block_based_table_factory.cc 24.9 KB
Newer Older
1
//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
S
Siying Dong 已提交
2 3 4
//  This source code is licensed under both the GPLv2 (found in the
//  COPYING file in the root directory) and Apache 2.0 License
//  (found in the LICENSE.Apache file in the root directory).
S
Siying Dong 已提交
5 6 7 8 9
//
// 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.

10
#include <cinttypes>
Y
Yi Wu 已提交
11 12
#include <stdint.h>

S
Siying Dong 已提交
13
#include <memory>
14 15
#include <string>

S
Siying Dong 已提交
16
#include "options/options_helper.h"
17
#include "port/port.h"
18
#include "rocksdb/cache.h"
S
Siying Dong 已提交
19 20
#include "rocksdb/convenience.h"
#include "rocksdb/flush_block_policy.h"
21
#include "table/block_based/block_based_table_builder.h"
22
#include "table/block_based/block_based_table_factory.h"
23
#include "table/block_based/block_based_table_reader.h"
24
#include "table/format.h"
25
#include "util/mutexlock.h"
S
Siying Dong 已提交
26
#include "util/string_util.h"
S
Siying Dong 已提交
27 28 29

namespace rocksdb {

30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159
void TailPrefetchStats::RecordEffectiveSize(size_t len) {
  MutexLock l(&mutex_);
  if (num_records_ < kNumTracked) {
    num_records_++;
  }
  records_[next_++] = len;
  if (next_ == kNumTracked) {
    next_ = 0;
  }
}

size_t TailPrefetchStats::GetSuggestedPrefetchSize() {
  std::vector<size_t> sorted;
  {
    MutexLock l(&mutex_);

    if (num_records_ == 0) {
      return 0;
    }
    sorted.assign(records_, records_ + num_records_);
  }

  // Of the historic size, we find the maximum one that satisifis the condtiion
  // that if prefetching all, less than 1/8 will be wasted.
  std::sort(sorted.begin(), sorted.end());

  // Assuming we have 5 data points, and after sorting it looks like this:
  //
  //                                     +---+
  //                             +---+   |   |
  //                             |   |   |   |
  //                             |   |   |   |
  //                             |   |   |   |
  //                             |   |   |   |
  //                    +---+    |   |   |   |
  //                    |   |    |   |   |   |
  //           +---+    |   |    |   |   |   |
  //           |   |    |   |    |   |   |   |
  //  +---+    |   |    |   |    |   |   |   |
  //  |   |    |   |    |   |    |   |   |   |
  //  |   |    |   |    |   |    |   |   |   |
  //  |   |    |   |    |   |    |   |   |   |
  //  |   |    |   |    |   |    |   |   |   |
  //  |   |    |   |    |   |    |   |   |   |
  //  +---+    +---+    +---+    +---+   +---+
  //
  // and we use every of the value as a candidate, and estimate how much we
  // wasted, compared to read. For example, when we use the 3rd record
  // as candiate. This area is what we read:
  //                                     +---+
  //                             +---+   |   |
  //                             |   |   |   |
  //                             |   |   |   |
  //                             |   |   |   |
  //                             |   |   |   |
  //  ***  ***  ***  ***+ ***  ***  *** *** **
  //  *                 |   |    |   |   |   |
  //           +---+    |   |    |   |   |   *
  //  *        |   |    |   |    |   |   |   |
  //  +---+    |   |    |   |    |   |   |   *
  //  *   |    |   |    | X |    |   |   |   |
  //  |   |    |   |    |   |    |   |   |   *
  //  *   |    |   |    |   |    |   |   |   |
  //  |   |    |   |    |   |    |   |   |   *
  //  *   |    |   |    |   |    |   |   |   |
  //  *** *** ***-***  ***--*** ***--*** +****
  // which is (size of the record) X (number of records).
  //
  // While wasted is this area:
  //                                     +---+
  //                             +---+   |   |
  //                             |   |   |   |
  //                             |   |   |   |
  //                             |   |   |   |
  //                             |   |   |   |
  //  ***  ***  ***  ****---+    |   |   |   |
  //  *                 *   |    |   |   |   |
  //  *        *-***  ***   |    |   |   |   |
  //  *        *   |    |   |    |   |   |   |
  //  *--**  ***   |    |   |    |   |   |   |
  //  |   |    |   |    | X |    |   |   |   |
  //  |   |    |   |    |   |    |   |   |   |
  //  |   |    |   |    |   |    |   |   |   |
  //  |   |    |   |    |   |    |   |   |   |
  //  |   |    |   |    |   |    |   |   |   |
  //  +---+    +---+    +---+    +---+   +---+
  //
  // Which can be calculated iteratively.
  // The difference between wasted using 4st and 3rd record, will
  // be following area:
  //                                     +---+
  //  +--+  +-+   ++  +-+  +-+   +---+   |   |
  //  + xxxxxxxxxxxxxxxxxxxxxxxx |   |   |   |
  //    xxxxxxxxxxxxxxxxxxxxxxxx |   |   |   |
  //  + xxxxxxxxxxxxxxxxxxxxxxxx |   |   |   |
  //  | xxxxxxxxxxxxxxxxxxxxxxxx |   |   |   |
  //  +-+ +-+  +-+  ++  +---+ +--+   |   |   |
  //  |                 |   |    |   |   |   |
  //           +---+ ++ |   |    |   |   |   |
  //  |        |   |    |   |    | X |   |   |
  //  +---+ ++ |   |    |   |    |   |   |   |
  //  |   |    |   |    |   |    |   |   |   |
  //  |   |    |   |    |   |    |   |   |   |
  //  |   |    |   |    |   |    |   |   |   |
  //  |   |    |   |    |   |    |   |   |   |
  //  |   |    |   |    |   |    |   |   |   |
  //  +---+    +---+    +---+    +---+   +---+
  //
  // which will be the size difference between 4st and 3rd record,
  // times 3, which is number of records before the 4st.
  // Here we assume that all data within the prefetch range will be useful. In
  // reality, it may not be the case when a partial block is inside the range,
  // or there are data in the middle that is not read. We ignore those cases
  // for simplicity.
  assert(!sorted.empty());
  size_t prev_size = sorted[0];
  size_t max_qualified_size = sorted[0];
  size_t wasted = 0;
  for (size_t i = 1; i < sorted.size(); i++) {
    size_t read = sorted[i] * sorted.size();
    wasted += (sorted[i] - prev_size) * i;
    if (wasted <= read / 8) {
      max_qualified_size = sorted[i];
    }
    prev_size = sorted[i];
  }
  const size_t kMaxPrefetchSize = 512 * 1024;  // Never exceed 512KB
  return std::min(kMaxPrefetchSize, max_qualified_size);
}

160
BlockBasedTableFactory::BlockBasedTableFactory(
161 162
    const BlockBasedTableOptions& _table_options)
    : table_options_(_table_options) {
163 164 165 166
  if (table_options_.flush_block_policy_factory == nullptr) {
    table_options_.flush_block_policy_factory.reset(
        new FlushBlockBySizePolicyFactory());
  }
167 168 169 170 171 172 173 174 175
  if (table_options_.no_block_cache) {
    table_options_.block_cache.reset();
  } else if (table_options_.block_cache == nullptr) {
    table_options_.block_cache = NewLRUCache(8 << 20);
  }
  if (table_options_.block_size_deviation < 0 ||
      table_options_.block_size_deviation > 100) {
    table_options_.block_size_deviation = 0;
  }
176 177 178
  if (table_options_.block_restart_interval < 1) {
    table_options_.block_restart_interval = 1;
  }
179 180 181
  if (table_options_.index_block_restart_interval < 1) {
    table_options_.index_block_restart_interval = 1;
  }
M
Maysam Yabandeh 已提交
182 183 184 185 186 187
  if (table_options_.partition_filters &&
      table_options_.index_type !=
          BlockBasedTableOptions::kTwoLevelIndexSearch) {
    // We do not support partitioned filters without partitioning indexes
    table_options_.partition_filters = false;
  }
188 189
}

K
kailiu 已提交
190
Status BlockBasedTableFactory::NewTableReader(
191
    const TableReaderOptions& table_reader_options,
192 193
    std::unique_ptr<RandomAccessFileReader>&& file, uint64_t file_size,
    std::unique_ptr<TableReader>* table_reader,
194
    bool prefetch_index_and_filter_in_cache) const {
195 196 197
  return BlockBasedTable::Open(
      table_reader_options.ioptions, table_reader_options.env_options,
      table_options_, table_reader_options.internal_comparator, std::move(file),
198 199
      file_size, table_reader, table_reader_options.prefix_extractor,
      prefetch_index_and_filter_in_cache, table_reader_options.skip_filters,
200
      table_reader_options.level, table_reader_options.immortal,
201
      table_reader_options.largest_seqno, &tail_prefetch_stats_);
S
Siying Dong 已提交
202 203
}

K
kailiu 已提交
204
TableBuilder* BlockBasedTableFactory::NewTableBuilder(
205
    const TableBuilderOptions& table_builder_options, uint32_t column_family_id,
206
    WritableFileWriter* file) const {
207
  auto table_builder = new BlockBasedTableBuilder(
208 209
      table_builder_options.ioptions, table_builder_options.moptions,
      table_options_, table_builder_options.internal_comparator,
210 211
      table_builder_options.int_tbl_prop_collector_factories, column_family_id,
      file, table_builder_options.compression_type,
212
      table_builder_options.sample_for_compression,
213
      table_builder_options.compression_opts,
214
      table_builder_options.skip_filters,
S
Sagar Vemuri 已提交
215
      table_builder_options.column_family_name,
Y
Yi Wu 已提交
216
      table_builder_options.creation_time,
217
      table_builder_options.oldest_key_time,
S
Sagar Vemuri 已提交
218 219
      table_builder_options.target_file_size,
      table_builder_options.file_creation_time);
220 221

  return table_builder;
S
Siying Dong 已提交
222
}
223

224
Status BlockBasedTableFactory::SanitizeOptions(
M
Maysam Yabandeh 已提交
225
    const DBOptions& db_opts, const ColumnFamilyOptions& cf_opts) const {
226 227
  if (table_options_.index_type == BlockBasedTableOptions::kHashSearch &&
      cf_opts.prefix_extractor == nullptr) {
228 229
    return Status::InvalidArgument(
        "Hash index is specified for block-based "
230 231
        "table, but prefix_extractor is not given");
  }
232 233
  if (table_options_.cache_index_and_filter_blocks &&
      table_options_.no_block_cache) {
234 235
    return Status::InvalidArgument(
        "Enable cache_index_and_filter_blocks, "
236 237
        ", but block cache is disabled");
  }
238 239 240 241 242 243
  if (table_options_.pin_l0_filter_and_index_blocks_in_cache &&
      table_options_.no_block_cache) {
    return Status::InvalidArgument(
        "Enable pin_l0_filter_and_index_blocks_in_cache, "
        ", but block cache is disabled");
  }
244
  if (!BlockBasedTableSupportedVersion(table_options_.format_version)) {
245
    return Status::InvalidArgument(
246 247
        "Unsupported BlockBasedTable format_version. Please check "
        "include/rocksdb/table.h for more info");
248
  }
249
  if (table_options_.block_align && (cf_opts.compression != kNoCompression)) {
250 251
    return Status::InvalidArgument(
        "Enable block_align, but compression "
252 253 254 255 256 257 258
        "enabled");
  }
  if (table_options_.block_align &&
      (table_options_.block_size & (table_options_.block_size - 1))) {
    return Status::InvalidArgument(
        "Block alignment requested but block size is not a power of 2");
  }
259
  if (table_options_.data_block_index_type ==
260
          BlockBasedTableOptions::kDataBlockBinaryAndHash &&
261 262 263 264 265
      table_options_.data_block_hash_table_util_ratio <= 0) {
    return Status::InvalidArgument(
        "data_block_hash_table_util_ratio should be greater than 0 when "
        "data_block_index_type is set to kDataBlockBinaryAndHash");
  }
M
Maysam Yabandeh 已提交
266 267 268 269 270 271
  if (db_opts.unordered_write && cf_opts.max_successive_merges > 0) {
    // TODO(myabandeh): support it
    return Status::InvalidArgument(
        "max_successive_merges larger than 0 is currently inconsistent with "
        "unordered_write");
  }
272 273 274
  return Status::OK();
}

L
Lei Jin 已提交
275 276 277 278 279 280 281 282
std::string BlockBasedTableFactory::GetPrintableTableOptions() const {
  std::string ret;
  ret.reserve(20000);
  const int kBufferSize = 200;
  char buffer[kBufferSize];

  snprintf(buffer, kBufferSize, "  flush_block_policy_factory: %s (%p)\n",
           table_options_.flush_block_policy_factory->Name(),
283
           static_cast<void*>(table_options_.flush_block_policy_factory.get()));
L
Lei Jin 已提交
284 285 286 287
  ret.append(buffer);
  snprintf(buffer, kBufferSize, "  cache_index_and_filter_blocks: %d\n",
           table_options_.cache_index_and_filter_blocks);
  ret.append(buffer);
Y
Yi Wu 已提交
288 289 290 291
  snprintf(buffer, kBufferSize,
           "  cache_index_and_filter_blocks_with_high_priority: %d\n",
           table_options_.cache_index_and_filter_blocks_with_high_priority);
  ret.append(buffer);
292 293 294 295
  snprintf(buffer, kBufferSize,
           "  pin_l0_filter_and_index_blocks_in_cache: %d\n",
           table_options_.pin_l0_filter_and_index_blocks_in_cache);
  ret.append(buffer);
296 297 298
  snprintf(buffer, kBufferSize, "  pin_top_level_index_and_filter: %d\n",
           table_options_.pin_top_level_index_and_filter);
  ret.append(buffer);
L
Lei Jin 已提交
299 300 301
  snprintf(buffer, kBufferSize, "  index_type: %d\n",
           table_options_.index_type);
  ret.append(buffer);
302 303 304
  snprintf(buffer, kBufferSize, "  data_block_index_type: %d\n",
           table_options_.data_block_index_type);
  ret.append(buffer);
305 306 307
  snprintf(buffer, kBufferSize, "  index_shortening: %d\n",
           static_cast<int>(table_options_.index_shortening));
  ret.append(buffer);
308 309 310
  snprintf(buffer, kBufferSize, "  data_block_hash_table_util_ratio: %lf\n",
           table_options_.data_block_hash_table_util_ratio);
  ret.append(buffer);
L
Lei Jin 已提交
311 312 313
  snprintf(buffer, kBufferSize, "  hash_index_allow_collision: %d\n",
           table_options_.hash_index_allow_collision);
  ret.append(buffer);
314
  snprintf(buffer, kBufferSize, "  checksum: %d\n", table_options_.checksum);
L
Lei Jin 已提交
315 316 317 318 319
  ret.append(buffer);
  snprintf(buffer, kBufferSize, "  no_block_cache: %d\n",
           table_options_.no_block_cache);
  ret.append(buffer);
  snprintf(buffer, kBufferSize, "  block_cache: %p\n",
320
           static_cast<void*>(table_options_.block_cache.get()));
L
Lei Jin 已提交
321 322
  ret.append(buffer);
  if (table_options_.block_cache) {
Y
Yi Wu 已提交
323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344
    const char* block_cache_name = table_options_.block_cache->Name();
    if (block_cache_name != nullptr) {
      snprintf(buffer, kBufferSize, "  block_cache_name: %s\n",
               block_cache_name);
      ret.append(buffer);
    }
    ret.append("  block_cache_options:\n");
    ret.append(table_options_.block_cache->GetPrintableOptions());
  }
  snprintf(buffer, kBufferSize, "  block_cache_compressed: %p\n",
           static_cast<void*>(table_options_.block_cache_compressed.get()));
  ret.append(buffer);
  if (table_options_.block_cache_compressed) {
    const char* block_cache_compressed_name =
        table_options_.block_cache_compressed->Name();
    if (block_cache_compressed_name != nullptr) {
      snprintf(buffer, kBufferSize, "  block_cache_name: %s\n",
               block_cache_compressed_name);
      ret.append(buffer);
    }
    ret.append("  block_cache_compressed_options:\n");
    ret.append(table_options_.block_cache_compressed->GetPrintableOptions());
L
Lei Jin 已提交
345
  }
Y
Yi Wu 已提交
346 347 348 349 350 351 352 353
  snprintf(buffer, kBufferSize, "  persistent_cache: %p\n",
           static_cast<void*>(table_options_.persistent_cache.get()));
  ret.append(buffer);
  if (table_options_.persistent_cache) {
    snprintf(buffer, kBufferSize, "  persistent_cache_options:\n");
    ret.append(buffer);
    ret.append(table_options_.persistent_cache->GetPrintableOptions());
  }
D
Dmitri Smirnov 已提交
354
  snprintf(buffer, kBufferSize, "  block_size: %" ROCKSDB_PRIszt "\n",
L
Lei Jin 已提交
355 356 357 358 359 360 361 362
           table_options_.block_size);
  ret.append(buffer);
  snprintf(buffer, kBufferSize, "  block_size_deviation: %d\n",
           table_options_.block_size_deviation);
  ret.append(buffer);
  snprintf(buffer, kBufferSize, "  block_restart_interval: %d\n",
           table_options_.block_restart_interval);
  ret.append(buffer);
363 364 365
  snprintf(buffer, kBufferSize, "  index_block_restart_interval: %d\n",
           table_options_.index_block_restart_interval);
  ret.append(buffer);
Y
Yi Wu 已提交
366 367 368 369 370 371 372 373 374
  snprintf(buffer, kBufferSize, "  metadata_block_size: %" PRIu64 "\n",
           table_options_.metadata_block_size);
  ret.append(buffer);
  snprintf(buffer, kBufferSize, "  partition_filters: %d\n",
           table_options_.partition_filters);
  ret.append(buffer);
  snprintf(buffer, kBufferSize, "  use_delta_encoding: %d\n",
           table_options_.use_delta_encoding);
  ret.append(buffer);
L
Lei Jin 已提交
375
  snprintf(buffer, kBufferSize, "  filter_policy: %s\n",
376 377 378
           table_options_.filter_policy == nullptr
               ? "nullptr"
               : table_options_.filter_policy->Name());
L
Lei Jin 已提交
379 380 381
  ret.append(buffer);
  snprintf(buffer, kBufferSize, "  whole_key_filtering: %d\n",
           table_options_.whole_key_filtering);
382
  ret.append(buffer);
Y
Yi Wu 已提交
383 384 385 386 387 388
  snprintf(buffer, kBufferSize, "  verify_compression: %d\n",
           table_options_.verify_compression);
  ret.append(buffer);
  snprintf(buffer, kBufferSize, "  read_amp_bytes_per_bit: %d\n",
           table_options_.read_amp_bytes_per_bit);
  ret.append(buffer);
389 390
  snprintf(buffer, kBufferSize, "  format_version: %d\n",
           table_options_.format_version);
L
Lei Jin 已提交
391
  ret.append(buffer);
392 393 394
  snprintf(buffer, kBufferSize, "  enable_index_compression: %d\n",
           table_options_.enable_index_compression);
  ret.append(buffer);
395 396 397
  snprintf(buffer, kBufferSize, "  block_align: %d\n",
           table_options_.block_align);
  ret.append(buffer);
L
Lei Jin 已提交
398 399 400
  return ret;
}

S
Siying Dong 已提交
401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444
#ifndef ROCKSDB_LITE
namespace {
bool SerializeSingleBlockBasedTableOption(
    std::string* opt_string, const BlockBasedTableOptions& bbt_options,
    const std::string& name, const std::string& delimiter) {
  auto iter = block_based_table_type_info.find(name);
  if (iter == block_based_table_type_info.end()) {
    return false;
  }
  auto& opt_info = iter->second;
  const char* opt_address =
      reinterpret_cast<const char*>(&bbt_options) + opt_info.offset;
  std::string value;
  bool result = SerializeSingleOptionHelper(opt_address, opt_info.type, &value);
  if (result) {
    *opt_string = name + "=" + value + delimiter;
  }
  return result;
}
}  // namespace

Status BlockBasedTableFactory::GetOptionString(
    std::string* opt_string, const std::string& delimiter) const {
  assert(opt_string);
  opt_string->clear();
  for (auto iter = block_based_table_type_info.begin();
       iter != block_based_table_type_info.end(); ++iter) {
    if (iter->second.verification == OptionVerificationType::kDeprecated) {
      // If the option is no longer used in rocksdb and marked as deprecated,
      // we skip it in the serialization.
      continue;
    }
    std::string single_output;
    bool result = SerializeSingleBlockBasedTableOption(
        &single_output, table_options_, iter->first, delimiter);
    assert(result);
    if (result) {
      opt_string->append(single_output);
    }
  }
  return Status::OK();
}
#else
Status BlockBasedTableFactory::GetOptionString(
445
    std::string* /*opt_string*/, const std::string& /*delimiter*/) const {
S
Siying Dong 已提交
446 447 448 449
  return Status::OK();
}
#endif  // !ROCKSDB_LITE

S
SherlockNoMad 已提交
450
const BlockBasedTableOptions& BlockBasedTableFactory::table_options() const {
451 452 453
  return table_options_;
}

S
Siying Dong 已提交
454 455 456 457 458 459 460 461 462 463 464 465
#ifndef ROCKSDB_LITE
namespace {
std::string ParseBlockBasedTableOption(const std::string& name,
                                       const std::string& org_value,
                                       BlockBasedTableOptions* new_options,
                                       bool input_strings_escaped = false,
                                       bool ignore_unknown_options = false) {
  const std::string& value =
      input_strings_escaped ? UnescapeOptionString(org_value) : org_value;
  if (!input_strings_escaped) {
    // if the input string is not escaped, it means this function is
    // invoked from SetOptions, which takes the old format.
466 467 468 469 470 471 472 473 474 475 476 477 478
    if (name == "block_cache" || name == "block_cache_compressed") {
      // cache options can be specified in the following format
      //   "block_cache={capacity=1M;num_shard_bits=4;
      //    strict_capacity_limit=true;high_pri_pool_ratio=0.5;}"
      // To support backward compatibility, the following format
      // is also supported.
      //   "block_cache=1M"
      std::shared_ptr<Cache> cache;
      // block_cache is specified in format block_cache=<cache_size>.
      if (value.find('=') == std::string::npos) {
        cache = NewLRUCache(ParseSizeT(value));
      } else {
        LRUCacheOptions cache_opts;
479 480
        if (!ParseOptionHelper(reinterpret_cast<char*>(&cache_opts),
                               OptionType::kLRUCacheOptions, value)) {
481 482 483 484 485 486 487 488 489 490
          return "Invalid cache options";
        }
        cache = NewLRUCache(cache_opts);
      }

      if (name == "block_cache") {
        new_options->block_cache = cache;
      } else {
        new_options->block_cache_compressed = cache;
      }
S
Siying Dong 已提交
491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562
      return "";
    } else if (name == "filter_policy") {
      // Expect the following format
      // bloomfilter:int:bool
      const std::string kName = "bloomfilter:";
      if (value.compare(0, kName.size(), kName) != 0) {
        return "Invalid filter policy name";
      }
      size_t pos = value.find(':', kName.size());
      if (pos == std::string::npos) {
        return "Invalid filter policy config, missing bits_per_key";
      }
      int bits_per_key =
          ParseInt(trim(value.substr(kName.size(), pos - kName.size())));
      bool use_block_based_builder =
          ParseBoolean("use_block_based_builder", trim(value.substr(pos + 1)));
      new_options->filter_policy.reset(
          NewBloomFilterPolicy(bits_per_key, use_block_based_builder));
      return "";
    }
  }
  const auto iter = block_based_table_type_info.find(name);
  if (iter == block_based_table_type_info.end()) {
    if (ignore_unknown_options) {
      return "";
    } else {
      return "Unrecognized option";
    }
  }
  const auto& opt_info = iter->second;
  if (opt_info.verification != OptionVerificationType::kDeprecated &&
      !ParseOptionHelper(reinterpret_cast<char*>(new_options) + opt_info.offset,
                         opt_info.type, value)) {
    return "Invalid value";
  }
  return "";
}
}  // namespace

Status GetBlockBasedTableOptionsFromString(
    const BlockBasedTableOptions& table_options, const std::string& opts_str,
    BlockBasedTableOptions* new_table_options) {
  std::unordered_map<std::string, std::string> opts_map;
  Status s = StringToMap(opts_str, &opts_map);
  if (!s.ok()) {
    return s;
  }

  return GetBlockBasedTableOptionsFromMap(table_options, opts_map,
                                          new_table_options);
}

Status GetBlockBasedTableOptionsFromMap(
    const BlockBasedTableOptions& table_options,
    const std::unordered_map<std::string, std::string>& opts_map,
    BlockBasedTableOptions* new_table_options, bool input_strings_escaped,
    bool ignore_unknown_options) {
  assert(new_table_options);
  *new_table_options = table_options;
  for (const auto& o : opts_map) {
    auto error_message = ParseBlockBasedTableOption(
        o.first, o.second, new_table_options, input_strings_escaped,
        ignore_unknown_options);
    if (error_message != "") {
      const auto iter = block_based_table_type_info.find(o.first);
      if (iter == block_based_table_type_info.end() ||
          !input_strings_escaped ||  // !input_strings_escaped indicates
                                     // the old API, where everything is
                                     // parsable.
          (iter->second.verification != OptionVerificationType::kByName &&
           iter->second.verification !=
               OptionVerificationType::kByNameAllowNull &&
563 564
           iter->second.verification !=
               OptionVerificationType::kByNameAllowFromNull &&
S
Siying Dong 已提交
565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613
           iter->second.verification != OptionVerificationType::kDeprecated)) {
        // Restore "new_options" to the default "base_options".
        *new_table_options = table_options;
        return Status::InvalidArgument("Can't parse BlockBasedTableOptions:",
                                       o.first + " " + error_message);
      }
    }
  }
  return Status::OK();
}

Status VerifyBlockBasedTableFactory(
    const BlockBasedTableFactory* base_tf,
    const BlockBasedTableFactory* file_tf,
    OptionsSanityCheckLevel sanity_check_level) {
  if ((base_tf != nullptr) != (file_tf != nullptr) &&
      sanity_check_level > kSanityLevelNone) {
    return Status::Corruption(
        "[RocksDBOptionsParser]: Inconsistent TableFactory class type");
  }
  if (base_tf == nullptr) {
    return Status::OK();
  }
  assert(file_tf != nullptr);

  const auto& base_opt = base_tf->table_options();
  const auto& file_opt = file_tf->table_options();

  for (auto& pair : block_based_table_type_info) {
    if (pair.second.verification == OptionVerificationType::kDeprecated) {
      // We skip checking deprecated variables as they might
      // contain random values since they might not be initialized
      continue;
    }
    if (BBTOptionSanityCheckLevel(pair.first) <= sanity_check_level) {
      if (!AreEqualOptions(reinterpret_cast<const char*>(&base_opt),
                           reinterpret_cast<const char*>(&file_opt),
                           pair.second, pair.first, nullptr)) {
        return Status::Corruption(
            "[RocksDBOptionsParser]: "
            "failed the verification on BlockBasedTableOptions::",
            pair.first);
      }
    }
  }
  return Status::OK();
}
#endif  // !ROCKSDB_LITE

K
kailiu 已提交
614
TableFactory* NewBlockBasedTableFactory(
615 616
    const BlockBasedTableOptions& _table_options) {
  return new BlockBasedTableFactory(_table_options);
K
kailiu 已提交
617 618
}

S
Siying Dong 已提交
619
const std::string BlockBasedTableFactory::kName = "BlockBasedTable";
620 621
const std::string BlockBasedTablePropertyNames::kIndexType =
    "rocksdb.block.based.table.index.type";
622 623 624 625
const std::string BlockBasedTablePropertyNames::kWholeKeyFiltering =
    "rocksdb.block.based.table.whole.key.filtering";
const std::string BlockBasedTablePropertyNames::kPrefixFiltering =
    "rocksdb.block.based.table.prefix.filtering";
K
Kai Liu 已提交
626 627 628
const std::string kHashIndexPrefixesBlock = "rocksdb.hashindex.prefixes";
const std::string kHashIndexPrefixesMetadataBlock =
    "rocksdb.hashindex.metadata";
629 630
const std::string kPropTrue = "1";
const std::string kPropFalse = "0";
631

S
Siying Dong 已提交
632
}  // namespace rocksdb