@nxtedition/rocksdb 5.2.1 → 5.2.2

package/deps/rocksdb/rocksdb/table/plain/plain_table_factory.cc

@@ -0,0 +1,263 @@
+// Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved.
+// 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.
+
+#ifndef ROCKSDB_LITE
+#include "table/plain/plain_table_factory.h"
+
+#include <stdint.h>
+
+#include <memory>
+
+#include "db/dbformat.h"
+#include "options/configurable_helper.h"
+#include "port/port.h"
+#include "rocksdb/convenience.h"
+#include "rocksdb/utilities/options_type.h"
+#include "table/plain/plain_table_builder.h"
+#include "table/plain/plain_table_reader.h"
+#include "util/string_util.h"
+
+namespace ROCKSDB_NAMESPACE {
+static std::unordered_map<std::string, OptionTypeInfo> plain_table_type_info = {
+    {"user_key_len",
+     {offsetof(struct PlainTableOptions, user_key_len), OptionType::kUInt32T,
+      OptionVerificationType::kNormal, OptionTypeFlags::kNone}},
+    {"bloom_bits_per_key",
+     {offsetof(struct PlainTableOptions, bloom_bits_per_key), OptionType::kInt,
+      OptionVerificationType::kNormal, OptionTypeFlags::kNone}},
+    {"hash_table_ratio",
+     {offsetof(struct PlainTableOptions, hash_table_ratio), OptionType::kDouble,
+      OptionVerificationType::kNormal, OptionTypeFlags::kNone}},
+    {"index_sparseness",
+     {offsetof(struct PlainTableOptions, index_sparseness), OptionType::kSizeT,
+      OptionVerificationType::kNormal, OptionTypeFlags::kNone}},
+    {"huge_page_tlb_size",
+     {offsetof(struct PlainTableOptions, huge_page_tlb_size),
+      OptionType::kSizeT, OptionVerificationType::kNormal,
+      OptionTypeFlags::kNone}},
+    {"encoding_type",
+     {offsetof(struct PlainTableOptions, encoding_type),
+      OptionType::kEncodingType, OptionVerificationType::kNormal,
+      OptionTypeFlags::kNone}},
+    {"full_scan_mode",
+     {offsetof(struct PlainTableOptions, full_scan_mode), OptionType::kBoolean,
+      OptionVerificationType::kNormal, OptionTypeFlags::kNone}},
+    {"store_index_in_file",
+     {offsetof(struct PlainTableOptions, store_index_in_file),
+      OptionType::kBoolean, OptionVerificationType::kNormal,
+      OptionTypeFlags::kNone}},
+};
+
+PlainTableFactory::PlainTableFactory(const PlainTableOptions& options)
+    : table_options_(options) {
+  ConfigurableHelper::RegisterOptions(*this, &table_options_,
+                                      &plain_table_type_info);
+}
+
+Status PlainTableFactory::NewTableReader(
+    const ReadOptions& /*ro*/, const TableReaderOptions& table_reader_options,
+    std::unique_ptr<RandomAccessFileReader>&& file, uint64_t file_size,
+    std::unique_ptr<TableReader>* table,
+    bool /*prefetch_index_and_filter_in_cache*/) const {
+  return PlainTableReader::Open(
+      table_reader_options.ioptions, table_reader_options.env_options,
+      table_reader_options.internal_comparator, std::move(file), file_size,
+      table, table_options_.bloom_bits_per_key, table_options_.hash_table_ratio,
+      table_options_.index_sparseness, table_options_.huge_page_tlb_size,
+      table_options_.full_scan_mode, table_reader_options.immortal,
+      table_reader_options.prefix_extractor);
+}
+
+TableBuilder* PlainTableFactory::NewTableBuilder(
+    const TableBuilderOptions& table_builder_options, uint32_t column_family_id,
+    WritableFileWriter* file) const {
+  // Ignore the skip_filters flag. PlainTable format is optimized for small
+  // in-memory dbs. The skip_filters optimization is not useful for plain
+  // tables
+  //
+  return new PlainTableBuilder(
+      table_builder_options.ioptions, table_builder_options.moptions,
+      table_builder_options.int_tbl_prop_collector_factories, column_family_id,
+      file, table_options_.user_key_len, table_options_.encoding_type,
+      table_options_.index_sparseness, table_options_.bloom_bits_per_key,
+      table_builder_options.column_family_name, 6,
+      table_options_.huge_page_tlb_size, table_options_.hash_table_ratio,
+      table_options_.store_index_in_file, table_builder_options.db_id,
+      table_builder_options.db_session_id);
+}
+
+std::string PlainTableFactory::GetPrintableOptions() const {
+  std::string ret;
+  ret.reserve(20000);
+  const int kBufferSize = 200;
+  char buffer[kBufferSize];
+
+  snprintf(buffer, kBufferSize, "  user_key_len: %u\n",
+           table_options_.user_key_len);
+  ret.append(buffer);
+  snprintf(buffer, kBufferSize, "  bloom_bits_per_key: %d\n",
+           table_options_.bloom_bits_per_key);
+  ret.append(buffer);
+  snprintf(buffer, kBufferSize, "  hash_table_ratio: %lf\n",
+           table_options_.hash_table_ratio);
+  ret.append(buffer);
+  snprintf(buffer, kBufferSize, "  index_sparseness: %" ROCKSDB_PRIszt "\n",
+           table_options_.index_sparseness);
+  ret.append(buffer);
+  snprintf(buffer, kBufferSize, "  huge_page_tlb_size: %" ROCKSDB_PRIszt "\n",
+           table_options_.huge_page_tlb_size);
+  ret.append(buffer);
+  snprintf(buffer, kBufferSize, "  encoding_type: %d\n",
+           table_options_.encoding_type);
+  ret.append(buffer);
+  snprintf(buffer, kBufferSize, "  full_scan_mode: %d\n",
+           table_options_.full_scan_mode);
+  ret.append(buffer);
+  snprintf(buffer, kBufferSize, "  store_index_in_file: %d\n",
+           table_options_.store_index_in_file);
+  ret.append(buffer);
+  return ret;
+}
+
+Status GetPlainTableOptionsFromString(const PlainTableOptions& table_options,
+                                      const std::string& opts_str,
+                                      PlainTableOptions* new_table_options) {
+  ConfigOptions config_options;
+  config_options.input_strings_escaped = false;
+  config_options.ignore_unknown_options = false;
+  config_options.invoke_prepare_options = false;
+  return GetPlainTableOptionsFromString(config_options, table_options, opts_str,
+                                        new_table_options);
+}
+
+Status GetPlainTableOptionsFromString(const ConfigOptions& config_options,
+                                      const PlainTableOptions& table_options,
+                                      const std::string& opts_str,
+                                      PlainTableOptions* new_table_options) {
+  std::unordered_map<std::string, std::string> opts_map;
+  Status s = StringToMap(opts_str, &opts_map);
+  if (!s.ok()) {
+    return s;
+  }
+
+  s = GetPlainTableOptionsFromMap(config_options, table_options, opts_map,
+                                  new_table_options);
+  // Translate any errors (NotFound, NotSupported, to InvalidArgument
+  if (s.ok() || s.IsInvalidArgument()) {
+    return s;
+  } else {
+    return Status::InvalidArgument(s.getState());
+  }
+}
+
+Status GetMemTableRepFactoryFromString(
+    const std::string& opts_str,
+    std::unique_ptr<MemTableRepFactory>* new_mem_factory) {
+  std::vector<std::string> opts_list = StringSplit(opts_str, ':');
+  size_t len = opts_list.size();
+
+  if (opts_list.empty() || opts_list.size() > 2) {
+    return Status::InvalidArgument("Can't parse memtable_factory option ",
+                                   opts_str);
+  }
+
+  MemTableRepFactory* mem_factory = nullptr;
+
+  if (opts_list[0] == "skip_list" || opts_list[0] == "SkipListFactory") {
+    // Expecting format
+    // skip_list:<lookahead>
+    if (2 == len) {
+      size_t lookahead = ParseSizeT(opts_list[1]);
+      mem_factory = new SkipListFactory(lookahead);
+    } else if (1 == len) {
+      mem_factory = new SkipListFactory();
+    }
+  } else if (opts_list[0] == "prefix_hash" ||
+             opts_list[0] == "HashSkipListRepFactory") {
+    // Expecting format
+    // prfix_hash:<hash_bucket_count>
+    if (2 == len) {
+      size_t hash_bucket_count = ParseSizeT(opts_list[1]);
+      mem_factory = NewHashSkipListRepFactory(hash_bucket_count);
+    } else if (1 == len) {
+      mem_factory = NewHashSkipListRepFactory();
+    }
+  } else if (opts_list[0] == "hash_linkedlist" ||
+             opts_list[0] == "HashLinkListRepFactory") {
+    // Expecting format
+    // hash_linkedlist:<hash_bucket_count>
+    if (2 == len) {
+      size_t hash_bucket_count = ParseSizeT(opts_list[1]);
+      mem_factory = NewHashLinkListRepFactory(hash_bucket_count);
+    } else if (1 == len) {
+      mem_factory = NewHashLinkListRepFactory();
+    }
+  } else if (opts_list[0] == "vector" || opts_list[0] == "VectorRepFactory") {
+    // Expecting format
+    // vector:<count>
+    if (2 == len) {
+      size_t count = ParseSizeT(opts_list[1]);
+      mem_factory = new VectorRepFactory(count);
+    } else if (1 == len) {
+      mem_factory = new VectorRepFactory();
+    }
+  } else if (opts_list[0] == "cuckoo") {
+    return Status::NotSupported(
+        "cuckoo hash memtable is not supported anymore.");
+  } else {
+    return Status::InvalidArgument("Unrecognized memtable_factory option ",
+                                   opts_str);
+  }
+
+  if (mem_factory != nullptr) {
+    new_mem_factory->reset(mem_factory);
+  }
+
+  return Status::OK();
+}
+
+Status GetPlainTableOptionsFromMap(
+    const PlainTableOptions& table_options,
+    const std::unordered_map<std::string, std::string>& opts_map,
+    PlainTableOptions* new_table_options, bool input_strings_escaped,
+    bool ignore_unknown_options) {
+  ConfigOptions config_options;
+  config_options.input_strings_escaped = input_strings_escaped;
+  config_options.ignore_unknown_options = ignore_unknown_options;
+  return GetPlainTableOptionsFromMap(config_options, table_options, opts_map,
+                                     new_table_options);
+}
+
+Status GetPlainTableOptionsFromMap(
+    const ConfigOptions& config_options, const PlainTableOptions& table_options,
+    const std::unordered_map<std::string, std::string>& opts_map,
+    PlainTableOptions* new_table_options) {
+  assert(new_table_options);
+  PlainTableFactory ptf(table_options);
+  Status s = ptf.ConfigureFromMap(config_options, opts_map);
+  if (s.ok()) {
+    *new_table_options = *(ptf.GetOptions<PlainTableOptions>());
+  } else {
+    // Restore "new_options" to the default "base_options".
+    *new_table_options = table_options;
+  }
+  return s;
+}
+
+extern TableFactory* NewPlainTableFactory(const PlainTableOptions& options) {
+  return new PlainTableFactory(options);
+}
+
+const std::string PlainTablePropertyNames::kEncodingType =
+    "rocksdb.plain.table.encoding.type";
+
+const std::string PlainTablePropertyNames::kBloomVersion =
+    "rocksdb.plain.table.bloom.version";
+
+const std::string PlainTablePropertyNames::kNumBloomBlocks =
+    "rocksdb.plain.table.bloom.numblocks";
+
+}  // namespace ROCKSDB_NAMESPACE
+#endif  // ROCKSDB_LITE

package/deps/rocksdb/rocksdb/table/plain/plain_table_factory.h

@@ -0,0 +1,182 @@
+// Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved.
+// 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.
+
+#pragma once
+
+#ifndef ROCKSDB_LITE
+#include <memory>
+#include <string>
+#include <stdint.h>
+
+#include "rocksdb/table.h"
+
+namespace ROCKSDB_NAMESPACE {
+
+struct EnvOptions;
+
+class Status;
+class RandomAccessFile;
+class WritableFile;
+class Table;
+class TableBuilder;
+
+// PlainTableFactory is the entrance function to the PlainTable format of
+// SST files. It returns instances PlainTableBuilder as the builder
+// class and PlainTableReader as the reader class, where the format is
+// actually implemented.
+//
+// The PlainTable is designed for memory-mapped file systems, e.g. tmpfs.
+// Data is not organized in blocks, which allows fast access. Because of
+// following downsides
+// 1. Data compression is not supported.
+// 2. Data is not checksumed.
+// it is not recommended to use this format on other type of file systems.
+//
+// PlainTable requires fixed length key, configured as a constructor
+// parameter of the factory class. Output file format:
+// +-------------+-----------------+
+// | version     | user_key_length |
+// +------------++------------+-----------------+  <= key1 offset
+// |  encoded key1            | value_size  |   |
+// +------------+-------------+-------------+   |
+// | value1                                     |
+// |                                            |
+// +--------------------------+-------------+---+  <= key2 offset
+// | encoded key2             | value_size  |   |
+// +------------+-------------+-------------+   |
+// | value2                                     |
+// |                                            |
+// |        ......                              |
+// +-----------------+--------------------------+
+//
+// When the key encoding type is kPlain. Key part is encoded as:
+// +------------+--------------------+
+// | [key_size] |  internal key      |
+// +------------+--------------------+
+// for the case of user_key_len = kPlainTableVariableLength case,
+// and simply:
+// +----------------------+
+// |  internal key        |
+// +----------------------+
+// for user_key_len != kPlainTableVariableLength case.
+//
+// If key encoding type is kPrefix. Keys are encoding in this format.
+// There are three ways to encode a key:
+// (1) Full Key
+// +---------------+---------------+-------------------+
+// | Full Key Flag | Full Key Size | Full Internal Key |
+// +---------------+---------------+-------------------+
+// which simply encodes a full key
+//
+// (2) A key shared the same prefix as the previous key, which is encoded as
+//     format of (1).
+// +-------------+-------------+-------------+-------------+------------+
+// | Prefix Flag | Prefix Size | Suffix Flag | Suffix Size | Key Suffix |
+// +-------------+-------------+-------------+-------------+------------+
+// where key is the suffix part of the key, including the internal bytes.
+// the actual key will be constructed by concatenating prefix part of the
+// previous key, with the suffix part of the key here, with sizes given here.
+//
+// (3) A key shared the same prefix as the previous key, which is encoded as
+//     the format of (2).
+// +-----------------+-----------------+------------------------+
+// | Key Suffix Flag | Key Suffix Size | Suffix of Internal Key |
+// +-----------------+-----------------+------------------------+
+// The key will be constructed by concatenating previous key's prefix (which is
+// also a prefix which the last key encoded in the format of (1)) and the
+// key given here.
+//
+// For example, we for following keys (prefix and suffix are separated by
+// spaces):
+//   0000 0001
+//   0000 00021
+//   0000 0002
+//   00011 00
+//   0002 0001
+// Will be encoded like this:
+//   FK 8 00000001
+//   PF 4 SF 5 00021
+//   SF 4 0002
+//   FK 7 0001100
+//   FK 8 00020001
+// (where FK means full key flag, PF means prefix flag and SF means suffix flag)
+//
+// All those "key flag + key size" shown above are in this format:
+// The 8 bits of the first byte:
+// +----+----+----+----+----+----+----+----+
+// |  Type   |            Size             |
+// +----+----+----+----+----+----+----+----+
+// Type indicates: full key, prefix, or suffix.
+// The last 6 bits are for size. If the size bits are not all 1, it means the
+// size of the key. Otherwise, varint32 is read after this byte. This varint
+// value + 0x3F (the value of all 1) will be the key size.
+//
+// For example, full key with length 16 will be encoded as (binary):
+//     00 010000
+// (00 means full key)
+// and a prefix with 100 bytes will be encoded as:
+//     01 111111    00100101
+//         (63)       (37)
+// (01 means key suffix)
+//
+// All the internal keys above (including kPlain and kPrefix) are encoded in
+// this format:
+// There are two types:
+// (1) normal internal key format
+// +----------- ...... -------------+----+---+---+---+---+---+---+---+
+// |       user key                 |type|      sequence ID          |
+// +----------- ..... --------------+----+---+---+---+---+---+---+---+
+// (2) Special case for keys whose sequence ID is 0 and is value type
+// +----------- ...... -------------+----+
+// |       user key                 |0x80|
+// +----------- ..... --------------+----+
+// To save 7 bytes for the special case where sequence ID = 0.
+//
+//
+class PlainTableFactory : public TableFactory {
+ public:
+  ~PlainTableFactory() {}
+  // user_key_len is the length of the user key. If it is set to be
+  // kPlainTableVariableLength, then it means variable length. Otherwise, all
+  // the keys need to have the fix length of this value. bloom_bits_per_key is
+  // number of bits used for bloom filer per key. hash_table_ratio is
+  // the desired utilization of the hash table used for prefix hashing.
+  // hash_table_ratio = number of prefixes / #buckets in the hash table
+  // hash_table_ratio = 0 means skip hash table but only replying on binary
+  // search.
+  // index_sparseness determines index interval for keys
+  // inside the same prefix. It will be the maximum number of linear search
+  // required after hash and binary search.
+  // index_sparseness = 0 means index for every key.
+  // huge_page_tlb_size determines whether to allocate hash indexes from huge
+  // page TLB and the page size if allocating from there. See comments of
+  // Arena::AllocateAligned() for details.
+  explicit PlainTableFactory(
+      const PlainTableOptions& _table_options = PlainTableOptions());
+
+  // Method to allow CheckedCast to work for this class
+  static const char* kClassName() { return kPlainTableName(); }
+  const char* Name() const override { return kPlainTableName(); }
+  using TableFactory::NewTableReader;
+  Status NewTableReader(const ReadOptions& ro,
+                        const TableReaderOptions& table_reader_options,
+                        std::unique_ptr<RandomAccessFileReader>&& file,
+                        uint64_t file_size, std::unique_ptr<TableReader>* table,
+                        bool prefetch_index_and_filter_in_cache) const override;
+
+  TableBuilder* NewTableBuilder(
+      const TableBuilderOptions& table_builder_options,
+      uint32_t column_family_id, WritableFileWriter* file) const override;
+
+  std::string GetPrintableOptions() const override;
+  static const char kValueTypeSeqId0 = char(~0);
+
+ private:
+  PlainTableOptions table_options_;
+};
+
+
+}  // namespace ROCKSDB_NAMESPACE
+#endif  // ROCKSDB_LITE

package/deps/rocksdb/rocksdb/table/plain/plain_table_index.cc

@@ -0,0 +1,211 @@
+//  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).
+
+#ifndef ROCKSDB_LITE
+
+#include <cinttypes>
+
+#include "table/plain/plain_table_index.h"
+#include "util/coding.h"
+#include "util/hash.h"
+
+namespace ROCKSDB_NAMESPACE {
+
+namespace {
+inline uint32_t GetBucketIdFromHash(uint32_t hash, uint32_t num_buckets) {
+  assert(num_buckets > 0);
+  return hash % num_buckets;
+}
+}
+
+Status PlainTableIndex::InitFromRawData(Slice data) {
+  if (!GetVarint32(&data, &index_size_)) {
+    return Status::Corruption("Couldn't read the index size!");
+  }
+  assert(index_size_ > 0);
+  if (!GetVarint32(&data, &num_prefixes_)) {
+    return Status::Corruption("Couldn't read the index size!");
+  }
+  sub_index_size_ =
+      static_cast<uint32_t>(data.size()) - index_size_ * kOffsetLen;
+
+  char* index_data_begin = const_cast<char*>(data.data());
+  index_ = reinterpret_cast<uint32_t*>(index_data_begin);
+  sub_index_ = reinterpret_cast<char*>(index_ + index_size_);
+  return Status::OK();
+}
+
+PlainTableIndex::IndexSearchResult PlainTableIndex::GetOffset(
+    uint32_t prefix_hash, uint32_t* bucket_value) const {
+  int bucket = GetBucketIdFromHash(prefix_hash, index_size_);
+  GetUnaligned(index_ + bucket, bucket_value);
+  if ((*bucket_value & kSubIndexMask) == kSubIndexMask) {
+    *bucket_value ^= kSubIndexMask;
+    return kSubindex;
+  }
+  if (*bucket_value >= kMaxFileSize) {
+    return kNoPrefixForBucket;
+  } else {
+    // point directly to the file
+    return kDirectToFile;
+  }
+}
+
+void PlainTableIndexBuilder::IndexRecordList::AddRecord(uint32_t hash,
+                                                        uint32_t offset) {
+  if (num_records_in_current_group_ == kNumRecordsPerGroup) {
+    current_group_ = AllocateNewGroup();
+    num_records_in_current_group_ = 0;
+  }
+  auto& new_record = current_group_[num_records_in_current_group_++];
+  new_record.hash = hash;
+  new_record.offset = offset;
+  new_record.next = nullptr;
+}
+
+void PlainTableIndexBuilder::AddKeyPrefix(Slice key_prefix_slice,
+                                          uint32_t key_offset) {
+  if (is_first_record_ || prev_key_prefix_ != key_prefix_slice.ToString()) {
+    ++num_prefixes_;
+    if (!is_first_record_) {
+      keys_per_prefix_hist_.Add(num_keys_per_prefix_);
+    }
+    num_keys_per_prefix_ = 0;
+    prev_key_prefix_ = key_prefix_slice.ToString();
+    prev_key_prefix_hash_ = GetSliceHash(key_prefix_slice);
+    due_index_ = true;
+  }
+
+  if (due_index_) {
+    // Add an index key for every kIndexIntervalForSamePrefixKeys keys
+    record_list_.AddRecord(prev_key_prefix_hash_, key_offset);
+    due_index_ = false;
+  }
+
+  num_keys_per_prefix_++;
+  if (index_sparseness_ == 0 || num_keys_per_prefix_ % index_sparseness_ == 0) {
+    due_index_ = true;
+  }
+  is_first_record_ = false;
+}
+
+Slice PlainTableIndexBuilder::Finish() {
+  AllocateIndex();
+  std::vector<IndexRecord*> hash_to_offsets(index_size_, nullptr);
+  std::vector<uint32_t> entries_per_bucket(index_size_, 0);
+  BucketizeIndexes(&hash_to_offsets, &entries_per_bucket);
+
+  keys_per_prefix_hist_.Add(num_keys_per_prefix_);
+  ROCKS_LOG_INFO(ioptions_.info_log, "Number of Keys per prefix Histogram: %s",
+                 keys_per_prefix_hist_.ToString().c_str());
+
+  // From the temp data structure, populate indexes.
+  return FillIndexes(hash_to_offsets, entries_per_bucket);
+}
+
+void PlainTableIndexBuilder::AllocateIndex() {
+  if (prefix_extractor_ == nullptr || hash_table_ratio_ <= 0) {
+    // Fall back to pure binary search if the user fails to specify a prefix
+    // extractor.
+    index_size_ = 1;
+  } else {
+    double hash_table_size_multipier = 1.0 / hash_table_ratio_;
+    index_size_ =
+      static_cast<uint32_t>(num_prefixes_ * hash_table_size_multipier) + 1;
+    assert(index_size_ > 0);
+  }
+}
+
+void PlainTableIndexBuilder::BucketizeIndexes(
+    std::vector<IndexRecord*>* hash_to_offsets,
+    std::vector<uint32_t>* entries_per_bucket) {
+  bool first = true;
+  uint32_t prev_hash = 0;
+  size_t num_records = record_list_.GetNumRecords();
+  for (size_t i = 0; i < num_records; i++) {
+    IndexRecord* index_record = record_list_.At(i);
+    uint32_t cur_hash = index_record->hash;
+    if (first || prev_hash != cur_hash) {
+      prev_hash = cur_hash;
+      first = false;
+    }
+    uint32_t bucket = GetBucketIdFromHash(cur_hash, index_size_);
+    IndexRecord* prev_bucket_head = (*hash_to_offsets)[bucket];
+    index_record->next = prev_bucket_head;
+    (*hash_to_offsets)[bucket] = index_record;
+    (*entries_per_bucket)[bucket]++;
+  }
+
+  sub_index_size_ = 0;
+  for (auto entry_count : *entries_per_bucket) {
+    if (entry_count <= 1) {
+      continue;
+    }
+    // Only buckets with more than 1 entry will have subindex.
+    sub_index_size_ += VarintLength(entry_count);
+    // total bytes needed to store these entries' in-file offsets.
+    sub_index_size_ += entry_count * PlainTableIndex::kOffsetLen;
+  }
+}
+
+Slice PlainTableIndexBuilder::FillIndexes(
+    const std::vector<IndexRecord*>& hash_to_offsets,
+    const std::vector<uint32_t>& entries_per_bucket) {
+  ROCKS_LOG_DEBUG(ioptions_.info_log,
+                  "Reserving %" PRIu32 " bytes for plain table's sub_index",
+                  sub_index_size_);
+  auto total_allocate_size = GetTotalSize();
+  char* allocated = arena_->AllocateAligned(
+      total_allocate_size, huge_page_tlb_size_, ioptions_.info_log);
+
+  auto temp_ptr = EncodeVarint32(allocated, index_size_);
+  uint32_t* index =
+      reinterpret_cast<uint32_t*>(EncodeVarint32(temp_ptr, num_prefixes_));
+  char* sub_index = reinterpret_cast<char*>(index + index_size_);
+
+  uint32_t sub_index_offset = 0;
+  for (uint32_t i = 0; i < index_size_; i++) {
+    uint32_t num_keys_for_bucket = entries_per_bucket[i];
+    switch (num_keys_for_bucket) {
+      case 0:
+        // No key for bucket
+        PutUnaligned(index + i, (uint32_t)PlainTableIndex::kMaxFileSize);
+        break;
+      case 1:
+        // point directly to the file offset
+        PutUnaligned(index + i, hash_to_offsets[i]->offset);
+        break;
+      default:
+        // point to second level indexes.
+        PutUnaligned(index + i, sub_index_offset | PlainTableIndex::kSubIndexMask);
+        char* prev_ptr = &sub_index[sub_index_offset];
+        char* cur_ptr = EncodeVarint32(prev_ptr, num_keys_for_bucket);
+        sub_index_offset += static_cast<uint32_t>(cur_ptr - prev_ptr);
+        char* sub_index_pos = &sub_index[sub_index_offset];
+        IndexRecord* record = hash_to_offsets[i];
+        int j;
+        for (j = num_keys_for_bucket - 1; j >= 0 && record;
+             j--, record = record->next) {
+          EncodeFixed32(sub_index_pos + j * sizeof(uint32_t), record->offset);
+        }
+        assert(j == -1 && record == nullptr);
+        sub_index_offset += PlainTableIndex::kOffsetLen * num_keys_for_bucket;
+        assert(sub_index_offset <= sub_index_size_);
+        break;
+    }
+  }
+  assert(sub_index_offset == sub_index_size_);
+
+  ROCKS_LOG_DEBUG(ioptions_.info_log,
+                  "hash table size: %" PRIu32 ", suffix_map length %" PRIu32,
+                  index_size_, sub_index_size_);
+  return Slice(allocated, GetTotalSize());
+}
+
+const std::string PlainTableIndexBuilder::kPlainTableIndexBlock =
+    "PlainTableIndexBlock";
+};  // namespace ROCKSDB_NAMESPACE
+
+#endif  // ROCKSDB_LITE