db_stress.cc 67.1 KB
Newer Older
1 2 3 4 5
//  Copyright (c) 2013, Facebook, Inc.  All rights reserved.
//  This source code is licensed under the BSD-style license found in the
//  LICENSE file in the root directory of this source tree. An additional grant
//  of patent rights can be found in the PATENTS file in the same directory.
//
6 7 8
// Copyright (c) 2011 The LevelDB Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file. See the AUTHORS file for names of contributors.
9
//
X
Xing Jin 已提交
10 11 12 13 14 15 16 17 18 19 20 21
// The test uses an array to compare against values written to the database.
// Keys written to the array are in 1:1 correspondence to the actual values in
// the database according to the formula in the function GenerateValue.

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

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

31 32 33
#include <sys/types.h>
#include <stdio.h>
#include <stdlib.h>
34
#include <exception>
35
#include <gflags/gflags.h>
36 37
#include "db/db_impl.h"
#include "db/version_set.h"
I
Igor Canadi 已提交
38
#include "rocksdb/statistics.h"
39
#include "rocksdb/cache.h"
40
#include "rocksdb/utilities/db_ttl.h"
41 42
#include "rocksdb/env.h"
#include "rocksdb/write_batch.h"
43 44
#include "rocksdb/slice.h"
#include "rocksdb/slice_transform.h"
45
#include "port/port.h"
46
#include "util/coding.h"
47 48 49 50 51
#include "util/crc32c.h"
#include "util/histogram.h"
#include "util/mutexlock.h"
#include "util/random.h"
#include "util/testutil.h"
52
#include "util/logging.h"
53
#include "hdfs/env_hdfs.h"
D
Deon Nicholas 已提交
54
#include "utilities/merge_operators.h"
55

56 57 58
using GFLAGS::ParseCommandLineFlags;
using GFLAGS::RegisterFlagValidator;
using GFLAGS::SetUsageMessage;
59

60
static const long KB = 1024;
61

62 63
static bool ValidateUint32Range(const char* flagname, uint64_t value) {
  if (value > std::numeric_limits<uint32_t>::max()) {
K
kailiu 已提交
64 65 66 67
    fprintf(stderr,
            "Invalid value for --%s: %lu, overflow\n",
            flagname,
            (unsigned long)value);
68 69 70 71
    return false;
  }
  return true;
}
72

73
DEFINE_uint64(seed, 2341234, "Seed for PRNG");
I
Igor Canadi 已提交
74
static const bool FLAGS_seed_dummy __attribute__((unused)) =
75
    RegisterFlagValidator(&FLAGS_seed, &ValidateUint32Range);
76

I
Igor Canadi 已提交
77
DEFINE_int64(max_key, 1 * KB* KB,
78
             "Max number of key/values to place in database");
79

I
Igor Canadi 已提交
80 81
DEFINE_int32(column_families, 10, "Number of column families");

82
DEFINE_bool(test_batches_snapshots, false,
I
Igor Canadi 已提交
83
            "If set, the test uses MultiGet(), Multiut() and MultiDelete()"
84 85 86 87 88 89 90 91 92
            " which read/write/delete multiple keys in a batch. In this mode,"
            " we do not verify db content by comparing the content with the "
            "pre-allocated array. Instead, we do partial verification inside"
            " MultiGet() by checking various values in a batch. Benefit of"
            " this mode:\n"
            "\t(a) No need to acquire mutexes during writes (less cache "
            "flushes in multi-core leading to speed up)\n"
            "\t(b) No long validation at the end (more speed up)\n"
            "\t(c) Test snapshot and atomicity of batch writes");
93

94
DEFINE_int32(threads, 32, "Number of concurrent threads to run.");
95

96 97 98 99
DEFINE_int32(ttl, -1,
             "Opens the db with this ttl value if this is not -1. "
             "Carefully specify a large value such that verifications on "
             "deleted values don't fail");
100

101 102
DEFINE_int32(value_size_mult, 8,
             "Size of value will be this number times rand_int(1,3) bytes");
103

104
DEFINE_bool(verify_before_write, false, "Verify before write");
105

106
DEFINE_bool(histogram, false, "Print histogram of operation timings");
107

108 109
DEFINE_bool(destroy_db_initially, true,
            "Destroys the database dir before start if this is true");
110

111 112 113 114
DEFINE_bool(verbose, false, "Verbose");

DEFINE_bool(progress_reports, true,
            "If true, db_stress will report number of finished operations");
115

116 117
DEFINE_int32(write_buffer_size, rocksdb::Options().write_buffer_size,
             "Number of bytes to buffer in memtable before compacting");
118

119 120 121 122
DEFINE_int32(max_write_buffer_number,
             rocksdb::Options().max_write_buffer_number,
             "The number of in-memory memtables. "
             "Each memtable is of size FLAGS_write_buffer_size.");
123

124 125 126 127 128 129 130 131 132 133
DEFINE_int32(min_write_buffer_number_to_merge,
             rocksdb::Options().min_write_buffer_number_to_merge,
             "The minimum number of write buffers that will be merged together "
             "before writing to storage. This is cheap because it is an "
             "in-memory merge. If this feature is not enabled, then all these "
             "write buffers are flushed to L0 as separate files and this "
             "increases read amplification because a get request has to check "
             "in all of these files. Also, an in-memory merge may result in "
             "writing less data to storage if there are duplicate records in"
             " each of these individual write buffers.");
134

135 136 137
DEFINE_int32(open_files, rocksdb::Options().max_open_files,
             "Maximum number of files to keep open at the same time "
             "(use default if == 0)");
138

139 140 141 142
DEFINE_int64(compressed_cache_size, -1,
             "Number of bytes to use as a cache of compressed data."
             " Negative means use default settings.");

143
DEFINE_int32(compaction_style, rocksdb::Options().compaction_style, "");
144

145 146 147
DEFINE_int32(level0_file_num_compaction_trigger,
             rocksdb::Options().level0_file_num_compaction_trigger,
             "Level0 compaction start trigger");
148

149 150 151
DEFINE_int32(level0_slowdown_writes_trigger,
             rocksdb::Options().level0_slowdown_writes_trigger,
             "Number of files in level-0 that will slow down writes");
152

153 154 155
DEFINE_int32(level0_stop_writes_trigger,
             rocksdb::Options().level0_stop_writes_trigger,
             "Number of files in level-0 that will trigger put stop.");
156

157
DEFINE_int32(block_size, rocksdb::BlockBasedTableOptions().block_size,
158
             "Number of bytes in a block.");
159

160 161 162 163
DEFINE_int32(max_background_compactions,
             rocksdb::Options().max_background_compactions,
             "The maximum number of concurrent background compactions "
             "that can occur in parallel.");
164

165 166 167 168 169 170 171 172
DEFINE_int32(compaction_thread_pool_adjust_interval, 0,
             "The interval (in milliseconds) to adjust compaction thread pool "
             "size. Don't change it periodically if the value is 0.");

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

I
Igor Canadi 已提交
173 174 175 176
DEFINE_int32(max_background_flushes, rocksdb::Options().max_background_flushes,
             "The maximum number of concurrent background flushes "
             "that can occur in parallel.");

177 178
DEFINE_int32(universal_size_ratio, 0, "The ratio of file sizes that trigger"
             " compaction in universal style");
179

180 181
DEFINE_int32(universal_min_merge_width, 0, "The minimum number of files to "
             "compact in universal style compaction");
182

183 184
DEFINE_int32(universal_max_merge_width, 0, "The max number of files to compact"
             " in universal style compaction");
185

186 187
DEFINE_int32(universal_max_size_amplification_percent, 0,
             "The max size amplification for universal style compaction");
188

I
Igor Canadi 已提交
189 190 191 192 193
DEFINE_int32(clear_column_family_one_in, 1000000,
             "With a chance of 1/N, delete a column family and then recreate "
             "it again. If N == 0, never drop/create column families. "
             "When test_batches_snapshots is true, this flag has no effect");

L
Lei Jin 已提交
194 195 196
DEFINE_int32(set_options_one_in, 0,
             "With a chance of 1/N, change some random options");

197 198 199
DEFINE_int32(set_in_place_one_in, 0,
             "With a chance of 1/N, toggle in place support option");

200 201
DEFINE_int64(cache_size, 2 * KB * KB * KB,
             "Number of bytes to use as a cache of uncompressed data.");
202

203 204 205 206 207 208 209 210 211
static bool ValidateInt32Positive(const char* flagname, int32_t value) {
  if (value < 0) {
    fprintf(stderr, "Invalid value for --%s: %d, must be >=0\n",
            flagname, value);
    return false;
  }
  return true;
}
DEFINE_int32(reopen, 10, "Number of times database reopens");
I
Igor Canadi 已提交
212
static const bool FLAGS_reopen_dummy __attribute__((unused)) =
213
    RegisterFlagValidator(&FLAGS_reopen, &ValidateInt32Positive);
214

215 216
DEFINE_int32(bloom_bits, 10, "Bloom filter bits per key. "
             "Negative means use default settings.");
217

218 219 220
DEFINE_bool(use_block_based_filter, false, "use block based filter"
              "instead of full filter for block based table");

221
DEFINE_string(db, "", "Use the db with the following name.");
222

223 224
DEFINE_bool(verify_checksum, false,
            "Verify checksum for every block read from storage");
225

226 227
DEFINE_bool(mmap_read, rocksdb::EnvOptions().use_mmap_reads,
            "Allow reads to occur via mmap-ing files");
228

229 230 231
// Database statistics
static std::shared_ptr<rocksdb::Statistics> dbstats;
DEFINE_bool(statistics, false, "Create database statistics");
232

233
DEFINE_bool(sync, false, "Sync all writes to disk");
234

235 236
DEFINE_bool(disable_data_sync, false,
            "If true, do not wait until data is synced to disk.");
A
amayank 已提交
237

238
DEFINE_bool(use_fsync, false, "If true, issue fsync instead of fdatasync");
239

240 241 242
DEFINE_int32(kill_random_test, 0,
             "If non-zero, kill at various points in source code with "
             "probability 1/this");
I
Igor Canadi 已提交
243
static const bool FLAGS_kill_random_test_dummy __attribute__((unused)) =
244
    RegisterFlagValidator(&FLAGS_kill_random_test, &ValidateInt32Positive);
245
extern int rocksdb_kill_odds;
246

247
DEFINE_bool(disable_wal, false, "If true, do not write WAL for write.");
248

249 250
DEFINE_int32(target_file_size_base, 64 * KB,
             "Target level-1 file size for compaction");
251

252 253
DEFINE_int32(target_file_size_multiplier, 1,
             "A multiplier to compute targe level-N file size (N >= 2)");
254

255
DEFINE_uint64(max_bytes_for_level_base, 256 * KB, "Max bytes for level-1");
256

257 258
DEFINE_int32(max_bytes_for_level_multiplier, 2,
             "A multiplier to compute max bytes for level-N (N >= 2)");
259

260 261 262 263 264 265 266 267 268 269
static bool ValidateInt32Percent(const char* flagname, int32_t value) {
  if (value < 0 || value>100) {
    fprintf(stderr, "Invalid value for --%s: %d, 0<= pct <=100 \n",
            flagname, value);
    return false;
  }
  return true;
}
DEFINE_int32(readpercent, 10,
             "Ratio of reads to total workload (expressed as a percentage)");
I
Igor Canadi 已提交
270
static const bool FLAGS_readpercent_dummy __attribute__((unused)) =
271
    RegisterFlagValidator(&FLAGS_readpercent, &ValidateInt32Percent);
272 273 274 275

DEFINE_int32(prefixpercent, 20,
             "Ratio of prefix iterators to total workload (expressed as a"
             " percentage)");
I
Igor Canadi 已提交
276
static const bool FLAGS_prefixpercent_dummy __attribute__((unused)) =
277
    RegisterFlagValidator(&FLAGS_prefixpercent, &ValidateInt32Percent);
278 279 280

DEFINE_int32(writepercent, 45,
             " Ratio of deletes to total workload (expressed as a percentage)");
I
Igor Canadi 已提交
281
static const bool FLAGS_writepercent_dummy __attribute__((unused)) =
282
    RegisterFlagValidator(&FLAGS_writepercent, &ValidateInt32Percent);
283 284 285

DEFINE_int32(delpercent, 15,
             "Ratio of deletes to total workload (expressed as a percentage)");
I
Igor Canadi 已提交
286
static const bool FLAGS_delpercent_dummy __attribute__((unused)) =
287
    RegisterFlagValidator(&FLAGS_delpercent, &ValidateInt32Percent);
288 289 290

DEFINE_int32(iterpercent, 10, "Ratio of iterations to total workload"
             " (expressed as a percentage)");
I
Igor Canadi 已提交
291
static const bool FLAGS_iterpercent_dummy __attribute__((unused)) =
292
    RegisterFlagValidator(&FLAGS_iterpercent, &ValidateInt32Percent);
293 294

DEFINE_uint64(num_iterations, 10, "Number of iterations per MultiIterate run");
I
Igor Canadi 已提交
295
static const bool FLAGS_num_iterations_dummy __attribute__((unused)) =
296
    RegisterFlagValidator(&FLAGS_num_iterations, &ValidateUint32Range);
297

I
Igor Canadi 已提交
298
namespace {
299 300 301 302 303 304 305 306 307 308 309
enum rocksdb::CompressionType StringToCompressionType(const char* ctype) {
  assert(ctype);

  if (!strcasecmp(ctype, "none"))
    return rocksdb::kNoCompression;
  else if (!strcasecmp(ctype, "snappy"))
    return rocksdb::kSnappyCompression;
  else if (!strcasecmp(ctype, "zlib"))
    return rocksdb::kZlibCompression;
  else if (!strcasecmp(ctype, "bzip2"))
    return rocksdb::kBZip2Compression;
A
Albert Strasheim 已提交
310 311 312 313
  else if (!strcasecmp(ctype, "lz4"))
    return rocksdb::kLZ4Compression;
  else if (!strcasecmp(ctype, "lz4hc"))
    return rocksdb::kLZ4HCCompression;
314 315 316 317

  fprintf(stdout, "Cannot parse compression type '%s'\n", ctype);
  return rocksdb::kSnappyCompression; //default value
}
I
Igor Canadi 已提交
318 319
}  // namespace

320 321 322
DEFINE_string(compression_type, "snappy",
              "Algorithm to use to compress the database");
static enum rocksdb::CompressionType FLAGS_compression_type_e =
323
    rocksdb::kSnappyCompression;
324

325
DEFINE_string(hdfs, "", "Name of hdfs environment");
326
// posix or hdfs environment
327
static rocksdb::Env* FLAGS_env = rocksdb::Env::Default();
328

I
Igor Canadi 已提交
329 330
DEFINE_uint64(ops_per_thread, 1200000, "Number of operations per thread.");
static const bool FLAGS_ops_per_thread_dummy __attribute__((unused)) =
331
    RegisterFlagValidator(&FLAGS_ops_per_thread, &ValidateUint32Range);
332

333
DEFINE_uint64(log2_keys_per_lock, 2, "Log2 of number of keys per lock");
I
Igor Canadi 已提交
334
static const bool FLAGS_log2_keys_per_lock_dummy __attribute__((unused)) =
335
    RegisterFlagValidator(&FLAGS_log2_keys_per_lock, &ValidateUint32Range);
336

337 338 339
DEFINE_int32(purge_redundant_percent, 50,
             "Percentage of times we want to purge redundant keys in memory "
             "before flushing");
I
Igor Canadi 已提交
340
static const bool FLAGS_purge_redundant_percent_dummy __attribute__((unused)) =
341 342
    RegisterFlagValidator(&FLAGS_purge_redundant_percent,
                          &ValidateInt32Percent);
343

344 345
DEFINE_bool(filter_deletes, false, "On true, deletes use KeyMayExist to drop"
            " the delete if key not present");
346

347 348
DEFINE_bool(in_place_update, false, "On true, does inplace update in memtable");

J
Jim Paton 已提交
349 350
enum RepFactory {
  kSkipList,
I
Igor Canadi 已提交
351
  kHashSkipList,
J
Jim Paton 已提交
352 353
  kVectorRep
};
I
Igor Canadi 已提交
354 355

namespace {
356 357 358 359 360 361
enum RepFactory StringToRepFactory(const char* ctype) {
  assert(ctype);

  if (!strcasecmp(ctype, "skip_list"))
    return kSkipList;
  else if (!strcasecmp(ctype, "prefix_hash"))
I
Igor Canadi 已提交
362
    return kHashSkipList;
363 364 365 366 367 368
  else if (!strcasecmp(ctype, "vector"))
    return kVectorRep;

  fprintf(stdout, "Cannot parse memreptable %s\n", ctype);
  return kSkipList;
}
I
Igor Canadi 已提交
369 370
}  // namespace

J
Jim Paton 已提交
371
static enum RepFactory FLAGS_rep_factory;
L
Lei Jin 已提交
372
DEFINE_string(memtablerep, "prefix_hash", "");
J
Jim Paton 已提交
373

374
static bool ValidatePrefixSize(const char* flagname, int32_t value) {
L
Lei Jin 已提交
375 376
  if (value < 0 || value > 8) {
    fprintf(stderr, "Invalid value for --%s: %d. 0 <= PrefixSize <= 8\n",
377 378 379 380 381
            flagname, value);
    return false;
  }
  return true;
}
I
Igor Canadi 已提交
382
DEFINE_int32(prefix_size, 7, "Control the prefix size for HashSkipListRep");
L
Lei Jin 已提交
383
static const bool FLAGS_prefix_size_dummy __attribute__((unused)) =
384
    RegisterFlagValidator(&FLAGS_prefix_size, &ValidatePrefixSize);
385 386 387

DEFINE_bool(use_merge, false, "On true, replaces all writes with a Merge "
            "that behaves like a Put");
J
Jim Paton 已提交
388

D
Deon Nicholas 已提交
389

390
namespace rocksdb {
391

392 393 394 395 396 397 398 399 400 401 402 403 404
// convert long to a big-endian slice key
static std::string Key(long val) {
  std::string little_endian_key;
  std::string big_endian_key;
  PutFixed64(&little_endian_key, val);
  assert(little_endian_key.size() == sizeof(val));
  big_endian_key.resize(sizeof(val));
  for (int i=0; i<(int)sizeof(val); i++) {
    big_endian_key[i] = little_endian_key[sizeof(val) - 1 - i];
  }
  return big_endian_key;
}

405 406 407 408 409 410 411 412 413 414 415
static std::string StringToHex(const std::string& str) {
  std::string result = "0x";
  char buf[10];
  for (size_t i = 0; i < str.length(); i++) {
    snprintf(buf, 10, "%02X", (unsigned char)str[i]);
    result += buf;
  }
  return result;
}


416 417 418 419 420 421 422 423 424
class StressTest;
namespace {

class Stats {
 private:
  double start_;
  double finish_;
  double seconds_;
  long done_;
425 426
  long gets_;
  long prefixes_;
427
  long writes_;
A
amayank 已提交
428
  long deletes_;
429
  long iterator_size_sums_;
430
  long founds_;
431
  long iterations_;
432
  long errors_;
433 434 435
  int next_report_;
  size_t bytes_;
  double last_op_finish_;
436
  HistogramImpl hist_;
437 438 439 440 441 442 443 444

 public:
  Stats() { }

  void Start() {
    next_report_ = 100;
    hist_.Clear();
    done_ = 0;
445 446
    gets_ = 0;
    prefixes_ = 0;
447
    writes_ = 0;
A
amayank 已提交
448
    deletes_ = 0;
449
    iterator_size_sums_ = 0;
450
    founds_ = 0;
451
    iterations_ = 0;
452
    errors_ = 0;
453 454 455 456 457 458 459 460 461 462
    bytes_ = 0;
    seconds_ = 0;
    start_ = FLAGS_env->NowMicros();
    last_op_finish_ = start_;
    finish_ = start_;
  }

  void Merge(const Stats& other) {
    hist_.Merge(other.hist_);
    done_ += other.done_;
463 464
    gets_ += other.gets_;
    prefixes_ += other.prefixes_;
465
    writes_ += other.writes_;
A
amayank 已提交
466
    deletes_ += other.deletes_;
467
    iterator_size_sums_ += other.iterator_size_sums_;
468
    founds_ += other.founds_;
469
    iterations_ += other.iterations_;
470
    errors_ += other.errors_;
471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487
    bytes_ += other.bytes_;
    seconds_ += other.seconds_;
    if (other.start_ < start_) start_ = other.start_;
    if (other.finish_ > finish_) finish_ = other.finish_;
  }

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

  void FinishedSingleOp() {
    if (FLAGS_histogram) {
      double now = FLAGS_env->NowMicros();
      double micros = now - last_op_finish_;
      hist_.Add(micros);
      if (micros > 20000) {
488
        fprintf(stdout, "long op: %.1f micros%30s\r", micros, "");
489 490 491 492
      }
      last_op_finish_ = now;
    }

493
      done_++;
494
    if (FLAGS_progress_reports) {
495 496 497 498 499 500 501 502 503 504
      if (done_ >= next_report_) {
        if      (next_report_ < 1000)   next_report_ += 100;
        else if (next_report_ < 5000)   next_report_ += 500;
        else if (next_report_ < 10000)  next_report_ += 1000;
        else if (next_report_ < 50000)  next_report_ += 5000;
        else if (next_report_ < 100000) next_report_ += 10000;
        else if (next_report_ < 500000) next_report_ += 50000;
        else                            next_report_ += 100000;
        fprintf(stdout, "... finished %ld ops%30s\r", done_, "");
      }
505 506 507
    }
  }

508 509 510 511 512
  void AddBytesForWrites(int nwrites, size_t nbytes) {
    writes_ += nwrites;
    bytes_ += nbytes;
  }

513 514 515
  void AddGets(int ngets, int nfounds) {
    founds_ += nfounds;
    gets_ += ngets;
516 517
  }

518 519 520 521 522
  void AddPrefixes(int nprefixes, int count) {
    prefixes_ += nprefixes;
    iterator_size_sums_ += count;
  }

523 524 525 526
  void AddIterations(int n) {
    iterations_ += n;
  }

527 528 529 530
  void AddDeletes(int n) {
    deletes_ += n;
  }

531 532
  void AddErrors(int n) {
    errors_ += n;
A
amayank 已提交
533 534
  }

535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551
  void Report(const char* name) {
    std::string extra;
    if (bytes_ < 1 || done_ < 1) {
      fprintf(stderr, "No writes or ops?\n");
      return;
    }

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

    fprintf(stdout, "%-12s: ", name);
    fprintf(stdout, "%.3f micros/op %ld ops/sec\n",
            seconds_ * 1e6 / done_, (long)throughput);
    fprintf(stdout, "%-12s: Wrote %.2f MB (%.2f MB/sec) (%ld%% of %ld ops)\n",
            "", bytes_mb, rate, (100*writes_)/done_, done_);
552
    fprintf(stdout, "%-12s: Wrote %ld times\n", "", writes_);
A
amayank 已提交
553
    fprintf(stdout, "%-12s: Deleted %ld times\n", "", deletes_);
X
Xing Jin 已提交
554 555
    fprintf(stdout, "%-12s: %ld read and %ld found the key\n", "",
            gets_, founds_);
556 557 558
    fprintf(stdout, "%-12s: Prefix scanned %ld times\n", "", prefixes_);
    fprintf(stdout, "%-12s: Iterator size sum is %ld\n", "",
            iterator_size_sums_);
559
    fprintf(stdout, "%-12s: Iterated %ld times\n", "", iterations_);
560
    fprintf(stdout, "%-12s: Got errors %ld times\n", "", errors_);
561 562 563 564 565 566 567 568 569 570 571

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

// State shared by all concurrent executions of the same benchmark.
class SharedState {
 public:
572
  static const uint32_t SENTINEL;
573

574 575 576 577 578 579 580 581 582 583 584 585
  explicit SharedState(StressTest* stress_test)
      : cv_(&mu_),
        seed_(FLAGS_seed),
        max_key_(FLAGS_max_key),
        log2_keys_per_lock_(FLAGS_log2_keys_per_lock),
        num_threads_(FLAGS_threads),
        num_initialized_(0),
        num_populated_(0),
        vote_reopen_(0),
        num_done_(0),
        start_(false),
        start_verify_(false),
586 587
        should_stop_bg_thread_(false),
        bg_thread_finished_(false),
588 589
        stress_test_(stress_test),
        verification_failure_(false) {
590 591 592 593
    if (FLAGS_test_batches_snapshots) {
      fprintf(stdout, "No lock creation because test_batches_snapshots set\n");
      return;
    }
I
Igor Canadi 已提交
594 595 596 597
    values_.resize(FLAGS_column_families);

    for (int i = 0; i < FLAGS_column_families; ++i) {
      values_[i] = std::vector<uint32_t>(max_key_, SENTINEL);
598
    }
599

600 601
    long num_locks = (max_key_ >> log2_keys_per_lock_);
    if (max_key_ & ((1 << log2_keys_per_lock_) - 1)) {
I
Igor Canadi 已提交
602 603 604 605 606 607
      num_locks++;
    }
    fprintf(stdout, "Creating %ld locks\n", num_locks * FLAGS_column_families);
    key_locks_.resize(FLAGS_column_families);
    for (int i = 0; i < FLAGS_column_families; ++i) {
      key_locks_[i] = std::vector<port::Mutex>(num_locks);
608 609 610
    }
  }

I
Igor Canadi 已提交
611
  ~SharedState() {}
612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636

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

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

  StressTest* GetStressTest() const {
    return stress_test_;
  }

  long GetMaxKey() const {
    return max_key_;
  }

  uint32_t GetNumThreads() const {
    return num_threads_;
  }

  void IncInitialized() {
    num_initialized_++;
  }

A
amayank 已提交
637
  void IncOperated() {
638 639 640 641 642 643 644
    num_populated_++;
  }

  void IncDone() {
    num_done_++;
  }

645 646 647 648
  void IncVotedReopen() {
    vote_reopen_ = (vote_reopen_ + 1) % num_threads_;
  }

649 650 651 652
  bool AllInitialized() const {
    return num_initialized_ >= num_threads_;
  }

A
amayank 已提交
653
  bool AllOperated() const {
654 655 656 657 658 659 660
    return num_populated_ >= num_threads_;
  }

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

661 662 663 664
  bool AllVotedReopen() {
    return (vote_reopen_ == 0);
  }

665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680
  void SetStart() {
    start_ = true;
  }

  void SetStartVerify() {
    start_verify_ = true;
  }

  bool Started() const {
    return start_;
  }

  bool VerifyStarted() const {
    return start_verify_;
  }

I
Igor Canadi 已提交
681 682 683 684
  void SetVerificationFailure() { verification_failure_.store(true); }

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

I
Igor Canadi 已提交
685 686
  port::Mutex* GetMutexForKey(int cf, long key) {
    return &key_locks_[cf][key >> log2_keys_per_lock_];
687 688
  }

I
Igor Canadi 已提交
689 690 691 692
  void LockColumnFamily(int cf) {
    for (auto& mutex : key_locks_[cf]) {
      mutex.Lock();
    }
693 694
  }

I
Igor Canadi 已提交
695 696 697 698
  void UnlockColumnFamily(int cf) {
    for (auto& mutex : key_locks_[cf]) {
      mutex.Unlock();
    }
699 700
  }

I
Igor Canadi 已提交
701 702
  void ClearColumnFamily(int cf) {
    std::fill(values_[cf].begin(), values_[cf].end(), SENTINEL);
A
amayank 已提交
703 704
  }

I
Igor Canadi 已提交
705 706
  void Put(int cf, long key, uint32_t value_base) {
    values_[cf][key] = value_base;
707 708
  }

I
Igor Canadi 已提交
709 710 711 712 713 714
  uint32_t Get(int cf, long key) const { return values_[cf][key]; }

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

  uint32_t GetSeed() const { return seed_; }

715 716 717 718 719 720 721 722
  void SetShouldStopBgThread() { should_stop_bg_thread_ = true; }

  bool ShoudStopBgThread() { return should_stop_bg_thread_; }

  void SetBgThreadFinish() { bg_thread_finished_ = true; }

  bool BgThreadFinished() const { return bg_thread_finished_; }

723 724 725 726 727 728 729 730 731
 private:
  port::Mutex mu_;
  port::CondVar cv_;
  const uint32_t seed_;
  const long max_key_;
  const uint32_t log2_keys_per_lock_;
  const int num_threads_;
  long num_initialized_;
  long num_populated_;
732
  long vote_reopen_;
733 734 735
  long num_done_;
  bool start_;
  bool start_verify_;
736 737
  bool should_stop_bg_thread_;
  bool bg_thread_finished_;
738
  StressTest* stress_test_;
I
Igor Canadi 已提交
739
  std::atomic<bool> verification_failure_;
740

I
Igor Canadi 已提交
741 742
  std::vector<std::vector<uint32_t>> values_;
  std::vector<std::vector<port::Mutex>> key_locks_;
743 744
};

745 746
const uint32_t SharedState::SENTINEL = 0xffffffff;

747 748 749 750 751 752 753
// Per-thread state for concurrent executions of the same benchmark.
struct ThreadState {
  uint32_t tid; // 0..n-1
  Random rand;  // Has different seeds for different threads
  SharedState* shared;
  Stats stats;

I
Igor Canadi 已提交
754
  ThreadState(uint32_t index, SharedState* _shared)
I
Igor Canadi 已提交
755
      : tid(index), rand(1000 + index + _shared->GetSeed()), shared(_shared) {}
756 757 758 759 760 761 762 763
};

}  // namespace

class StressTest {
 public:
  StressTest()
      : cache_(NewLRUCache(FLAGS_cache_size)),
I
Igor Canadi 已提交
764 765 766
        compressed_cache_(FLAGS_compressed_cache_size >= 0
                              ? NewLRUCache(FLAGS_compressed_cache_size)
                              : nullptr),
767
        filter_policy_(FLAGS_bloom_bits >= 0
768 769 770 771
                   ? FLAGS_use_block_based_filter
                     ? NewBloomFilterPolicy(FLAGS_bloom_bits, true)
                     : NewBloomFilterPolicy(FLAGS_bloom_bits, false)
                   : nullptr),
772
        db_(nullptr),
773
        new_column_family_name_(1),
774
        num_times_reopened_(0) {
775 776 777 778 779
    if (FLAGS_destroy_db_initially) {
      std::vector<std::string> files;
      FLAGS_env->GetChildren(FLAGS_db, &files);
      for (unsigned int i = 0; i < files.size(); i++) {
        if (Slice(files[i]).starts_with("heap-")) {
780
          FLAGS_env->DeleteFile(FLAGS_db + "/" + files[i]);
781
        }
782
      }
783
      DestroyDB(FLAGS_db, Options());
784 785 786 787
    }
  }

  ~StressTest() {
I
Igor Canadi 已提交
788 789 790 791
    for (auto cf : column_families_) {
      delete cf;
    }
    column_families_.clear();
792 793 794
    delete db_;
  }

L
Lei Jin 已提交
795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914
  bool BuildOptionsTable() {
    if (FLAGS_set_options_one_in <= 0) {
      return true;
    }
    options_table_ = {
      {"write_buffer_size",
        {
          std::to_string(FLAGS_write_buffer_size),
          std::to_string(FLAGS_write_buffer_size * 2),
          std::to_string(FLAGS_write_buffer_size * 4)
        }
      },
      {"max_write_buffer_number",
        {
          std::to_string(FLAGS_max_write_buffer_number),
          std::to_string(FLAGS_max_write_buffer_number * 2),
          std::to_string(FLAGS_max_write_buffer_number * 4)
        }
      },
      {"arena_block_size",
        {
          std::to_string(Options().arena_block_size),
          std::to_string(FLAGS_write_buffer_size / 4),
          std::to_string(FLAGS_write_buffer_size / 8),
        }
      },
      {"memtable_prefix_bloom_bits", {"0", "8", "10"}},
      {"memtable_prefix_bloom_probes", {"4", "5", "6"}},
      {"memtable_prefix_bloom_huge_page_tlb_size",
        {
          "0",
          std::to_string(2 * 1024 * 1024)
        }
      },
      {"max_successive_merges", {"0", "2", "4"}},
      {"filter_deletes", {"0", "1"}},
      {"inplace_update_num_locks", {"100", "200", "300"}},
      // TODO(ljin): enable test for this option
      // {"disable_auto_compactions", {"100", "200", "300"}},
      {"soft_rate_limit", {"0", "0.5", "0.9"}},
      {"hard_rate_limit", {"0", "1.1", "2.0"}},
      {"level0_file_num_compaction_trigger",
        {
          std::to_string(FLAGS_level0_file_num_compaction_trigger),
          std::to_string(FLAGS_level0_file_num_compaction_trigger + 2),
          std::to_string(FLAGS_level0_file_num_compaction_trigger + 4),
        }
      },
      {"level0_slowdown_writes_trigger",
        {
          std::to_string(FLAGS_level0_slowdown_writes_trigger),
          std::to_string(FLAGS_level0_slowdown_writes_trigger + 2),
          std::to_string(FLAGS_level0_slowdown_writes_trigger + 4),
        }
      },
      {"level0_stop_writes_trigger",
        {
          std::to_string(FLAGS_level0_stop_writes_trigger),
          std::to_string(FLAGS_level0_stop_writes_trigger + 2),
          std::to_string(FLAGS_level0_stop_writes_trigger + 4),
        }
      },
      {"max_grandparent_overlap_factor",
        {
          std::to_string(Options().max_grandparent_overlap_factor - 5),
          std::to_string(Options().max_grandparent_overlap_factor),
          std::to_string(Options().max_grandparent_overlap_factor + 5),
        }
      },
      {"expanded_compaction_factor",
        {
          std::to_string(Options().expanded_compaction_factor - 5),
          std::to_string(Options().expanded_compaction_factor),
          std::to_string(Options().expanded_compaction_factor + 5),
        }
      },
      {"source_compaction_factor",
        {
          std::to_string(Options().source_compaction_factor),
          std::to_string(Options().source_compaction_factor * 2),
          std::to_string(Options().source_compaction_factor * 4),
        }
      },
      {"target_file_size_base",
        {
          std::to_string(FLAGS_target_file_size_base),
          std::to_string(FLAGS_target_file_size_base * 2),
          std::to_string(FLAGS_target_file_size_base * 4),
        }
      },
      {"target_file_size_multiplier",
        {
          std::to_string(FLAGS_target_file_size_multiplier),
          "1",
          "2",
        }
      },
      {"max_bytes_for_level_base",
        {
          std::to_string(FLAGS_max_bytes_for_level_base / 2),
          std::to_string(FLAGS_max_bytes_for_level_base),
          std::to_string(FLAGS_max_bytes_for_level_base * 2),
        }
      },
      {"max_bytes_for_level_multiplier",
        {
          std::to_string(FLAGS_max_bytes_for_level_multiplier),
          "1",
          "2",
        }
      },
      {"max_mem_compaction_level", {"0", "1", "2"}},
      {"max_sequential_skip_in_iterations", {"4", "8", "12"}},
    };
    for (const auto& iter : options_table_) {
      options_index_.push_back(iter.first);
    }
    return true;
  }

I
Igor Canadi 已提交
915
  bool Run() {
916
    PrintEnv();
L
Lei Jin 已提交
917
    BuildOptionsTable();
918 919 920 921 922 923 924 925 926
    Open();
    SharedState shared(this);
    uint32_t n = shared.GetNumThreads();

    std::vector<ThreadState*> threads(n);
    for (uint32_t i = 0; i < n; i++) {
      threads[i] = new ThreadState(i, &shared);
      FLAGS_env->StartThread(ThreadBody, threads[i]);
    }
927 928 929 930 931
    ThreadState bg_thread(0, &shared);
    if (FLAGS_compaction_thread_pool_adjust_interval > 0) {
      FLAGS_env->StartThread(PoolSizeChangeThread, &bg_thread);
    }

932
    // Each thread goes through the following states:
A
amayank 已提交
933 934
    // initializing -> wait for others to init -> read/populate/depopulate
    // wait for others to operate -> verify -> done
935 936 937 938 939 940 941

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

942 943 944
      double now = FLAGS_env->NowMicros();
      fprintf(stdout, "%s Starting database operations\n",
              FLAGS_env->TimeToString((uint64_t) now/1000000).c_str());
A
amayank 已提交
945

946 947
      shared.SetStart();
      shared.GetCondVar()->SignalAll();
A
amayank 已提交
948
      while (!shared.AllOperated()) {
949 950 951
        shared.GetCondVar()->Wait();
      }

952
      now = FLAGS_env->NowMicros();
953 954 955 956 957 958 959
      if (FLAGS_test_batches_snapshots) {
        fprintf(stdout, "%s Limited verification already done during gets\n",
                FLAGS_env->TimeToString((uint64_t) now/1000000).c_str());
      } else {
        fprintf(stdout, "%s Starting verification\n",
                FLAGS_env->TimeToString((uint64_t) now/1000000).c_str());
      }
960

961 962 963 964 965 966 967
      shared.SetStartVerify();
      shared.GetCondVar()->SignalAll();
      while (!shared.AllDone()) {
        shared.GetCondVar()->Wait();
      }
    }

968
    for (unsigned int i = 1; i < n; i++) {
969 970 971 972
      threads[0]->stats.Merge(threads[i]->stats);
    }
    threads[0]->stats.Report("Stress Test");

973
    for (unsigned int i = 0; i < n; i++) {
974
      delete threads[i];
975
      threads[i] = nullptr;
976
    }
977
    double now = FLAGS_env->NowMicros();
978 979 980 981
    if (!FLAGS_test_batches_snapshots) {
      fprintf(stdout, "%s Verification successful\n",
              FLAGS_env->TimeToString((uint64_t) now/1000000).c_str());
    }
982
    PrintStatistics();
I
Igor Canadi 已提交
983

984 985 986 987 988 989 990 991
    if (FLAGS_compaction_thread_pool_adjust_interval > 0) {
      MutexLock l(shared.GetMutex());
      shared.SetShouldStopBgThread();
      while (!shared.BgThreadFinished()) {
        shared.GetCondVar()->Wait();
      }
    }

I
Igor Canadi 已提交
992 993 994 995 996
    if (shared.HasVerificationFailedYet()) {
      printf("Verification failed :(\n");
      return false;
    }
    return true;
997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014
  }

 private:

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

    {
      MutexLock l(shared->GetMutex());
      shared->IncInitialized();
      if (shared->AllInitialized()) {
        shared->GetCondVar()->SignalAll();
      }
      while (!shared->Started()) {
        shared->GetCondVar()->Wait();
      }
    }
A
amayank 已提交
1015
    thread->shared->GetStressTest()->OperateDb(thread);
1016 1017 1018

    {
      MutexLock l(shared->GetMutex());
A
amayank 已提交
1019 1020
      shared->IncOperated();
      if (shared->AllOperated()) {
1021 1022 1023 1024 1025 1026 1027
        shared->GetCondVar()->SignalAll();
      }
      while (!shared->VerifyStarted()) {
        shared->GetCondVar()->Wait();
      }
    }

1028
    if (!FLAGS_test_batches_snapshots) {
M
Mayank Agarwal 已提交
1029
      thread->shared->GetStressTest()->VerifyDb(thread);
1030
    }
1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041

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

  }

1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073
  static void PoolSizeChangeThread(void* v) {
    assert(FLAGS_compaction_thread_pool_adjust_interval > 0);
    ThreadState* thread = reinterpret_cast<ThreadState*>(v);
    SharedState* shared = thread->shared;

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

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

1074 1075 1076
  // Given a key K and value V, this puts ("0"+K, "0"+V), ("1"+K, "1"+V), ...
  // ("9"+K, "9"+V) in DB atomically i.e in a single batch.
  // Also refer MultiGet.
I
Igor Canadi 已提交
1077 1078 1079
  Status MultiPut(ThreadState* thread, const WriteOptions& writeoptions,
                  ColumnFamilyHandle* column_family, const Slice& key,
                  const Slice& value, size_t sz) {
1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090
    std::string keys[10] = {"9", "8", "7", "6", "5",
                            "4", "3", "2", "1", "0"};
    std::string values[10] = {"9", "8", "7", "6", "5",
                              "4", "3", "2", "1", "0"};
    Slice value_slices[10];
    WriteBatch batch;
    Status s;
    for (int i = 0; i < 10; i++) {
      keys[i] += key.ToString();
      values[i] += value.ToString();
      value_slices[i] = values[i];
1091
      if (FLAGS_use_merge) {
I
Igor Canadi 已提交
1092
        batch.Merge(column_family, keys[i], value_slices[i]);
D
Deon Nicholas 已提交
1093
      } else {
I
Igor Canadi 已提交
1094
        batch.Put(column_family, keys[i], value_slices[i]);
D
Deon Nicholas 已提交
1095
      }
1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111
    }

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

    return s;
  }

  // Given a key K, this deletes ("0"+K), ("1"+K),... ("9"+K)
  // in DB atomically i.e in a single batch. Also refer MultiGet.
I
Igor Canadi 已提交
1112 1113
  Status MultiDelete(ThreadState* thread, const WriteOptions& writeoptions,
                     ColumnFamilyHandle* column_family, const Slice& key) {
1114 1115 1116 1117 1118 1119 1120
    std::string keys[10] = {"9", "7", "5", "3", "1",
                            "8", "6", "4", "2", "0"};

    WriteBatch batch;
    Status s;
    for (int i = 0; i < 10; i++) {
      keys[i] += key.ToString();
I
Igor Canadi 已提交
1121
      batch.Delete(column_family, keys[i]);
1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138
    }

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

    return s;
  }

  // Given a key K, this gets values for "0"+K, "1"+K,..."9"+K
  // in the same snapshot, and verifies that all the values are of the form
  // "0"+V, "1"+V,..."9"+V.
  // ASSUMES that MultiPut was used to put (K, V) into the DB.
I
Igor Canadi 已提交
1139 1140 1141
  Status MultiGet(ThreadState* thread, const ReadOptions& readoptions,
                  ColumnFamilyHandle* column_family, const Slice& key,
                  std::string* value) {
1142 1143 1144 1145 1146 1147 1148 1149 1150
    std::string keys[10] = {"0", "1", "2", "3", "4", "5", "6", "7", "8", "9"};
    Slice key_slices[10];
    std::string values[10];
    ReadOptions readoptionscopy = readoptions;
    readoptionscopy.snapshot = db_->GetSnapshot();
    Status s;
    for (int i = 0; i < 10; i++) {
      keys[i] += key.ToString();
      key_slices[i] = keys[i];
I
Igor Canadi 已提交
1151
      s = db_->Get(readoptionscopy, column_family, key_slices[i], value);
1152 1153 1154 1155 1156 1157 1158 1159
      if (!s.ok() && !s.IsNotFound()) {
        fprintf(stderr, "get error: %s\n", s.ToString().c_str());
        values[i] = "";
        thread->stats.AddErrors(1);
        // we continue after error rather than exiting so that we can
        // find more errors if any
      } else if (s.IsNotFound()) {
        values[i] = "";
1160
        thread->stats.AddGets(1, 0);
1161 1162 1163 1164 1165 1166
      } else {
        values[i] = *value;

        char expected_prefix = (keys[i])[0];
        char actual_prefix = (values[i])[0];
        if (actual_prefix != expected_prefix) {
1167
          fprintf(stderr, "error expected prefix = %c actual = %c\n",
1168 1169 1170
                  expected_prefix, actual_prefix);
        }
        (values[i])[0] = ' '; // blank out the differing character
1171
        thread->stats.AddGets(1, 1);
1172 1173 1174 1175 1176 1177 1178
      }
    }
    db_->ReleaseSnapshot(readoptionscopy.snapshot);

    // Now that we retrieved all values, check that they all match
    for (int i = 1; i < 10; i++) {
      if (values[i] != values[0]) {
1179
        fprintf(stderr, "error : inconsistent values for key %s: %s, %s\n",
1180 1181
                key.ToString(true).c_str(), StringToHex(values[0]).c_str(),
                StringToHex(values[i]).c_str());
1182 1183 1184 1185 1186 1187 1188 1189
      // we continue after error rather than exiting so that we can
      // find more errors if any
      }
    }

    return s;
  }

L
Lei Jin 已提交
1190 1191 1192 1193 1194
  // Given a key, this does prefix scans for "0"+P, "1"+P,..."9"+P
  // in the same snapshot where P is the first FLAGS_prefix_size - 1 bytes
  // of the key. Each of these 10 scans returns a series of values;
  // each series should be the same length, and it is verified for each
  // index i that all the i'th values are of the form "0"+V, "1"+V,..."9"+V.
1195
  // ASSUMES that MultiPut was used to put (K, V)
I
Igor Canadi 已提交
1196 1197
  Status MultiPrefixScan(ThreadState* thread, const ReadOptions& readoptions,
                         ColumnFamilyHandle* column_family,
L
Lei Jin 已提交
1198
                         const Slice& key) {
1199 1200 1201 1202 1203 1204 1205 1206
    std::string prefixes[10] = {"0", "1", "2", "3", "4",
                                "5", "6", "7", "8", "9"};
    Slice prefix_slices[10];
    ReadOptions readoptionscopy[10];
    const Snapshot* snapshot = db_->GetSnapshot();
    Iterator* iters[10];
    Status s = Status::OK();
    for (int i = 0; i < 10; i++) {
L
Lei Jin 已提交
1207 1208 1209
      prefixes[i] += key.ToString();
      prefixes[i].resize(FLAGS_prefix_size);
      prefix_slices[i] = Slice(prefixes[i]);
1210 1211
      readoptionscopy[i] = readoptions;
      readoptionscopy[i].snapshot = snapshot;
I
Igor Canadi 已提交
1212
      iters[i] = db_->NewIterator(readoptionscopy[i], column_family);
I
Igor Canadi 已提交
1213
      iters[i]->Seek(prefix_slices[i]);
1214 1215 1216
    }

    int count = 0;
1217
    while (iters[0]->Valid() && iters[0]->key().starts_with(prefix_slices[0])) {
1218 1219 1220 1221 1222
      count++;
      std::string values[10];
      // get list of all values for this iteration
      for (int i = 0; i < 10; i++) {
        // no iterator should finish before the first one
1223 1224
        assert(iters[i]->Valid() &&
               iters[i]->key().starts_with(prefix_slices[i]));
1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238
        values[i] = iters[i]->value().ToString();

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

        if (actual_first != expected_first) {
          fprintf(stderr, "error expected first = %c actual = %c\n",
                  expected_first, actual_first);
        }
        (values[i])[0] = ' '; // blank out the differing character
      }
      // make sure all values are equivalent
      for (int i = 0; i < 10; i++) {
        if (values[i] != values[0]) {
1239 1240 1241
          fprintf(stderr, "error : %d, inconsistent values for prefix %s: %s, %s\n",
                  i, prefixes[i].c_str(), StringToHex(values[0]).c_str(),
                  StringToHex(values[i]).c_str());
1242 1243 1244 1245 1246 1247 1248 1249 1250 1251
          // we continue after error rather than exiting so that we can
          // find more errors if any
        }
        iters[i]->Next();
      }
    }

    // cleanup iterators and snapshot
    for (int i = 0; i < 10; i++) {
      // if the first iterator finished, they should have all finished
1252 1253
      assert(!iters[i]->Valid() ||
             !iters[i]->key().starts_with(prefix_slices[i]));
1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267
      assert(iters[i]->status().ok());
      delete iters[i];
    }
    db_->ReleaseSnapshot(snapshot);

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

    return s;
  }

1268 1269
  // Given a key K, this creates an iterator which scans to K and then
  // does a random sequence of Next/Prev operations.
I
Igor Canadi 已提交
1270 1271
  Status MultiIterate(ThreadState* thread, const ReadOptions& readoptions,
                      ColumnFamilyHandle* column_family, const Slice& key) {
1272 1273 1274 1275
    Status s;
    const Snapshot* snapshot = db_->GetSnapshot();
    ReadOptions readoptionscopy = readoptions;
    readoptionscopy.snapshot = snapshot;
I
Igor Canadi 已提交
1276
    unique_ptr<Iterator> iter(db_->NewIterator(readoptionscopy, column_family));
1277 1278

    iter->Seek(key);
1279
    for (uint64_t i = 0; i < FLAGS_num_iterations && iter->Valid(); i++) {
1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297
      if (thread->rand.OneIn(2)) {
        iter->Next();
      } else {
        iter->Prev();
      }
    }

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

    db_->ReleaseSnapshot(snapshot);

    return s;
  }

1298
  Status SetOptions(ThreadState* thread) {
L
Lei Jin 已提交
1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324
    assert(FLAGS_set_options_one_in > 0);
    std::unordered_map<std::string, std::string> opts;
    std::string name = options_index_[
      thread->rand.Next() % options_index_.size()];
    int value_idx = thread->rand.Next() % options_table_[name].size();
    if (name == "soft_rate_limit" || name == "hard_rate_limit") {
      opts["soft_rate_limit"] = options_table_["soft_rate_limit"][value_idx];
      opts["hard_rate_limit"] = options_table_["hard_rate_limit"][value_idx];
    } else if (name == "level0_file_num_compaction_trigger" ||
               name == "level0_slowdown_writes_trigger" ||
               name == "level0_stop_writes_trigger") {
      opts["level0_file_num_compaction_trigger"] =
        options_table_["level0_file_num_compaction_trigger"][value_idx];
      opts["level0_slowdown_writes_trigger"] =
        options_table_["level0_slowdown_writes_trigger"][value_idx];
      opts["level0_stop_writes_trigger"] =
        options_table_["level0_stop_writes_trigger"][value_idx];
    } else {
      opts[name] = options_table_[name][value_idx];
    }

    int rand_cf_idx = thread->rand.Next() % FLAGS_column_families;
    auto cfh = column_families_[rand_cf_idx];
    return db_->SetOptions(cfh, opts);
  }

A
amayank 已提交
1325
  void OperateDb(ThreadState* thread) {
1326 1327
    ReadOptions read_opts(FLAGS_verify_checksum, true);
    WriteOptions write_opts;
A
amayank 已提交
1328
    char value[100];
1329 1330 1331 1332 1333 1334
    long max_key = thread->shared->GetMaxKey();
    std::string from_db;
    if (FLAGS_sync) {
      write_opts.sync = true;
    }
    write_opts.disableWAL = FLAGS_disable_wal;
1335 1336 1337
    const int prefixBound = (int)FLAGS_readpercent + (int)FLAGS_prefixpercent;
    const int writeBound = prefixBound + (int)FLAGS_writepercent;
    const int delBound = writeBound + (int)FLAGS_delpercent;
1338 1339

    thread->stats.Start();
1340
    for (uint64_t i = 0; i < FLAGS_ops_per_thread; i++) {
I
Igor Canadi 已提交
1341 1342 1343 1344
      if (thread->shared->HasVerificationFailedYet()) {
        break;
      }
      if (i != 0 && (i % (FLAGS_ops_per_thread / (FLAGS_reopen + 1))) == 0) {
1345
        {
1346
          thread->stats.FinishedSingleOp();
1347 1348 1349 1350 1351 1352 1353 1354 1355
          MutexLock l(thread->shared->GetMutex());
          thread->shared->IncVotedReopen();
          if (thread->shared->AllVotedReopen()) {
            thread->shared->GetStressTest()->Reopen();
            thread->shared->GetCondVar()->SignalAll();
          }
          else {
            thread->shared->GetCondVar()->Wait();
          }
1356 1357
          // Commenting this out as we don't want to reset stats on each open.
          // thread->stats.Start();
1358 1359
        }
      }
1360

L
Lei Jin 已提交
1361 1362 1363 1364 1365 1366
      // Change Options
      if (FLAGS_set_options_one_in > 0 &&
          thread->rand.OneIn(FLAGS_set_options_one_in)) {
        SetOptions(thread);
      }

1367 1368 1369 1370 1371
      if (FLAGS_set_in_place_one_in > 0 &&
          thread->rand.OneIn(FLAGS_set_in_place_one_in)) {
        options_.inplace_update_support ^= options_.inplace_update_support;
      }

I
Igor Canadi 已提交
1372
      if (!FLAGS_test_batches_snapshots &&
1373
          FLAGS_clear_column_family_one_in != 0 && FLAGS_column_families > 1) {
I
Igor Canadi 已提交
1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389
        if (thread->rand.OneIn(FLAGS_clear_column_family_one_in)) {
          // drop column family and then create it again (can't drop default)
          int cf = thread->rand.Next() % (FLAGS_column_families - 1) + 1;
          std::string new_name =
              std::to_string(new_column_family_name_.fetch_add(1));
          {
            MutexLock l(thread->shared->GetMutex());
            fprintf(
                stdout,
                "[CF %d] Dropping and recreating column family. new name: %s\n",
                cf, new_name.c_str());
          }
          thread->shared->LockColumnFamily(cf);
          Status s __attribute__((unused));
          s = db_->DropColumnFamily(column_families_[cf]);
          delete column_families_[cf];
1390 1391 1392 1393 1394
          if (!s.ok()) {
            fprintf(stderr, "dropping column family error: %s\n",
                s.ToString().c_str());
            std::terminate();
          }
I
Igor Canadi 已提交
1395 1396 1397 1398
          s = db_->CreateColumnFamily(ColumnFamilyOptions(options_), new_name,
                                      &column_families_[cf]);
          column_family_names_[cf] = new_name;
          thread->shared->ClearColumnFamily(cf);
1399 1400 1401 1402 1403
          if (!s.ok()) {
            fprintf(stderr, "creating column family error: %s\n",
                s.ToString().c_str());
            std::terminate();
          }
I
Igor Canadi 已提交
1404 1405 1406 1407
          thread->shared->UnlockColumnFamily(cf);
        }
      }

1408
      long rand_key = thread->rand.Next() % max_key;
I
Igor Canadi 已提交
1409
      int rand_column_family = thread->rand.Next() % FLAGS_column_families;
1410 1411 1412
      std::string keystr = Key(rand_key);
      Slice key = keystr;
      int prob_op = thread->rand.Uniform(100);
I
Igor Canadi 已提交
1413 1414 1415 1416 1417 1418
      std::unique_ptr<MutexLock> l;
      if (!FLAGS_test_batches_snapshots) {
        l.reset(new MutexLock(
            thread->shared->GetMutexForKey(rand_column_family, rand_key)));
      }
      auto column_family = column_families_[rand_column_family];
1419 1420

      if (prob_op >= 0 && prob_op < (int)FLAGS_readpercent) {
1421
        // OPERATION read
1422
        if (!FLAGS_test_batches_snapshots) {
I
Igor Canadi 已提交
1423
          Status s = db_->Get(read_opts, column_family, key, &from_db);
1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434
          if (s.ok()) {
            // found case
            thread->stats.AddGets(1, 1);
          } else if (s.IsNotFound()) {
            // not found case
            thread->stats.AddGets(1, 0);
          } else {
            // errors case
            thread->stats.AddErrors(1);
          }
        } else {
I
Igor Canadi 已提交
1435
          MultiGet(thread, read_opts, column_family, key, &from_db);
1436
        }
1437 1438
      } else if ((int)FLAGS_readpercent <= prob_op && prob_op < prefixBound) {
        // OPERATION prefix scan
L
Lei Jin 已提交
1439 1440
        // keys are 8 bytes long, prefix size is FLAGS_prefix_size. There are
        // (8 - FLAGS_prefix_size) bytes besides the prefix. So there will
I
Igor Canadi 已提交
1441 1442
        // be 2 ^ ((8 - FLAGS_prefix_size) * 8) possible keys with the same
        // prefix
1443
        if (!FLAGS_test_batches_snapshots) {
L
Lei Jin 已提交
1444
          Slice prefix = Slice(key.data(), FLAGS_prefix_size);
I
Igor Canadi 已提交
1445
          Iterator* iter = db_->NewIterator(read_opts, column_family);
L
Lei Jin 已提交
1446
          int64_t count = 0;
I
Igor Canadi 已提交
1447 1448
          for (iter->Seek(prefix);
               iter->Valid() && iter->key().starts_with(prefix); iter->Next()) {
L
Lei Jin 已提交
1449
            ++count;
1450
          }
I
Igor Canadi 已提交
1451 1452
          assert(count <=
                 (static_cast<int64_t>(1) << ((8 - FLAGS_prefix_size) * 8)));
1453 1454 1455 1456 1457 1458
          if (iter->status().ok()) {
            thread->stats.AddPrefixes(1, count);
          } else {
            thread->stats.AddErrors(1);
          }
          delete iter;
1459
        } else {
1460
          MultiPrefixScan(thread, read_opts, column_family, key);
A
amayank 已提交
1461
        }
1462 1463
      } else if (prefixBound <= prob_op && prob_op < writeBound) {
        // OPERATION write
1464 1465 1466
        uint32_t value_base = thread->rand.Next();
        size_t sz = GenerateValue(value_base, value, sizeof(value));
        Slice v(value, sz);
1467
        if (!FLAGS_test_batches_snapshots) {
1468
          if (FLAGS_verify_before_write) {
I
Igor Canadi 已提交
1469 1470
            std::string keystr2 = Key(rand_key);
            Slice k = keystr2;
I
Igor Canadi 已提交
1471
            Status s = db_->Get(read_opts, column_family, k, &from_db);
I
Igor Canadi 已提交
1472 1473 1474 1475
            if (VerifyValue(rand_column_family, rand_key, read_opts,
                            thread->shared, from_db, s, true) == false) {
              break;
            }
1476
          }
I
Igor Canadi 已提交
1477
          thread->shared->Put(rand_column_family, rand_key, value_base);
1478
          Status s;
1479
          if (FLAGS_use_merge) {
1480
            s = db_->Merge(write_opts, column_family, key, v);
D
Deon Nicholas 已提交
1481
          } else {
1482 1483 1484 1485 1486
            s = db_->Put(write_opts, column_family, key, v);
          }
          if (!s.ok()) {
            fprintf(stderr, "put or merge error: %s\n", s.ToString().c_str());
            std::terminate();
D
Deon Nicholas 已提交
1487
          }
1488 1489
          thread->stats.AddBytesForWrites(1, sz);
        } else {
I
Igor Canadi 已提交
1490
          MultiPut(thread, write_opts, column_family, key, v, sz);
1491
        }
I
Igor Canadi 已提交
1492
        PrintKeyValue(rand_column_family, rand_key, value, sz);
1493 1494
      } else if (writeBound <= prob_op && prob_op < delBound) {
        // OPERATION delete
1495
        if (!FLAGS_test_batches_snapshots) {
I
Igor Canadi 已提交
1496
          thread->shared->Delete(rand_column_family, rand_key);
1497
          Status s = db_->Delete(write_opts, column_family, key);
1498
          thread->stats.AddDeletes(1);
1499 1500 1501 1502
          if (!s.ok()) {
            fprintf(stderr, "delete error: %s\n", s.ToString().c_str());
            std::terminate();
          }
1503
        } else {
I
Igor Canadi 已提交
1504
          MultiDelete(thread, write_opts, column_family, key);
1505
        }
1506 1507
      } else {
        // OPERATION iterate
I
Igor Canadi 已提交
1508
        MultiIterate(thread, read_opts, column_family, key);
1509
      }
1510 1511
      thread->stats.FinishedSingleOp();
    }
1512

1513 1514 1515
    thread->stats.Stop();
  }

M
Mayank Agarwal 已提交
1516
  void VerifyDb(ThreadState* thread) const {
1517
    ReadOptions options(FLAGS_verify_checksum, true);
I
Igor Canadi 已提交
1518 1519 1520
    auto shared = thread->shared;
    static const long max_key = shared->GetMaxKey();
    static const long keys_per_thread = max_key / shared->GetNumThreads();
M
Mayank Agarwal 已提交
1521
    long start = keys_per_thread * thread->tid;
1522
    long end = start + keys_per_thread;
I
Igor Canadi 已提交
1523
    if (thread->tid == shared->GetNumThreads() - 1) {
1524 1525
      end = max_key;
    }
I
Igor Canadi 已提交
1526
    for (size_t cf = 0; cf < column_families_.size(); ++cf) {
I
Igor Canadi 已提交
1527 1528 1529
      if (thread->shared->HasVerificationFailedYet()) {
        break;
      }
I
Igor Canadi 已提交
1530 1531 1532 1533 1534 1535
      if (!thread->rand.OneIn(2)) {
        // Use iterator to verify this range
        unique_ptr<Iterator> iter(
            db_->NewIterator(options, column_families_[cf]));
        iter->Seek(Key(start));
        for (long i = start; i < end; i++) {
I
Igor Canadi 已提交
1536 1537 1538
          if (thread->shared->HasVerificationFailedYet()) {
            break;
          }
1539 1540 1541 1542 1543 1544
          // TODO(ljin): update "long" to uint64_t
          // Reseek when the prefix changes
          if (i % (static_cast<int64_t>(1) << 8 * (8 - FLAGS_prefix_size)) ==
              0) {
            iter->Seek(Key(i));
          }
I
Igor Canadi 已提交
1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555
          std::string from_db;
          std::string keystr = Key(i);
          Slice k = keystr;
          Status s = iter->status();
          if (iter->Valid()) {
            if (iter->key().compare(k) > 0) {
              s = Status::NotFound(Slice());
            } else if (iter->key().compare(k) == 0) {
              from_db = iter->value().ToString();
              iter->Next();
            } else if (iter->key().compare(k) < 0) {
I
Igor Canadi 已提交
1556
              VerificationAbort(shared, "An out of range key was found", cf, i);
I
Igor Canadi 已提交
1557 1558 1559 1560
            }
          } else {
            // The iterator found no value for the key in question, so do not
            // move to the next item in the iterator
M
Mayank Agarwal 已提交
1561 1562
            s = Status::NotFound(Slice());
          }
I
Igor Canadi 已提交
1563 1564 1565 1566
          VerifyValue(cf, i, options, shared, from_db, s, true);
          if (from_db.length()) {
            PrintKeyValue(cf, i, from_db.data(), from_db.length());
          }
M
Mayank Agarwal 已提交
1567
        }
I
Igor Canadi 已提交
1568 1569 1570
      } else {
        // Use Get to verify this range
        for (long i = start; i < end; i++) {
I
Igor Canadi 已提交
1571 1572 1573
          if (thread->shared->HasVerificationFailedYet()) {
            break;
          }
I
Igor Canadi 已提交
1574 1575 1576 1577
          std::string from_db;
          std::string keystr = Key(i);
          Slice k = keystr;
          Status s = db_->Get(options, column_families_[cf], k, &from_db);
1578
          VerifyValue(cf, i, options, shared, from_db, s, true);
I
Igor Canadi 已提交
1579 1580 1581
          if (from_db.length()) {
            PrintKeyValue(cf, i, from_db.data(), from_db.length());
          }
M
Mayank Agarwal 已提交
1582
        }
1583 1584 1585 1586
      }
    }
  }

I
Igor Canadi 已提交
1587 1588 1589 1590 1591
  void VerificationAbort(SharedState* shared, std::string msg, int cf,
                         long key) const {
    printf("Verification failed for column family %d key %ld: %s\n", cf, key,
           msg.c_str());
    shared->SetVerificationFailure();
1592 1593
  }

I
Igor Canadi 已提交
1594 1595
  bool VerifyValue(int cf, long key, const ReadOptions& opts,
                   SharedState* shared, const std::string& value_from_db,
I
Igor Canadi 已提交
1596
                   Status s, bool strict = false) const {
I
Igor Canadi 已提交
1597 1598 1599
    if (shared->HasVerificationFailedYet()) {
      return false;
    }
1600
    // compare value_from_db with the value in the shared state
A
amayank 已提交
1601
    char value[100];
I
Igor Canadi 已提交
1602
    uint32_t value_base = shared->Get(cf, key);
1603
    if (value_base == SharedState::SENTINEL && !strict) {
I
Igor Canadi 已提交
1604
      return true;
1605 1606
    }

1607
    if (s.ok()) {
1608
      if (value_base == SharedState::SENTINEL) {
I
Igor Canadi 已提交
1609 1610
        VerificationAbort(shared, "Unexpected value found", cf, key);
        return false;
1611
      }
A
amayank 已提交
1612
      size_t sz = GenerateValue(value_base, value, sizeof(value));
1613
      if (value_from_db.length() != sz) {
I
Igor Canadi 已提交
1614 1615
        VerificationAbort(shared, "Length of value read is not equal", cf, key);
        return false;
1616
      }
1617
      if (memcmp(value_from_db.data(), value, sz) != 0) {
I
Igor Canadi 已提交
1618 1619 1620
        VerificationAbort(shared, "Contents of value read don't match", cf,
                          key);
        return false;
1621 1622 1623
      }
    } else {
      if (value_base != SharedState::SENTINEL) {
I
Igor Canadi 已提交
1624 1625
        VerificationAbort(shared, "Value not found: " + s.ToString(), cf, key);
        return false;
1626 1627
      }
    }
I
Igor Canadi 已提交
1628
    return true;
1629 1630
  }

I
Igor Canadi 已提交
1631 1632 1633 1634 1635 1636 1637
  static void PrintKeyValue(int cf, uint32_t key, const char* value,
                            size_t sz) {
    if (!FLAGS_verbose) {
      return;
    }
    fprintf(stdout, "[CF %d] %u ==> (%u) ", cf, key, (unsigned int)sz);
    for (size_t i = 0; i < sz; i++) {
1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649
      fprintf(stdout, "%X", value[i]);
    }
    fprintf(stdout, "\n");
  }

  static size_t GenerateValue(uint32_t rand, char *v, size_t max_sz) {
    size_t value_sz = ((rand % 3) + 1) * FLAGS_value_size_mult;
    assert(value_sz <= max_sz && value_sz >= sizeof(uint32_t));
    *((uint32_t*)v) = rand;
    for (size_t i=sizeof(uint32_t); i < value_sz; i++) {
      v[i] = (char)(rand ^ i);
    }
1650
    v[value_sz] = '\0';
1651 1652 1653 1654
    return value_sz; // the size of the value set.
  }

  void PrintEnv() const {
I
Igor Canadi 已提交
1655 1656 1657 1658 1659 1660 1661
    fprintf(stdout, "RocksDB version     : %d.%d\n", kMajorVersion,
            kMinorVersion);
    fprintf(stdout, "Column families     : %d\n", FLAGS_column_families);
    if (!FLAGS_test_batches_snapshots) {
      fprintf(stdout, "Clear CFs one in    : %d\n",
              FLAGS_clear_column_family_one_in);
    }
1662
    fprintf(stdout, "Number of threads   : %d\n", FLAGS_threads);
K
kailiu 已提交
1663 1664 1665
    fprintf(stdout,
            "Ops per thread      : %lu\n",
            (unsigned long)FLAGS_ops_per_thread);
1666 1667 1668 1669 1670
    std::string ttl_state("unused");
    if (FLAGS_ttl > 0) {
      ttl_state = NumberToString(FLAGS_ttl);
    }
    fprintf(stdout, "Time to live(sec)   : %s\n", ttl_state.c_str());
K
Kai Liu 已提交
1671 1672 1673 1674 1675
    fprintf(stdout, "Read percentage     : %d%%\n", FLAGS_readpercent);
    fprintf(stdout, "Prefix percentage   : %d%%\n", FLAGS_prefixpercent);
    fprintf(stdout, "Write percentage    : %d%%\n", FLAGS_writepercent);
    fprintf(stdout, "Delete percentage   : %d%%\n", FLAGS_delpercent);
    fprintf(stdout, "Iterate percentage  : %d%%\n", FLAGS_iterpercent);
1676
    fprintf(stdout, "Write-buffer-size   : %d\n", FLAGS_write_buffer_size);
K
kailiu 已提交
1677 1678 1679 1680 1681 1682
    fprintf(stdout,
            "Iterations          : %lu\n",
            (unsigned long)FLAGS_num_iterations);
    fprintf(stdout,
            "Max key             : %lu\n",
            (unsigned long)FLAGS_max_key);
X
Xing Jin 已提交
1683 1684
    fprintf(stdout, "Ratio #ops/#keys    : %f\n",
            (1.0 * FLAGS_ops_per_thread * FLAGS_threads)/FLAGS_max_key);
1685
    fprintf(stdout, "Num times DB reopens: %d\n", FLAGS_reopen);
1686 1687
    fprintf(stdout, "Batches/snapshots   : %d\n",
            FLAGS_test_batches_snapshots);
X
Xing Jin 已提交
1688
    fprintf(stdout, "Purge redundant %%   : %d\n",
1689
            FLAGS_purge_redundant_percent);
1690
    fprintf(stdout, "Deletes use filter  : %d\n",
1691
            FLAGS_filter_deletes);
1692 1693
    fprintf(stdout, "Do update in place  : %d\n",
            FLAGS_in_place_update);
1694
    fprintf(stdout, "Num keys per lock   : %d\n",
1695 1696
            1 << FLAGS_log2_keys_per_lock);

1697
    const char* compression = "";
1698
    switch (FLAGS_compression_type_e) {
1699
      case rocksdb::kNoCompression:
1700
        compression = "none";
1701
        break;
1702
      case rocksdb::kSnappyCompression:
1703
        compression = "snappy";
1704
        break;
1705
      case rocksdb::kZlibCompression:
1706
        compression = "zlib";
1707
        break;
1708
      case rocksdb::kBZip2Compression:
1709
        compression = "bzip2";
1710
        break;
A
Albert Strasheim 已提交
1711 1712 1713 1714 1715 1716
      case rocksdb::kLZ4Compression:
        compression = "lz4";
      case rocksdb::kLZ4HCCompression:
        compression = "lz4hc";
        break;
      }
1717 1718

    fprintf(stdout, "Compression         : %s\n", compression);
J
Jim Paton 已提交
1719 1720 1721 1722 1723 1724

    const char* memtablerep = "";
    switch (FLAGS_rep_factory) {
      case kSkipList:
        memtablerep = "skip_list";
        break;
I
Igor Canadi 已提交
1725
      case kHashSkipList:
J
Jim Paton 已提交
1726 1727 1728 1729 1730 1731 1732 1733 1734
        memtablerep = "prefix_hash";
        break;
      case kVectorRep:
        memtablerep = "vector";
        break;
    }

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

1735 1736 1737 1738
    fprintf(stdout, "------------------------------------------------\n");
  }

  void Open() {
1739
    assert(db_ == nullptr);
1740 1741 1742 1743 1744 1745 1746
    BlockBasedTableOptions block_based_options;
    block_based_options.block_cache = cache_;
    block_based_options.block_cache_compressed = compressed_cache_;
    block_based_options.block_size = FLAGS_block_size;
    block_based_options.filter_policy = filter_policy_;
    options_.table_factory.reset(
        NewBlockBasedTableFactory(block_based_options));
I
Igor Canadi 已提交
1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761
    options_.write_buffer_size = FLAGS_write_buffer_size;
    options_.max_write_buffer_number = FLAGS_max_write_buffer_number;
    options_.min_write_buffer_number_to_merge =
        FLAGS_min_write_buffer_number_to_merge;
    options_.max_background_compactions = FLAGS_max_background_compactions;
    options_.max_background_flushes = FLAGS_max_background_flushes;
    options_.compaction_style =
        static_cast<rocksdb::CompactionStyle>(FLAGS_compaction_style);
    options_.prefix_extractor.reset(NewFixedPrefixTransform(FLAGS_prefix_size));
    options_.max_open_files = FLAGS_open_files;
    options_.statistics = dbstats;
    options_.env = FLAGS_env;
    options_.disableDataSync = FLAGS_disable_data_sync;
    options_.use_fsync = FLAGS_use_fsync;
    options_.allow_mmap_reads = FLAGS_mmap_read;
1762
    rocksdb_kill_odds = FLAGS_kill_random_test;
I
Igor Canadi 已提交
1763 1764 1765 1766
    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 =
1767
        FLAGS_max_bytes_for_level_multiplier;
I
Igor Canadi 已提交
1768 1769 1770 1771 1772 1773 1774 1775 1776
    options_.level0_stop_writes_trigger = FLAGS_level0_stop_writes_trigger;
    options_.level0_slowdown_writes_trigger =
        FLAGS_level0_slowdown_writes_trigger;
    options_.level0_file_num_compaction_trigger =
        FLAGS_level0_file_num_compaction_trigger;
    options_.compression = FLAGS_compression_type_e;
    options_.create_if_missing = true;
    options_.max_manifest_file_size = 10 * 1024;
    options_.filter_deletes = FLAGS_filter_deletes;
1777
    options_.inplace_update_support = FLAGS_in_place_update;
I
Igor Canadi 已提交
1778
    if ((FLAGS_prefix_size == 0) == (FLAGS_rep_factory == kHashSkipList)) {
J
Jim Paton 已提交
1779 1780 1781 1782 1783
      fprintf(stderr,
            "prefix_size should be non-zero iff memtablerep == prefix_hash\n");
      exit(1);
    }
    switch (FLAGS_rep_factory) {
I
Igor Canadi 已提交
1784
      case kHashSkipList:
1785
        options_.memtable_factory.reset(NewHashSkipListRepFactory(10000));
J
Jim Paton 已提交
1786 1787 1788 1789 1790
        break;
      case kSkipList:
        // no need to do anything
        break;
      case kVectorRep:
I
Igor Canadi 已提交
1791
        options_.memtable_factory.reset(new VectorRepFactory());
J
Jim Paton 已提交
1792 1793
        break;
    }
1794
    static Random purge_percent(1000); // no benefit from non-determinism here
1795 1796
    if (static_cast<int32_t>(purge_percent.Uniform(100)) <
        FLAGS_purge_redundant_percent - 1) {
I
Igor Canadi 已提交
1797
      options_.purge_redundant_kvs_while_flush = false;
1798
    }
1799

1800
    if (FLAGS_use_merge) {
I
Igor Canadi 已提交
1801
      options_.merge_operator = MergeOperators::CreatePutOperator();
D
Deon Nicholas 已提交
1802 1803
    }

1804 1805
    // set universal style compaction configurations, if applicable
    if (FLAGS_universal_size_ratio != 0) {
I
Igor Canadi 已提交
1806 1807
      options_.compaction_options_universal.size_ratio =
          FLAGS_universal_size_ratio;
1808 1809
    }
    if (FLAGS_universal_min_merge_width != 0) {
I
Igor Canadi 已提交
1810 1811
      options_.compaction_options_universal.min_merge_width =
          FLAGS_universal_min_merge_width;
1812 1813
    }
    if (FLAGS_universal_max_merge_width != 0) {
I
Igor Canadi 已提交
1814 1815
      options_.compaction_options_universal.max_merge_width =
          FLAGS_universal_max_merge_width;
1816 1817
    }
    if (FLAGS_universal_max_size_amplification_percent != 0) {
I
Igor Canadi 已提交
1818 1819
      options_.compaction_options_universal.max_size_amplification_percent =
          FLAGS_universal_max_size_amplification_percent;
1820 1821
    }

1822
    fprintf(stdout, "DB path: [%s]\n", FLAGS_db.c_str());
1823

1824 1825
    Status s;
    if (FLAGS_ttl == -1) {
I
Igor Canadi 已提交
1826 1827 1828 1829 1830 1831 1832
      std::vector<std::string> existing_column_families;
      s = DB::ListColumnFamilies(DBOptions(options_), FLAGS_db,
                                 &existing_column_families);  // ignore errors
      if (!s.ok()) {
        // DB doesn't exist
        assert(existing_column_families.empty());
        assert(column_family_names_.empty());
1833
        column_family_names_.push_back(kDefaultColumnFamilyName);
I
Igor Canadi 已提交
1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860
      } else if (column_family_names_.empty()) {
        // this is the first call to the function Open()
        column_family_names_ = existing_column_families;
      } else {
        // this is a reopen. just assert that existing column_family_names are
        // equivalent to what we remember
        auto sorted_cfn = column_family_names_;
        sort(sorted_cfn.begin(), sorted_cfn.end());
        sort(existing_column_families.begin(), existing_column_families.end());
        if (sorted_cfn != existing_column_families) {
          fprintf(stderr,
                  "Expected column families differ from the existing:\n");
          printf("Expected: {");
          for (auto cf : sorted_cfn) {
            printf("%s ", cf.c_str());
          }
          printf("}\n");
          printf("Existing: {");
          for (auto cf : existing_column_families) {
            printf("%s ", cf.c_str());
          }
          printf("}\n");
        }
        assert(sorted_cfn == existing_column_families);
      }
      std::vector<ColumnFamilyDescriptor> cf_descriptors;
      for (auto name : column_family_names_) {
1861
        if (name != kDefaultColumnFamilyName) {
I
Igor Canadi 已提交
1862 1863 1864 1865 1866
          new_column_family_name_ =
              std::max(new_column_family_name_.load(), std::stoi(name) + 1);
        }
        cf_descriptors.emplace_back(name, ColumnFamilyOptions(options_));
      }
I
Igor Canadi 已提交
1867 1868 1869 1870 1871 1872 1873
      while (cf_descriptors.size() < (size_t)FLAGS_column_families) {
        std::string name = std::to_string(new_column_family_name_.load());
        new_column_family_name_++;
        cf_descriptors.emplace_back(name, ColumnFamilyOptions(options_));
        column_family_names_.push_back(name);
      }
      options_.create_missing_column_families = true;
I
Igor Canadi 已提交
1874 1875 1876 1877
      s = DB::Open(DBOptions(options_), FLAGS_db, cf_descriptors,
                   &column_families_, &db_);
      assert(!s.ok() || column_families_.size() ==
                            static_cast<size_t>(FLAGS_column_families));
1878
    } else {
I
Igor Canadi 已提交
1879 1880 1881
      DBWithTTL* db_with_ttl;
      s = DBWithTTL::Open(options_, FLAGS_db, &db_with_ttl, FLAGS_ttl);
      db_ = db_with_ttl;
1882
    }
1883 1884 1885 1886 1887 1888
    if (!s.ok()) {
      fprintf(stderr, "open error: %s\n", s.ToString().c_str());
      exit(1);
    }
  }

1889
  void Reopen() {
I
Igor Canadi 已提交
1890 1891
    for (auto cf : column_families_) {
      delete cf;
1892
    }
I
Igor Canadi 已提交
1893 1894
    column_families_.clear();
    delete db_;
1895
    db_ = nullptr;
1896 1897 1898 1899 1900 1901

    num_times_reopened_++;
    double now = FLAGS_env->NowMicros();
    fprintf(stdout, "%s Reopening database for the %dth time\n",
            FLAGS_env->TimeToString((uint64_t) now/1000000).c_str(),
            num_times_reopened_);
1902 1903 1904
    Open();
  }

1905 1906
  void PrintStatistics() {
    if (dbstats) {
1907
      fprintf(stdout, "STATISTICS:\n%s\n", dbstats->ToString().c_str());
1908 1909 1910 1911
    }
  }

 private:
1912 1913 1914
  std::shared_ptr<Cache> cache_;
  std::shared_ptr<Cache> compressed_cache_;
  std::shared_ptr<const FilterPolicy> filter_policy_;
1915
  DB* db_;
I
Igor Canadi 已提交
1916 1917 1918 1919
  Options options_;
  std::vector<ColumnFamilyHandle*> column_families_;
  std::vector<std::string> column_family_names_;
  std::atomic<int> new_column_family_name_;
1920
  int num_times_reopened_;
L
Lei Jin 已提交
1921 1922
  std::unordered_map<std::string, std::vector<std::string>> options_table_;
  std::vector<std::string> options_index_;
1923 1924
};

1925
}  // namespace rocksdb
1926 1927

int main(int argc, char** argv) {
1928 1929 1930
  SetUsageMessage(std::string("\nUSAGE:\n") + std::string(argv[0]) +
                  " [OPTIONS]...");
  ParseCommandLineFlags(&argc, &argv, true);
1931 1932 1933 1934 1935 1936 1937 1938

  if (FLAGS_statistics) {
    dbstats = rocksdb::CreateDBStatistics();
  }
  FLAGS_compression_type_e =
    StringToCompressionType(FLAGS_compression_type.c_str());
  if (!FLAGS_hdfs.empty()) {
    FLAGS_env  = new rocksdb::HdfsEnv(FLAGS_hdfs);
1939
  }
1940
  FLAGS_rep_factory = StringToRepFactory(FLAGS_memtablerep.c_str());
1941

1942 1943 1944 1945
  // The number of background threads should be at least as much the
  // max number of concurrent compactions.
  FLAGS_env->SetBackgroundThreads(FLAGS_max_background_compactions);

L
Lei Jin 已提交
1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957
  if (FLAGS_prefixpercent > 0 && FLAGS_prefix_size <= 0) {
    fprintf(stderr,
            "Error: prefixpercent is non-zero while prefix_size is "
            "not positive!\n");
    exit(1);
  }
  if (FLAGS_test_batches_snapshots && FLAGS_prefix_size <= 0) {
    fprintf(stderr,
            "Error: please specify prefix_size for "
            "test_batches_snapshots test!\n");
    exit(1);
  }
1958
  if ((FLAGS_readpercent + FLAGS_prefixpercent +
1959 1960 1961
       FLAGS_writepercent + FLAGS_delpercent + FLAGS_iterpercent) != 100) {
      fprintf(stderr,
              "Error: Read+Prefix+Write+Delete+Iterate percents != 100!\n");
A
amayank 已提交
1962 1963
      exit(1);
  }
1964 1965 1966 1967
  if (FLAGS_disable_wal == 1 && FLAGS_reopen > 0) {
      fprintf(stderr, "Error: Db cannot reopen safely with disable_wal set!\n");
      exit(1);
  }
1968
  if ((unsigned)FLAGS_reopen >= FLAGS_ops_per_thread) {
K
kailiu 已提交
1969 1970 1971 1972 1973
      fprintf(stderr,
              "Error: #DB-reopens should be < ops_per_thread\n"
              "Provided reopens = %d and ops_per_thread = %lu\n",
              FLAGS_reopen,
              (unsigned long)FLAGS_ops_per_thread);
1974 1975 1976
      exit(1);
  }

1977
  // Choose a location for the test database if none given with --db=<path>
1978 1979
  if (FLAGS_db.empty()) {
      std::string default_db_path;
1980
      rocksdb::Env::Default()->GetTestDirectory(&default_db_path);
1981
      default_db_path += "/dbstress";
1982
      FLAGS_db = default_db_path;
1983 1984
  }

1985
  rocksdb::StressTest stress;
I
Igor Canadi 已提交
1986 1987 1988 1989 1990
  if (stress.Run()) {
    return 0;
  } else {
    return 1;
  }
1991
}
1992 1993

#endif  // GFLAGS