已提交
557
  DB* db_;
558
  long num_;
559
  int value_size_;
560
  int key_size_;
561 562
  int entries_per_batch_;
  WriteOptions write_options_;
563
  long reads_;
564
  long writes_;
565
  long readwrites_;
J
jorlow@chromium.org 已提交
566
  int heap_counter_;
567
  char keyFormat_[100]; // this string will contain the format of key. e.g "%016d"
568 569
  void PrintHeader() {
    PrintEnvironment();
570
    fprintf(stdout, "Keys:       %d bytes each\n", FLAGS_key_size);
571 572 573
    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));
574
    fprintf(stdout, "Entries:    %ld\n", num_);
575
    fprintf(stdout, "RawSize:    %.1f MB (estimated)\n",
576
            ((static_cast<int64_t>(FLAGS_key_size + FLAGS_value_size) * num_)
577
             / 1048576.0));
578
    fprintf(stdout, "FileSize:   %.1f MB (estimated)\n",
579
            (((FLAGS_key_size + FLAGS_value_size * FLAGS_compression_ratio) * num_)
580
             / 1048576.0));
581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596

    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;
    }

597 598 599 600 601 602 603 604 605 606 607 608 609 610
    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
611

612 613 614 615 616
    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;
617
      const char* name = nullptr;
618 619 620 621 622 623 624
      std::string compressed;

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

      switch (FLAGS_compression_type) {
        case kSnappyCompression:
625 626
          result = port::Snappy_Compress(Options().compression_opts, text,
                                         strlen(text), &compressed);
627 628 629
          name = "Snappy";
          break;
        case kZlibCompression:
630 631
          result = port::Zlib_Compress(Options().compression_opts, text,
                                       strlen(text), &compressed);
632 633 634
          name = "Zlib";
          break;
        case kBZip2Compression:
635 636
          result = port::BZip2_Compress(Options().compression_opts, text,
                                        strlen(text), &compressed);
637 638
          name = "BZip2";
          break;
639 640 641
        case kNoCompression:
          assert(false); // cannot happen
          break;
642 643 644 645 646 647 648 649 650
      }

      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);
651
    }
652 653 654 655 656 657 658
  }

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

#if defined(__linux)
659
    time_t now = time(nullptr);
660 661 662
    fprintf(stderr, "Date:       %s", ctime(&now));  // ctime() adds newline

    FILE* cpuinfo = fopen("/proc/cpuinfo", "r");
663
    if (cpuinfo != nullptr) {
664 665 666 667
      char line[1000];
      int num_cpus = 0;
      std::string cpu_type;
      std::string cache_size;
668
      while (fgets(line, sizeof(line), cpuinfo) != nullptr) {
669
        const char* sep = strchr(line, ':');
670
        if (sep == nullptr) {
671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688
          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
  }

689 690 691 692 693 694 695 696
  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);
  }

697 698
  void PrintStatistics() {
    if (FLAGS_statistics) {
A
Abhishek Kona 已提交
699
      fprintf(stdout, "File opened:%ld closed:%ld errors:%ld\n"
700 701 702
          "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 "
703
          "Compaction key_drop_user: %ld\n",
704 705 706
          dbstats->getTickerCount(NO_FILE_OPENS),
          dbstats->getTickerCount(NO_FILE_CLOSES),
          dbstats->getTickerCount(NO_FILE_ERRORS),
A
Abhishek Kona 已提交
707
          dbstats->getTickerCount(BLOCK_CACHE_HIT),
708 709 710 711 712
          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));
713 714
      PrintHistogram(DB_GET, "DB_GET");
      PrintHistogram(DB_WRITE, "DB_WRITE");
715 716 717
    }
  }

J
jorlow@chromium.org 已提交
718
 public:
719
  Benchmark()
720 721 722
  : cache_(FLAGS_cache_size >= 0 ?
           (FLAGS_cache_numshardbits >= 1 ?
            NewLRUCache(FLAGS_cache_size, FLAGS_cache_numshardbits) :
723
            NewLRUCache(FLAGS_cache_size)) : nullptr),
S
Sanjay Ghemawat 已提交
724 725
    filter_policy_(FLAGS_bloom_bits >= 0
                   ? NewBloomFilterPolicy(FLAGS_bloom_bits)
726 727
                   : nullptr),
    db_(nullptr),
728
    num_(FLAGS_num),
729
    value_size_(FLAGS_value_size),
730
    key_size_(FLAGS_key_size),
731
    entries_per_batch_(1),
732
    reads_(FLAGS_reads < 0 ? FLAGS_num : FLAGS_reads),
733
    writes_(FLAGS_writes < 0 ? FLAGS_num : FLAGS_writes),
734
    readwrites_((FLAGS_writes < 0  && FLAGS_reads < 0)? FLAGS_num :
735
                ((FLAGS_writes > FLAGS_reads) ? FLAGS_writes : FLAGS_reads)
736
               ),
737
    heap_counter_(0) {
J
jorlow@chromium.org 已提交
738
    std::vector<std::string> files;
739
    FLAGS_env->GetChildren(FLAGS_db, &files);
740
    for (unsigned int i = 0; i < files.size(); i++) {
J
jorlow@chromium.org 已提交
741
      if (Slice(files[i]).starts_with("heap-")) {
742
        FLAGS_env->DeleteFile(std::string(FLAGS_db) + "/" + files[i]);
J
jorlow@chromium.org 已提交
743 744
      }
    }
745
    if (!FLAGS_use_existing_db) {
746
      DestroyDB(FLAGS_db, Options());
747
    }
J
jorlow@chromium.org 已提交
748 749 750 751
  }

  ~Benchmark() {
    delete db_;
S
Sanjay Ghemawat 已提交
752
    delete filter_policy_;
J
jorlow@chromium.org 已提交
753
  }
754 755 756 757 758 759 760
//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
761

762 763 764 765 766 767
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
768
  void Run() {
769 770
    PrintHeader();
    Open();
J
jorlow@chromium.org 已提交
771 772

    const char* benchmarks = FLAGS_benchmarks;
773
    while (benchmarks != nullptr) {
J
jorlow@chromium.org 已提交
774 775
      const char* sep = strchr(benchmarks, ',');
      Slice name;
776
      if (sep == nullptr) {
J
jorlow@chromium.org 已提交
777
        name = benchmarks;
778
        benchmarks = nullptr;
J
jorlow@chromium.org 已提交
779 780 781 782 783
      } else {
        name = Slice(benchmarks, sep - benchmarks);
        benchmarks = sep + 1;
      }

784 785
      // Reset parameters that may be overriddden bwlow
      num_ = FLAGS_num;
786
      reads_ = (FLAGS_reads < 0 ? FLAGS_num : FLAGS_reads);
787
      writes_ = (FLAGS_writes < 0 ? FLAGS_num : FLAGS_writes);
788
      value_size_ = FLAGS_value_size;
789 790
      key_size_ = FLAGS_key_size;
      ConstructStrFormatForKey(keyFormat_, key_size_);
791 792
      entries_per_batch_ = 1;
      write_options_ = WriteOptions();
793 794 795
      if (FLAGS_sync) {
        write_options_.sync = true;
      }
H
heyongqiang 已提交
796 797
      write_options_.disableWAL = FLAGS_disable_wal;

798
      void (Benchmark::*method)(ThreadState*) = nullptr;
799
      bool fresh_db = false;
800
      int num_threads = FLAGS_threads;
801 802

      if (name == Slice("fillseq")) {
803 804
        fresh_db = true;
        method = &Benchmark::WriteSeq;
805
      } else if (name == Slice("fillbatch")) {
806 807 808
        fresh_db = true;
        entries_per_batch_ = 1000;
        method = &Benchmark::WriteSeq;
809
      } else if (name == Slice("fillrandom")) {
810 811
        fresh_db = true;
        method = &Benchmark::WriteRandom;
812
      } else if (name == Slice("overwrite")) {
813 814
        fresh_db = false;
        method = &Benchmark::WriteRandom;
815
      } else if (name == Slice("fillsync")) {
816 817 818 819
        fresh_db = true;
        num_ /= 1000;
        write_options_.sync = true;
        method = &Benchmark::WriteRandom;
820
      } else if (name == Slice("fill100K")) {
821 822 823 824
        fresh_db = true;
        num_ /= 1000;
        value_size_ = 100 * 1000;
        method = &Benchmark::WriteRandom;
J
jorlow@chromium.org 已提交
825
      } else if (name == Slice("readseq")) {
826
        method = &Benchmark::ReadSequential;
J
jorlow@chromium.org 已提交
827
      } else if (name == Slice("readreverse")) {
828
        method = &Benchmark::ReadReverse;
J
jorlow@chromium.org 已提交
829
      } else if (name == Slice("readrandom")) {
830
        method = &Benchmark::ReadRandom;
S
Sanjay Ghemawat 已提交
831 832 833 834
      } else if (name == Slice("readmissing")) {
        method = &Benchmark::ReadMissing;
      } else if (name == Slice("seekrandom")) {
        method = &Benchmark::SeekRandom;
835
      } else if (name == Slice("readhot")) {
836
        method = &Benchmark::ReadHot;
837
      } else if (name == Slice("readrandomsmall")) {
838
        reads_ /= 1000;
839
        method = &Benchmark::ReadRandom;
S
Sanjay Ghemawat 已提交
840 841 842 843
      } else if (name == Slice("deleteseq")) {
        method = &Benchmark::DeleteSeq;
      } else if (name == Slice("deleterandom")) {
        method = &Benchmark::DeleteRandom;
844 845 846
      } else if (name == Slice("readwhilewriting")) {
        num_threads++;  // Add extra thread for writing
        method = &Benchmark::ReadWhileWriting;
847 848
      } else if (name == Slice("readrandomwriterandom")) {
        method = &Benchmark::ReadRandomWriteRandom;
M
Mark Callaghan 已提交
849 850
      } else if (name == Slice("updaterandom")) {
        method = &Benchmark::UpdateRandom;
851 852
      } else if (name == Slice("randomwithverify")) {
        method = &Benchmark::RandomWithVerify;
J
jorlow@chromium.org 已提交
853
      } else if (name == Slice("compact")) {
854
        method = &Benchmark::Compact;
J
jorlow@chromium.org 已提交
855
      } else if (name == Slice("crc32c")) {
856
        method = &Benchmark::Crc32c;
857
      } else if (name == Slice("acquireload")) {
858
        method = &Benchmark::AcquireLoad;
859
      } else if (name == Slice("snappycomp")) {
860
        method = &Benchmark::SnappyCompress;
861
      } else if (name == Slice("snappyuncomp")) {
862
        method = &Benchmark::SnappyUncompress;
J
jorlow@chromium.org 已提交
863 864
      } else if (name == Slice("heapprofile")) {
        HeapProfile();
865
      } else if (name == Slice("stats")) {
S
Sanjay Ghemawat 已提交
866
        PrintStats("leveldb.stats");
867 868
      } else if (name == Slice("levelstats")) {
        PrintStats("leveldb.levelstats");
S
Sanjay Ghemawat 已提交
869 870
      } else if (name == Slice("sstables")) {
        PrintStats("leveldb.sstables");
J
jorlow@chromium.org 已提交
871
      } else {
872 873 874 875
        if (name != Slice()) {  // No error message for empty name
          fprintf(stderr, "unknown benchmark '%s'\n", name.ToString().c_str());
        }
      }
876 877 878 879 880

      if (fresh_db) {
        if (FLAGS_use_existing_db) {
          fprintf(stdout, "%-12s : skipped (--use_existing_db is true)\n",
                  name.ToString().c_str());
881
          method = nullptr;
882 883
        } else {
          delete db_;
884
          db_ = nullptr;
885 886 887 888 889
          DestroyDB(FLAGS_db, Options());
          Open();
        }
      }

890
      if (method != nullptr) {
891
        RunBenchmark(num_threads, name, method);
J
jorlow@chromium.org 已提交
892 893
      }
    }
894
    PrintStatistics();
J
jorlow@chromium.org 已提交
895 896
  }

897
 private:
898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919
  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();
      }
    }

920
    thread->stats.Start(thread->tid);
921 922 923 924 925 926 927 928 929 930 931 932
    (arg->bm->*(arg->method))(thread);
    thread->stats.Stop();

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

933 934
  void RunBenchmark(int n, Slice name,
                    void (Benchmark::*method)(ThreadState*)) {
935 936 937 938 939 940 941 942 943 944 945 946
    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);
947
      arg[i].thread->shared = &shared;
948
      FLAGS_env->StartThread(ThreadBody, &arg[i]);
949 950 951 952 953 954 955 956 957 958 959 960 961 962
    }

    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();

963 964 965 966
    // Stats for some threads can be excluded.
    Stats merge_stats;
    for (int i = 0; i < n; i++) {
      merge_stats.Merge(arg[i].thread->stats);
967
    }
968
    merge_stats.Report(name);
969 970 971 972 973 974 975 976

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

  void Crc32c(ThreadState* thread) {
J
jorlow@chromium.org 已提交
977
    // Checksum about 500MB of data total
978 979
    const int size = 4096;
    const char* label = "(4K per op)";
J
jorlow@chromium.org 已提交
980
    std::string data(size, 'x');
J
jorlow@chromium.org 已提交
981 982 983 984
    int64_t bytes = 0;
    uint32_t crc = 0;
    while (bytes < 500 * 1048576) {
      crc = crc32c::Value(data.data(), size);
985
      thread->stats.FinishedSingleOp(nullptr);
J
jorlow@chromium.org 已提交
986 987 988 989 990
      bytes += size;
    }
    // Print so result is not dead
    fprintf(stderr, "... crc=0x%x\r", static_cast<unsigned int>(crc));

991 992
    thread->stats.AddBytes(bytes);
    thread->stats.AddMessage(label);
J
jorlow@chromium.org 已提交
993 994
  }

995
  void AcquireLoad(ThreadState* thread) {
996 997 998
    int dummy;
    port::AtomicPointer ap(&dummy);
    int count = 0;
999
    void *ptr = nullptr;
1000
    thread->stats.AddMessage("(each op is 1000 loads)");
1001 1002 1003 1004 1005
    while (count < 100000) {
      for (int i = 0; i < 1000; i++) {
        ptr = ap.Acquire_Load();
      }
      count++;
1006
      thread->stats.FinishedSingleOp(nullptr);
1007
    }
1008
    if (ptr == nullptr) exit(1); // Disable unused variable warning.
1009 1010
  }

1011 1012 1013
  void SnappyCompress(ThreadState* thread) {
    RandomGenerator gen;
    Slice input = gen.Generate(Options().block_size);
1014 1015 1016 1017 1018
    int64_t bytes = 0;
    int64_t produced = 0;
    bool ok = true;
    std::string compressed;
    while (ok && bytes < 1024 * 1048576) {  // Compress 1G
1019 1020
      ok = port::Snappy_Compress(Options().compression_opts, input.data(),
                                 input.size(), &compressed);
1021 1022
      produced += compressed.size();
      bytes += input.size();
1023
      thread->stats.FinishedSingleOp(nullptr);
1024 1025 1026
    }

    if (!ok) {
1027
      thread->stats.AddMessage("(snappy failure)");
1028 1029 1030 1031
    } else {
      char buf[100];
      snprintf(buf, sizeof(buf), "(output: %.1f%%)",
               (produced * 100.0) / bytes);
1032 1033
      thread->stats.AddMessage(buf);
      thread->stats.AddBytes(bytes);
1034 1035 1036
    }
  }

1037 1038 1039
  void SnappyUncompress(ThreadState* thread) {
    RandomGenerator gen;
    Slice input = gen.Generate(Options().block_size);
1040
    std::string compressed;
1041 1042
    bool ok = port::Snappy_Compress(Options().compression_opts, input.data(),
                                    input.size(), &compressed);
1043
    int64_t bytes = 0;
1044
    char* uncompressed = new char[input.size()];
1045 1046
    while (ok && bytes < 1024 * 1048576) {  // Compress 1G
      ok =  port::Snappy_Uncompress(compressed.data(), compressed.size(),
1047 1048
                                    uncompressed);
      bytes += input.size();
1049
      thread->stats.FinishedSingleOp(nullptr);
1050
    }
1051
    delete[] uncompressed;
1052 1053

    if (!ok) {
1054
      thread->stats.AddMessage("(snappy failure)");
1055
    } else {
1056
      thread->stats.AddBytes(bytes);
1057 1058 1059
    }
  }

1060
  void Open() {
1061
    assert(db_ == nullptr);
1062
    Options options;
1063
    options.create_if_missing = !FLAGS_use_existing_db;
1064
    options.block_cache = cache_;
1065
    if (cache_ == nullptr) {
1066 1067
      options.no_block_cache = true;
    }
1068
    options.write_buffer_size = FLAGS_write_buffer_size;
1069 1070
    options.max_write_buffer_number = FLAGS_max_write_buffer_number;
    options.max_background_compactions = FLAGS_max_background_compactions;
1071
    options.block_size = FLAGS_block_size;
S
Sanjay Ghemawat 已提交
1072
    options.filter_policy = filter_policy_;
1073 1074
    options.max_open_files = FLAGS_open_files;
    options.statistics = dbstats;
1075
    options.env = FLAGS_env;
H
heyongqiang 已提交
1076
    options.disableDataSync = FLAGS_disable_data_sync;
1077
    options.use_fsync = FLAGS_use_fsync;
1078
    options.num_levels = FLAGS_num_levels;
H
heyongqiang 已提交
1079 1080 1081 1082 1083
    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
1084
    options.level0_stop_writes_trigger = FLAGS_level0_stop_writes_trigger;
M
Mark Callaghan 已提交
1085
    options.level0_file_num_compaction_trigger =
1086
        FLAGS_level0_file_num_compaction_trigger;
H
heyongqiang 已提交
1087 1088
    options.level0_slowdown_writes_trigger =
      FLAGS_level0_slowdown_writes_trigger;
1089
    options.compression = FLAGS_compression_type;
1090
    options.WAL_ttl_seconds = FLAGS_WAL_ttl_seconds;
1091 1092
    if (FLAGS_min_level_to_compress >= 0) {
      assert(FLAGS_min_level_to_compress <= FLAGS_num_levels);
1093
      options.compression_per_level.resize(FLAGS_num_levels);
1094
      for (int i = 0; i < FLAGS_min_level_to_compress; i++) {
1095 1096
        options.compression_per_level[i] = kNoCompression;
      }
1097
      for (int i = FLAGS_min_level_to_compress;
1098 1099 1100 1101
           i < FLAGS_num_levels; i++) {
        options.compression_per_level[i] = FLAGS_compression_type;
      }
    }
1102
    options.disable_seek_compaction = FLAGS_disable_seek_compaction;
1103 1104
    options.delete_obsolete_files_period_micros =
      FLAGS_delete_obsolete_files_period_micros;
1105
    options.rate_limit = FLAGS_rate_limit;
1106
    options.rate_limit_delay_milliseconds = FLAGS_rate_limit_delay_milliseconds;
1107
    options.table_cache_numshardbits = FLAGS_table_cache_numshardbits;
A
Abhishek Kona 已提交
1108
    options.max_grandparent_overlap_factor =
1109
      FLAGS_max_grandparent_overlap_factor;
1110
    options.disable_auto_compactions = FLAGS_disable_auto_compactions;
1111
    options.source_compaction_factor = FLAGS_source_compaction_factor;
1112 1113 1114 1115 1116 1117 1118

    // fill storage options
    options.allow_os_buffer = FLAGS_use_os_buffer;
    options.allow_readahead = FLAGS_use_fsreadahead;
    options.allow_mmap_reads = FLAGS_use_mmap_reads;
    options.allow_mmap_writes = FLAGS_use_mmap_writes;
    options.allow_readahead_compactions = FLAGS_use_readahead_compactions;
H
heyongqiang 已提交
1119 1120 1121 1122 1123 1124
    Status s;
    if(FLAGS_read_only) {
      s = DB::OpenForReadOnly(options, FLAGS_db, &db_);
    } else {
      s = DB::Open(options, FLAGS_db, &db_);
    }
1125 1126 1127 1128
    if (!s.ok()) {
      fprintf(stderr, "open error: %s\n", s.ToString().c_str());
      exit(1);
    }
1129
    if (FLAGS_min_level_to_compress >= 0) {
1130
      options.compression_per_level.clear();
1131
    }
1132 1133
  }

1134 1135 1136
  void WriteSeq(ThreadState* thread) {
    DoWrite(thread, true);
  }
1137

1138 1139 1140 1141 1142
  void WriteRandom(ThreadState* thread) {
    DoWrite(thread, false);
  }

  void DoWrite(ThreadState* thread, bool seq) {
M
Mark Callaghan 已提交
1143 1144
    Duration duration(seq ? 0 : FLAGS_duration, writes_);

1145
    if (num_ != FLAGS_num) {
1146
      char msg[100];
1147
      snprintf(msg, sizeof(msg), "(%ld ops)", num_);
1148
      thread->stats.AddMessage(msg);
1149 1150
    }

1151
    RandomGenerator gen;
J
jorlow@chromium.org 已提交
1152 1153
    WriteBatch batch;
    Status s;
1154
    int64_t bytes = 0;
M
Mark Callaghan 已提交
1155 1156
    int i = 0;
    while (!duration.Done(entries_per_batch_)) {
J
jorlow@chromium.org 已提交
1157
      batch.Clear();
1158 1159
      for (int j = 0; j < entries_per_batch_; j++) {
        const int k = seq ? i+j : (thread->rand.Next() % FLAGS_num);
1160 1161 1162
        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
1163
        thread->stats.FinishedSingleOp(db_);
1164
      }
1165
      s = db_->Write(write_options_, &batch);
J
jorlow@chromium.org 已提交
1166 1167 1168 1169
      if (!s.ok()) {
        fprintf(stderr, "put error: %s\n", s.ToString().c_str());
        exit(1);
      }
M
Mark Callaghan 已提交
1170
      i += entries_per_batch_;
J
jorlow@chromium.org 已提交
1171
    }
1172
    thread->stats.AddBytes(bytes);
J
jorlow@chromium.org 已提交
1173 1174
  }

1175
  void ReadSequential(ThreadState* thread) {
1176
    Iterator* iter = db_->NewIterator(ReadOptions(FLAGS_verify_checksum, true));
1177
    long i = 0;
1178
    int64_t bytes = 0;
1179
    for (iter->SeekToFirst(); i < reads_ && iter->Valid(); iter->Next()) {
1180
      bytes += iter->key().size() + iter->value().size();
M
Mark Callaghan 已提交
1181
      thread->stats.FinishedSingleOp(db_);
1182 1183 1184
      ++i;
    }
    delete iter;
1185
    thread->stats.AddBytes(bytes);
1186 1187
  }

1188
  void ReadReverse(ThreadState* thread) {
1189
    Iterator* iter = db_->NewIterator(ReadOptions(FLAGS_verify_checksum, true));
1190
    long i = 0;
1191
    int64_t bytes = 0;
1192
    for (iter->SeekToLast(); i < reads_ && iter->Valid(); iter->Prev()) {
1193
      bytes += iter->key().size() + iter->value().size();
M
Mark Callaghan 已提交
1194
      thread->stats.FinishedSingleOp(db_);
1195 1196 1197
      ++i;
    }
    delete iter;
1198
    thread->stats.AddBytes(bytes);
1199 1200
  }

1201
  void ReadRandom(ThreadState* thread) {
1202
    ReadOptions options(FLAGS_verify_checksum, true);
1203
    Iterator* iter = db_->NewIterator(options);
M
Mark Callaghan 已提交
1204
    Duration duration(FLAGS_duration, reads_);
1205

1206
    std::string value;
1207
    long found = 0;
M
Mark Callaghan 已提交
1208
    while (!duration.Done(1)) {
1209
      const int k = thread->rand.Next() % FLAGS_num;
1210
      unique_ptr<char []> key = GenerateKeyFromInt(k);
M
Mark Callaghan 已提交
1211 1212 1213 1214 1215

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

1216
      if (FLAGS_read_range < 2) {
1217
        if (db_->Get(options, key.get(), &value).ok()) {
1218 1219 1220
          found++;
        }
      } else {
1221
        Slice skey(key.get());
1222
        int count = 1;
M
Mark Callaghan 已提交
1223 1224 1225 1226 1227 1228 1229 1230 1231 1232

        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);
        }

1233 1234 1235 1236 1237
        for (iter->Seek(skey);
             iter->Valid() && count <= FLAGS_read_range;
             ++count, iter->Next()) {
          found++;
        }
S
Sanjay Ghemawat 已提交
1238
      }
1239

M
Mark Callaghan 已提交
1240 1241 1242 1243 1244
      if (FLAGS_use_snapshot) {
        db_->ReleaseSnapshot(options.snapshot);
        options.snapshot = nullptr;
      }

M
Mark Callaghan 已提交
1245
      thread->stats.FinishedSingleOp(db_);
S
Sanjay Ghemawat 已提交
1246
    }
1247
    delete iter;
S
Sanjay Ghemawat 已提交
1248
    char msg[100];
1249
    snprintf(msg, sizeof(msg), "(%ld of %ld found)", found, num_);
S
Sanjay Ghemawat 已提交
1250 1251 1252 1253
    thread->stats.AddMessage(msg);
  }

  void ReadMissing(ThreadState* thread) {
M
Mark Callaghan 已提交
1254
    Duration duration(FLAGS_duration, reads_);
1255
    ReadOptions options(FLAGS_verify_checksum, true);
S
Sanjay Ghemawat 已提交
1256
    std::string value;
M
Mark Callaghan 已提交
1257
    while (!duration.Done(1)) {
S
Sanjay Ghemawat 已提交
1258
      const int k = thread->rand.Next() % FLAGS_num;
1259 1260
      unique_ptr<char []> key = GenerateKeyFromInt(k);
      db_->Get(options, key.get(), &value);
M
Mark Callaghan 已提交
1261
      thread->stats.FinishedSingleOp(db_);
J
jorlow@chromium.org 已提交
1262 1263 1264
    }
  }

1265
  void ReadHot(ThreadState* thread) {
M
Mark Callaghan 已提交
1266
    Duration duration(FLAGS_duration, reads_);
1267
    ReadOptions options(FLAGS_verify_checksum, true);
1268
    std::string value;
1269
    const long range = (FLAGS_num + 99) / 100;
M
Mark Callaghan 已提交
1270
    while (!duration.Done(1)) {
1271
      const int k = thread->rand.Next() % range;
1272 1273
      unique_ptr<char []> key = GenerateKeyFromInt(k);
      db_->Get(options, key.get(), &value);
M
Mark Callaghan 已提交
1274
      thread->stats.FinishedSingleOp(db_);
1275 1276 1277
    }
  }

S
Sanjay Ghemawat 已提交
1278
  void SeekRandom(ThreadState* thread) {
M
Mark Callaghan 已提交
1279
    Duration duration(FLAGS_duration, reads_);
1280
    ReadOptions options(FLAGS_verify_checksum, true);
S
Sanjay Ghemawat 已提交
1281
    std::string value;
1282
    long found = 0;
M
Mark Callaghan 已提交
1283
    while (!duration.Done(1)) {
S
Sanjay Ghemawat 已提交
1284 1285
      Iterator* iter = db_->NewIterator(options);
      const int k = thread->rand.Next() % FLAGS_num;
1286 1287 1288
      unique_ptr<char []> key = GenerateKeyFromInt(k);
      iter->Seek(key.get());
      if (iter->Valid() && iter->key() == key.get()) found++;
S
Sanjay Ghemawat 已提交
1289
      delete iter;
M
Mark Callaghan 已提交
1290
      thread->stats.FinishedSingleOp(db_);
S
Sanjay Ghemawat 已提交
1291 1292
    }
    char msg[100];
1293
    snprintf(msg, sizeof(msg), "(%ld of %ld found)", found, num_);
S
Sanjay Ghemawat 已提交
1294 1295 1296 1297 1298 1299 1300
    thread->stats.AddMessage(msg);
  }

  void DoDelete(ThreadState* thread, bool seq) {
    RandomGenerator gen;
    WriteBatch batch;
    Status s;
M
Mark Callaghan 已提交
1301 1302 1303
    Duration duration(seq ? 0 : FLAGS_duration, num_);
    long i = 0;
    while (!duration.Done(entries_per_batch_)) {
S
Sanjay Ghemawat 已提交
1304 1305 1306
      batch.Clear();
      for (int j = 0; j < entries_per_batch_; j++) {
        const int k = seq ? i+j : (thread->rand.Next() % FLAGS_num);
1307 1308
        unique_ptr<char []> key = GenerateKeyFromInt(k);
        batch.Delete(key.get());
M
Mark Callaghan 已提交
1309
        thread->stats.FinishedSingleOp(db_);
S
Sanjay Ghemawat 已提交
1310 1311 1312 1313 1314 1315
      }
      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
1316
      ++i;
S
Sanjay Ghemawat 已提交
1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327
    }
  }

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

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

1328 1329 1330 1331 1332 1333
  void ReadWhileWriting(ThreadState* thread) {
    if (thread->tid > 0) {
      ReadRandom(thread);
    } else {
      // Special thread that keeps writing until other threads are done.
      RandomGenerator gen;
1334 1335 1336 1337

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

1338 1339 1340 1341 1342 1343 1344 1345 1346 1347
      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;
1348 1349
        unique_ptr<char []> key = GenerateKeyFromInt(k);
        Status s = db_->Put(write_options_, key.get(), gen.Generate(value_size_));
1350 1351 1352 1353
        if (!s.ok()) {
          fprintf(stderr, "put error: %s\n", s.ToString().c_str());
          exit(1);
        }
1354
        thread->stats.FinishedSingleOp(db_);
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 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455
  // 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);
1456
      unique_ptr<char []> key = GenerateKeyFromInt(k);
1457 1458 1459 1460 1461 1462 1463 1464
      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
1465
        Status s = MultiGet(options, key.get(), &value);
1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477
        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
1478
        Status s = MultiPut(write_options_, key.get(), gen.Generate(value_size_));
1479 1480 1481 1482 1483 1484 1485
        if (!s.ok()) {
          fprintf(stderr, "multiput error: %s\n", s.ToString().c_str());
          exit(1);
        }
        put_weight--;
        puts_done++;
      } else if (delete_weight > 0) {
1486
        Status s = MultiDelete(write_options_, key.get());
1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502
        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);
  }

1503 1504
  //
  // This is diffferent from ReadWhileWriting because it does not use
1505
  // an extra thread.
1506 1507 1508 1509 1510 1511 1512 1513 1514 1515
  //
  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
1516 1517
    Duration duration(FLAGS_duration, readwrites_);

1518
    // the number of iterations is the larger of read_ or write_
M
Mark Callaghan 已提交
1519
    while (!duration.Done(1)) {
1520
      const int k = thread->rand.Next() % FLAGS_num;
1521
      unique_ptr<char []> key = GenerateKeyFromInt(k);
1522 1523 1524 1525 1526 1527
      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
1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542

        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);
        }

1543
        // do all the gets first
1544
        Status s = db_->Get(options, key.get(), &value);
1545 1546 1547 1548 1549 1550 1551 1552
        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++;
        }

1553
        if (db_->Get(options, key.get(), &value).ok()) {
1554 1555 1556 1557
          found++;
        }
        get_weight--;
        reads_done++;
M
Mark Callaghan 已提交
1558 1559 1560 1561 1562

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

1563 1564 1565
      } else  if (put_weight > 0) {
        // then do all the corresponding number of puts
        // for all the gets we have done earlier
1566
        Status s = db_->Put(write_options_, key.get(), gen.Generate(value_size_));
1567 1568 1569 1570 1571 1572 1573
        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
1574
      thread->stats.FinishedSingleOp(db_);
1575 1576
    }
    char msg[100];
1577 1578
    snprintf(msg, sizeof(msg), "( reads:%ld writes:%ld total:%ld found:%ld)",
             reads_done, writes_done, readwrites_, found);
1579 1580 1581
    thread->stats.AddMessage(msg);
  }

M
Mark Callaghan 已提交
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 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630
  //
  // 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);
  }

1631
  void Compact(ThreadState* thread) {
1632
    db_->CompactRange(nullptr, nullptr);
J
jorlow@chromium.org 已提交
1633 1634
  }

S
Sanjay Ghemawat 已提交
1635
  void PrintStats(const char* key) {
1636
    std::string stats;
S
Sanjay Ghemawat 已提交
1637
    if (!db_->GetProperty(key, &stats)) {
1638
      stats = "(failed)";
1639
    }
1640
    fprintf(stdout, "\n%s\n", stats.c_str());
1641 1642
  }

J
jorlow@chromium.org 已提交
1643 1644 1645 1646 1647 1648
  static void WriteToFile(void* arg, const char* buf, int n) {
    reinterpret_cast<WritableFile*>(arg)->Append(Slice(buf, n));
  }

  void HeapProfile() {
    char fname[100];
1649
    StorageOptions soptions;
1650
    snprintf(fname, sizeof(fname), "%s/heap-%04d", FLAGS_db, ++heap_counter_);
1651
    unique_ptr<WritableFile> file;
1652
    Status s = FLAGS_env->NewWritableFile(fname, &file, soptions);
J
jorlow@chromium.org 已提交
1653
    if (!s.ok()) {
1654
      fprintf(stderr, "%s\n", s.ToString().c_str());
J
jorlow@chromium.org 已提交
1655 1656
      return;
    }
1657
    bool ok = port::GetHeapProfile(WriteToFile, file.get());
J
jorlow@chromium.org 已提交
1658
    if (!ok) {
1659
      fprintf(stderr, "heap profiling not supported\n");
1660
      FLAGS_env->DeleteFile(fname);
J
jorlow@chromium.org 已提交
1661 1662 1663 1664
    }
  }
};

H
Hans Wennborg 已提交
1665
}  // namespace leveldb
J
jorlow@chromium.org 已提交
1666 1667

int main(int argc, char** argv) {
1668
  FLAGS_write_buffer_size = leveldb::Options().write_buffer_size;
1669
  FLAGS_max_write_buffer_number = leveldb::Options().max_write_buffer_number;
1670
  FLAGS_open_files = leveldb::Options().max_open_files;
A
Abhishek Kona 已提交
1671
  FLAGS_max_background_compactions =
1672
    leveldb::Options().max_background_compactions;
1673 1674
  // Compression test code above refers to FLAGS_block_size
  FLAGS_block_size = leveldb::Options().block_size;
1675 1676 1677 1678 1679 1680 1681
  FLAGS_use_os_buffer = leveldb::StorageOptions().UseOsBuffer();
  FLAGS_use_fsreadahead = leveldb::StorageOptions().UseReadahead();
  FLAGS_use_mmap_reads = leveldb::StorageOptions().UseMmapReads();
  FLAGS_use_mmap_writes = leveldb::StorageOptions().UseMmapWrites();
  FLAGS_use_readahead_compactions =
    leveldb::StorageOptions().UseReadaheadCompactions();

H
heyongqiang 已提交
1682
  std::string default_db_path;
1683

J
jorlow@chromium.org 已提交
1684 1685 1686
  for (int i = 1; i < argc; i++) {
    double d;
    int n;
1687
    long l;
J
jorlow@chromium.org 已提交
1688
    char junk;
1689
    char hdfsname[2048];
J
jorlow@chromium.org 已提交
1690 1691 1692 1693 1694 1695 1696
    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;
1697 1698 1699
    } else if (sscanf(argv[i], "--use_existing_db=%d%c", &n, &junk) == 1 &&
               (n == 0 || n == 1)) {
      FLAGS_use_existing_db = n;
1700 1701
    } else if (sscanf(argv[i], "--num=%ld%c", &l, &junk) == 1) {
      FLAGS_num = l;
1702 1703
    } else if (sscanf(argv[i], "--numdistinct=%ld%c", &l, &junk) == 1) {
      FLAGS_numdistinct = l;
1704 1705
    } else if (sscanf(argv[i], "--reads=%d%c", &n, &junk) == 1) {
      FLAGS_reads = n;
1706 1707
    } else if (sscanf(argv[i], "--read_range=%d%c", &n, &junk) == 1) {
      FLAGS_read_range = n;
M
Mark Callaghan 已提交
1708 1709
    } else if (sscanf(argv[i], "--duration=%d%c", &n, &junk) == 1) {
      FLAGS_duration = n;
1710 1711
    } else if (sscanf(argv[i], "--seed=%ld%c", &l, &junk) == 1) {
      FLAGS_seed = l;
1712 1713
    } else if (sscanf(argv[i], "--threads=%d%c", &n, &junk) == 1) {
      FLAGS_threads = n;
J
jorlow@chromium.org 已提交
1714 1715
    } else if (sscanf(argv[i], "--value_size=%d%c", &n, &junk) == 1) {
      FLAGS_value_size = n;
1716 1717 1718 1719 1720 1721 1722
    } 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
1723 1724
    } else if (sscanf(argv[i], "--write_buffer_size=%d%c", &n, &junk) == 1) {
      FLAGS_write_buffer_size = n;
1725 1726 1727 1728
    } 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;
1729 1730
    } else if (sscanf(argv[i], "--cache_size=%ld%c", &l, &junk) == 1) {
      FLAGS_cache_size = l;
1731 1732
    } else if (sscanf(argv[i], "--block_size=%d%c", &n, &junk) == 1) {
      FLAGS_block_size = n;
1733
    } else if (sscanf(argv[i], "--cache_numshardbits=%d%c", &n, &junk) == 1) {
1734 1735 1736 1737 1738 1739
      if (n < 20) {
        FLAGS_cache_numshardbits = n;
      } else {
        fprintf(stderr, "The cache cannot be sharded into 2**%d pieces\n", n);
        exit(1);
      }
1740
    } else if (sscanf(argv[i], "--table_cache_numshardbits=%d%c",
A
Abhishek Kona 已提交
1741
          &n, &junk) == 1) {
1742 1743 1744 1745 1746
      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
1747 1748
    } else if (sscanf(argv[i], "--bloom_bits=%d%c", &n, &junk) == 1) {
      FLAGS_bloom_bits = n;
1749 1750
    } else if (sscanf(argv[i], "--open_files=%d%c", &n, &junk) == 1) {
      FLAGS_open_files = n;
1751 1752
    } else if (strncmp(argv[i], "--db=", 5) == 0) {
      FLAGS_db = argv[i] + 5;
1753 1754 1755
    } else if (sscanf(argv[i], "--verify_checksum=%d%c", &n, &junk) == 1 &&
               (n == 0 || n == 1)) {
      FLAGS_verify_checksum = n;
1756 1757
    } else if (sscanf(argv[i], "--bufferedio=%d%c", &n, &junk) == 1 &&
               (n == 0 || n == 1)) {
1758
      FLAGS_use_os_buffer = n;
1759 1760
    } else if (sscanf(argv[i], "--mmap_read=%d%c", &n, &junk) == 1 &&
               (n == 0 || n == 1)) {
1761
      FLAGS_use_mmap_reads = n;
1762 1763
    } else if (sscanf(argv[i], "--mmap_write=%d%c", &n, &junk) == 1 &&
               (n == 0 || n == 1)) {
1764
      FLAGS_use_mmap_writes = n;
1765
    } else if (sscanf(argv[i], "--readahead=%d%c", &n, &junk) == 1 &&
1766
               (n == 0 || n == 1)) {
1767 1768 1769 1770
      FLAGS_use_fsreadahead = n;
    } else if (sscanf(argv[i], "--readahead_compactions=%d%c", &n, &junk) == 1&&
               (n == 0 || n == 1)) {
      FLAGS_use_readahead_compactions = n;
1771 1772 1773 1774 1775 1776
    } 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;
      }
1777 1778 1779 1780 1781
    } 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;
1782
    } else if (sscanf(argv[i], "--readwritepercent=%d%c", &n, &junk) == 1 &&
1783
               n > 0 && n < 100) {
1784
      FLAGS_readwritepercent = n;
1785 1786 1787
    } else if (sscanf(argv[i], "--deletepercent=%d%c", &n, &junk) == 1 &&
               n > 0 && n < 100) {
      FLAGS_deletepercent = n;
H
heyongqiang 已提交
1788 1789 1790
    } else if (sscanf(argv[i], "--disable_data_sync=%d%c", &n, &junk) == 1 &&
        (n == 0 || n == 1)) {
      FLAGS_disable_data_sync = n;
1791 1792 1793
    } 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
1794
    } else if (sscanf(argv[i], "--disable_wal=%d%c", &n, &junk) == 1 &&
1795
        (n == 0 || n == 1)) {
H
heyongqiang 已提交
1796
      FLAGS_disable_wal = n;
M
Mark Callaghan 已提交
1797 1798 1799 1800 1801 1802
    } 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;
1803
    } else if (sscanf(argv[i], "--hdfs=%s", hdfsname) == 1) {
1804
      FLAGS_env  = new leveldb::HdfsEnv(hdfsname);
1805 1806 1807
    } else if (sscanf(argv[i], "--num_levels=%d%c",
        &n, &junk) == 1) {
      FLAGS_num_levels = n;
H
heyongqiang 已提交
1808 1809 1810 1811 1812 1813 1814
    } 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 (
1815 1816
        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
1817 1818 1819
    } 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
1820 1821 1822 1823 1824 1825
    } 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;
1826 1827 1828
    } else if (sscanf(argv[i],"--level0_file_num_compaction_trigger=%d%c",
        &n, &junk) == 1) {
      FLAGS_level0_file_num_compaction_trigger = n;
1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841
    } 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]);
      }
1842 1843
    } else if (sscanf(argv[i], "--min_level_to_compress=%d%c", &n, &junk) == 1
        && n >= 0) {
1844
      FLAGS_min_level_to_compress = n;
1845 1846 1847
    } else if (sscanf(argv[i], "--disable_seek_compaction=%d%c", &n, &junk) == 1
        && (n == 0 || n == 1)) {
      FLAGS_disable_seek_compaction = n;
1848 1849
    } else if (sscanf(argv[i], "--delete_obsolete_files_period_micros=%ld%c",
                      &l, &junk) == 1) {
1850
      FLAGS_delete_obsolete_files_period_micros = l;
1851 1852 1853
    } else if (sscanf(argv[i], "--stats_interval=%d%c", &n, &junk) == 1 &&
               n >= 0 && n < 2000000000) {
      FLAGS_stats_interval = n;
1854 1855 1856 1857 1858 1859
    } 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;
1860 1861 1862 1863
    } 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
1864 1865 1866
    } else if (sscanf(argv[i], "--readonly=%d%c", &n, &junk) == 1 &&
        (n == 0 || n ==1 )) {
      FLAGS_read_only = n;
1867 1868 1869
    } 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
1870
    } else if (sscanf(argv[i], "--disable_auto_compactions=%d%c",
1871 1872
               &n, &junk) == 1 && (n == 0 || n ==1)) {
      FLAGS_disable_auto_compactions = n;
A
Abhishek Kona 已提交
1873
    } else if (sscanf(argv[i], "--source_compaction_factor=%d%c",
1874 1875
               &n, &junk) == 1 && n > 0) {
      FLAGS_source_compaction_factor = n;
1876 1877
    } else if (sscanf(argv[i], "--wal_ttl=%d%c", &n, &junk) == 1) {
      FLAGS_WAL_ttl_seconds = static_cast<uint64_t>(n);
1878
    } else {
J
jorlow@chromium.org 已提交
1879 1880 1881 1882 1883
      fprintf(stderr, "Invalid flag '%s'\n", argv[i]);
      exit(1);
    }
  }

1884 1885 1886 1887
  // 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
1888
  // Choose a location for the test database if none given with --db=<path>
1889
  if (FLAGS_db == nullptr) {
H
heyongqiang 已提交
1890 1891 1892 1893 1894
      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
1895 1896 1897 1898
  leveldb::Benchmark benchmark;
  benchmark.Run();
  return 0;
}