rocksdb/util/bloom.cc

//  Copyright (c) 2011-present, Facebook, Inc.  All rights reserved.
//  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).
//
// Copyright (c) 2012 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.

#include "rocksdb/filter_policy.h"

#include "rocksdb/slice.h"
#include "table/block_based/block_based_filter_block.h"
#include "table/block_based/full_filter_block.h"
#include "table/full_filter_bits_builder.h"
#include "third-party/folly/folly/ConstexprMath.h"
#include "util/bloom_impl.h"
#include "util/coding.h"
#include "util/hash.h"

namespace rocksdb {

typedef LegacyLocalityBloomImpl</*ExtraRotates*/ false> LegacyFullFilterImpl;
class BlockBasedFilterBlockBuilder;
class FullFilterBlockBuilder;

FullFilterBitsBuilder::FullFilterBitsBuilder(const int bits_per_key,
                                             const int num_probes)
    : bits_per_key_(bits_per_key), num_probes_(num_probes) {
  assert(bits_per_key_);
}

  FullFilterBitsBuilder::~FullFilterBitsBuilder() {}

  void FullFilterBitsBuilder::AddKey(const Slice& key) {
    uint32_t hash = BloomHash(key);
    if (hash_entries_.size() == 0 || hash != hash_entries_.back()) {
      hash_entries_.push_back(hash);
    }
  }

  Slice FullFilterBitsBuilder::Finish(std::unique_ptr<const char[]>* buf) {
    uint32_t total_bits, num_lines;
    char* data = ReserveSpace(static_cast<int>(hash_entries_.size()),
                              &total_bits, &num_lines);
    assert(data);

    if (total_bits != 0 && num_lines != 0) {
      for (auto h : hash_entries_) {
        AddHash(h, data, num_lines, total_bits);
      }
    }
    data[total_bits/8] = static_cast<char>(num_probes_);
    EncodeFixed32(data + total_bits/8 + 1, static_cast<uint32_t>(num_lines));

    const char* const_data = data;
    buf->reset(const_data);
    hash_entries_.clear();

    return Slice(data, total_bits / 8 + 5);
  }

uint32_t FullFilterBitsBuilder::GetTotalBitsForLocality(uint32_t total_bits) {
  uint32_t num_lines =
      (total_bits + CACHE_LINE_SIZE * 8 - 1) / (CACHE_LINE_SIZE * 8);

  // Make num_lines an odd number to make sure more bits are involved
  // when determining which block.
  if (num_lines % 2 == 0) {
    num_lines++;
  }
  return num_lines * (CACHE_LINE_SIZE * 8);
}

uint32_t FullFilterBitsBuilder::CalculateSpace(const int num_entry,
                                               uint32_t* total_bits,
                                               uint32_t* num_lines) {
  assert(bits_per_key_);
  if (num_entry != 0) {
    uint32_t total_bits_tmp = static_cast<uint32_t>(num_entry * bits_per_key_);

    *total_bits = GetTotalBitsForLocality(total_bits_tmp);
    *num_lines = *total_bits / (CACHE_LINE_SIZE * 8);
    assert(*total_bits > 0 && *total_bits % 8 == 0);
  } else {
    // filter is empty, just leave space for metadata
    *total_bits = 0;
    *num_lines = 0;
  }

  // Reserve space for Filter
  uint32_t sz = *total_bits / 8;
  sz += 5;  // 4 bytes for num_lines, 1 byte for num_probes
  return sz;
}

char* FullFilterBitsBuilder::ReserveSpace(const int num_entry,
                                          uint32_t* total_bits,
                                          uint32_t* num_lines) {
  uint32_t sz = CalculateSpace(num_entry, total_bits, num_lines);
  char* data = new char[sz];
  memset(data, 0, sz);
  return data;
}

int FullFilterBitsBuilder::CalculateNumEntry(const uint32_t space) {
  assert(bits_per_key_);
  assert(space > 0);
  uint32_t dont_care1, dont_care2;
  int high = static_cast<int>(space * 8 / bits_per_key_ + 1);
  int low = 1;
  int n = high;
  for (; n >= low; n--) {
    uint32_t sz = CalculateSpace(n, &dont_care1, &dont_care2);
    if (sz <= space) {
      break;
    }
  }
  assert(n < high);  // High should be an overestimation
  return n;
}

inline void FullFilterBitsBuilder::AddHash(uint32_t h, char* data,
    uint32_t num_lines, uint32_t total_bits) {
#ifdef NDEBUG
  static_cast<void>(total_bits);
#endif
  assert(num_lines > 0 && total_bits > 0);

  LegacyFullFilterImpl::AddHash(h, num_lines, num_probes_, data,
                                folly::constexpr_log2(CACHE_LINE_SIZE));
}

namespace {
class FullFilterBitsReader : public FilterBitsReader {
 public:
  explicit FullFilterBitsReader(const Slice& contents)
      : data_(contents.data()),
        data_len_(static_cast<uint32_t>(contents.size())),
        num_probes_(0),
        num_lines_(0),
        log2_cache_line_size_(0) {
    assert(data_);
    GetFilterMeta(contents, &num_probes_, &num_lines_);
    // Sanitize broken parameter
    if (num_lines_ != 0 && (data_len_-5) % num_lines_ != 0) {
      num_lines_ = 0;
      num_probes_ = 0;
    } else if (num_lines_ != 0) {
      while (true) {
        uint32_t num_lines_at_curr_cache_size =
            (data_len_ - 5) >> log2_cache_line_size_;
        if (num_lines_at_curr_cache_size == 0) {
          // The cache line size seems not a power of two. It's not supported
          // and indicates a corruption so disable using this filter.
          assert(false);
          num_lines_ = 0;
          num_probes_ = 0;
          break;
        }
        if (num_lines_at_curr_cache_size == num_lines_) {
          break;
        }
        ++log2_cache_line_size_;
      }
    }
  }
  // No Copy allowed
  FullFilterBitsReader(const FullFilterBitsReader&) = delete;
  void operator=(const FullFilterBitsReader&) = delete;

  ~FullFilterBitsReader() override {}

  // "contents" contains the data built by a preceding call to
  // FilterBitsBuilder::Finish. MayMatch must return true if the key was
  // passed to FilterBitsBuilder::AddKey. This method may return true or false
  // if the key was not on the list, but it should aim to return false with a
  // high probability.
  bool MayMatch(const Slice& key) override {
    if (data_len_ <= 5) {   // remain same with original filter
      return false;
    }
    // Other Error params, including a broken filter, regarded as match
    if (num_probes_ == 0 || num_lines_ == 0) return true;
    uint32_t hash = BloomHash(key);
    uint32_t byte_offset;
    LegacyFullFilterImpl::PrepareHashMayMatch(
        hash, num_lines_, data_, /*out*/ &byte_offset, log2_cache_line_size_);
    return LegacyFullFilterImpl::HashMayMatchPrepared(
        hash, num_probes_, data_ + byte_offset, log2_cache_line_size_);
  }

  virtual void MayMatch(int num_keys, Slice** keys, bool* may_match) override {
    if (data_len_ <= 5) {  // remain same with original filter
      for (int i = 0; i < num_keys; ++i) {
        may_match[i] = false;
      }
      return;
    }
    for (int i = 0; i < num_keys; ++i) {
      may_match[i] = true;
    }
    // Other Error params, including a broken filter, regarded as match
    if (num_probes_ == 0 || num_lines_ == 0) return;
    uint32_t hashes[MultiGetContext::MAX_BATCH_SIZE];
    uint32_t byte_offsets[MultiGetContext::MAX_BATCH_SIZE];
    for (int i = 0; i < num_keys; ++i) {
      hashes[i] = BloomHash(*keys[i]);
      LegacyFullFilterImpl::PrepareHashMayMatch(hashes[i], num_lines_, data_,
                                                /*out*/ &byte_offsets[i],
                                                log2_cache_line_size_);
    }

    for (int i = 0; i < num_keys; ++i) {
      if (!LegacyFullFilterImpl::HashMayMatchPrepared(hashes[i], num_probes_,
                                                      data_ + byte_offsets[i],
                                                      log2_cache_line_size_)) {
        may_match[i] = false;
      }
    }
  }

 private:
  // Filter meta data
  const char* data_;
  uint32_t data_len_;
  int num_probes_;
  uint32_t num_lines_;
  uint32_t log2_cache_line_size_;

  // Get num_probes, and num_lines from filter
  // If filter format broken, set both to 0.
  void GetFilterMeta(const Slice& filter, int* num_probes, uint32_t* num_lines);
};

void FullFilterBitsReader::GetFilterMeta(const Slice& filter, int* num_probes,
                                         uint32_t* num_lines) {
  uint32_t len = static_cast<uint32_t>(filter.size());
  if (len <= 5) {
    // filter is empty or broken
    *num_probes = 0;
    *num_lines = 0;
    return;
  }

  *num_probes = filter.data()[len - 5];
  *num_lines = DecodeFixed32(filter.data() + len - 4);
}

// An implementation of filter policy
class BloomFilterPolicy : public FilterPolicy {
 public:
  explicit BloomFilterPolicy(int bits_per_key, bool use_block_based_builder)
      : bits_per_key_(bits_per_key), hash_func_(BloomHash),
        use_block_based_builder_(use_block_based_builder) {
    initialize();
  }

  ~BloomFilterPolicy() override {}

  const char* Name() const override { return "rocksdb.BuiltinBloomFilter"; }

  void CreateFilter(const Slice* keys, int n, std::string* dst) const override {
    // Compute bloom filter size (in both bits and bytes)
    uint32_t bits = static_cast<uint32_t>(n * bits_per_key_);

    // For small n, we can see a very high false positive rate.  Fix it
    // by enforcing a minimum bloom filter length.
    if (bits < 64) bits = 64;

    uint32_t bytes = (bits + 7) / 8;
    bits = bytes * 8;

    const size_t init_size = dst->size();
    dst->resize(init_size + bytes, 0);
    dst->push_back(static_cast<char>(num_probes_));  // Remember # of probes
    char* array = &(*dst)[init_size];
    for (int i = 0; i < n; i++) {
      LegacyNoLocalityBloomImpl::AddHash(hash_func_(keys[i]), bits, num_probes_,
                                         array);
    }
  }

  bool KeyMayMatch(const Slice& key, const Slice& bloom_filter) const override {
    const size_t len = bloom_filter.size();
    if (len < 2 || len > 0xffffffffU) {
      return false;
    }

    const char* array = bloom_filter.data();
    const uint32_t bits = static_cast<uint32_t>(len - 1) * 8;

    // Use the encoded k so that we can read filters generated by
    // bloom filters created using different parameters.
    const int k = static_cast<uint8_t>(array[len - 1]);
    if (k > 30) {
      // Reserved for potentially new encodings for short bloom filters.
      // Consider it a match.
      return true;
    }
    // NB: using k not num_probes_
    return LegacyNoLocalityBloomImpl::HashMayMatch(hash_func_(key), bits, k,
                                                   array);
  }

  FilterBitsBuilder* GetFilterBitsBuilder() const override {
    if (use_block_based_builder_) {
      return nullptr;
    }

    return new FullFilterBitsBuilder(bits_per_key_, num_probes_);
  }

  FilterBitsReader* GetFilterBitsReader(const Slice& contents) const override {
    return new FullFilterBitsReader(contents);
  }

  // If choose to use block based builder
  bool UseBlockBasedBuilder() { return use_block_based_builder_; }

 private:
  int bits_per_key_;
  int num_probes_;
  uint32_t (*hash_func_)(const Slice& key);

  const bool use_block_based_builder_;

  void initialize() {
    // We intentionally round down to reduce probing cost a little bit
    num_probes_ = static_cast<int>(bits_per_key_ * 0.69);  // 0.69 =~ ln(2)
    if (num_probes_ < 1) num_probes_ = 1;
    if (num_probes_ > 30) num_probes_ = 30;
  }
};

}  // namespace

const FilterPolicy* NewBloomFilterPolicy(int bits_per_key,
                                         bool use_block_based_builder) {
  return new BloomFilterPolicy(bits_per_key, use_block_based_builder);
}

}  // namespace rocksdb