Revision 239d17a19c3cec16937aa4b6c56c90f4f217addf authored by Peter Dillinger on 18 December 2020, 22:29:48 UTC, committed by Facebook GitHub Bot on 18 December 2020, 22:31:03 UTC
Summary:
Primarily this change refactors the optimize_filters_for_memory
code for Bloom filters, based on malloc_usable_size, to also work for
Ribbon filters.

This change also replaces the somewhat slow but general
BuiltinFilterBitsBuilder::ApproximateNumEntries with
implementation-specific versions for Ribbon (new) and Legacy Bloom
(based on a recently deleted version). The reason is to emphasize
speed in ApproximateNumEntries rather than 100% accuracy.

Justification: ApproximateNumEntries (formerly CalculateNumEntry) is
only used by RocksDB for range-partitioned filters, called each time we
start to construct one. (In theory, it should be possible to reuse the
estimate, but the abstractions provided by FilterPolicy don't really
make that workable.) But this is only used as a heuristic estimate for
hitting a desired partitioned filter size because of alignment to data
blocks, which have various numbers of unique keys or prefixes. The two
factors lead us to prioritize reasonable speed over 100% accuracy.

optimize_filters_for_memory adds extra complication, because precisely
calculating num_entries for some allowed number of bytes depends on state
with optimize_filters_for_memory enabled. And the allocator-agnostic
implementation of optimize_filters_for_memory, using malloc_usable_size,
means we would have to actually allocate memory, many times, just to
precisely determine how many entries (keys) could be added and stay below
some size budget, for the current state. (In a draft, I got this
working, and then realized the balance of speed vs. accuracy was all
wrong.)

So related to that, I have made CalculateSpace, an internal-only API
only used for testing, non-authoritative also if
optimize_filters_for_memory is enabled. This simplifies some code.

Pull Request resolved: https://github.com/facebook/rocksdb/pull/7774

Test Plan:
unit test updated, and for FilterSize test, range of tested
values is greatly expanded (still super fast)

Also tested `db_bench -benchmarks=fillrandom,stats -bloom_bits=10 -num=1000000 -partition_index_and_filters -format_version=5 [-optimize_filters_for_memory] [-use_ribbon_filter]` with temporary debug output of generated filter sizes.

Bloom+optimize_filters_for_memory:

      1 Filter size: 197 (224 in memory)
    134 Filter size: 3525 (3584 in memory)
    107 Filter size: 4037 (4096 in memory)
    Total on disk: 904,506
    Total in memory: 918,752

Ribbon+optimize_filters_for_memory:

      1 Filter size: 3061 (3072 in memory)
    110 Filter size: 3573 (3584 in memory)
     58 Filter size: 4085 (4096 in memory)
    Total on disk: 633,021 (-30.0%)
    Total in memory: 634,880 (-30.9%)

Bloom (no offm):

      1 Filter size: 261 (320 in memory)
      1 Filter size: 3333 (3584 in memory)
    240 Filter size: 3717 (4096 in memory)
    Total on disk: 895,674 (-1% on disk vs. +offm; known tolerable overhead of offm)
    Total in memory: 986,944 (+7.4% vs. +offm)

Ribbon (no offm):

      1 Filter size: 2949 (3072 in memory)
      1 Filter size: 3381 (3584 in memory)
    167 Filter size: 3701 (4096 in memory)
    Total on disk: 624,397 (-30.3% vs. Bloom)
    Total in memory: 690,688 (-30.0% vs. Bloom)

Note that optimize_filters_for_memory is even more effective for Ribbon filter than for cache-local Bloom, because it can close the unused memory gap even tighter than Bloom filter, because of 16 byte increments for Ribbon vs. 64 byte increments for Bloom.

Reviewed By: jay-zhuang

Differential Revision: D25592970

Pulled By: pdillinger

fbshipit-source-id: 606fdaa025bb790d7e9c21601e8ea86e10541912
1 parent 04b3524
Raw File
table_reader_bench.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).

#ifndef GFLAGS
#include <cstdio>
int main() {
  fprintf(stderr, "Please install gflags to run rocksdb tools\n");
  return 1;
}
#else

#include "db/db_impl/db_impl.h"
#include "db/dbformat.h"
#include "env/composite_env_wrapper.h"
#include "file/random_access_file_reader.h"
#include "monitoring/histogram.h"
#include "rocksdb/db.h"
#include "rocksdb/slice_transform.h"
#include "rocksdb/table.h"
#include "table/block_based/block_based_table_factory.h"
#include "table/get_context.h"
#include "table/internal_iterator.h"
#include "table/plain/plain_table_factory.h"
#include "table/table_builder.h"
#include "test_util/testharness.h"
#include "test_util/testutil.h"
#include "util/gflags_compat.h"

using GFLAGS_NAMESPACE::ParseCommandLineFlags;
using GFLAGS_NAMESPACE::SetUsageMessage;

namespace ROCKSDB_NAMESPACE {

namespace {
// Make a key that i determines the first 4 characters and j determines the
// last 4 characters.
static std::string MakeKey(int i, int j, bool through_db) {
  char buf[100];
  snprintf(buf, sizeof(buf), "%04d__key___%04d", i, j);
  if (through_db) {
    return std::string(buf);
  }
  // If we directly query table, which operates on internal keys
  // instead of user keys, we need to add 8 bytes of internal
  // information (row type etc) to user key to make an internal
  // key.
  InternalKey key(std::string(buf), 0, ValueType::kTypeValue);
  return key.Encode().ToString();
}

uint64_t Now(Env* env, bool measured_by_nanosecond) {
  return measured_by_nanosecond ? env->NowNanos() : env->NowMicros();
}
}  // namespace

// A very simple benchmark that.
// Create a table with roughly numKey1 * numKey2 keys,
// where there are numKey1 prefixes of the key, each has numKey2 number of
// distinguished key, differing in the suffix part.
// If if_query_empty_keys = false, query the existing keys numKey1 * numKey2
// times randomly.
// If if_query_empty_keys = true, query numKey1 * numKey2 random empty keys.
// Print out the total time.
// If through_db=true, a full DB will be created and queries will be against
// it. Otherwise, operations will be directly through table level.
//
// If for_terator=true, instead of just query one key each time, it queries
// a range sharing the same prefix.
namespace {
void TableReaderBenchmark(Options& opts, EnvOptions& env_options,
                          ReadOptions& read_options, int num_keys1,
                          int num_keys2, int num_iter, int /*prefix_len*/,
                          bool if_query_empty_keys, bool for_iterator,
                          bool through_db, bool measured_by_nanosecond) {
  ROCKSDB_NAMESPACE::InternalKeyComparator ikc(opts.comparator);

  std::string file_name =
      test::PerThreadDBPath("rocksdb_table_reader_benchmark");
  std::string dbname = test::PerThreadDBPath("rocksdb_table_reader_bench_db");
  WriteOptions wo;
  Env* env = Env::Default();
  TableBuilder* tb = nullptr;
  DB* db = nullptr;
  Status s;
  const ImmutableCFOptions ioptions(opts);
  const ColumnFamilyOptions cfo(opts);
  const MutableCFOptions moptions(cfo);
  std::unique_ptr<WritableFileWriter> file_writer;
  if (!through_db) {
    std::unique_ptr<WritableFile> file;
    env->NewWritableFile(file_name, &file, env_options);

    std::vector<std::unique_ptr<IntTblPropCollectorFactory> >
        int_tbl_prop_collector_factories;

    file_writer.reset(new WritableFileWriter(
        NewLegacyWritableFileWrapper(std::move(file)), file_name, env_options));
    int unknown_level = -1;
    tb = opts.table_factory->NewTableBuilder(
        TableBuilderOptions(
            ioptions, moptions, ikc, &int_tbl_prop_collector_factories,
            CompressionType::kNoCompression, 0 /* sample_for_compression */,
            CompressionOptions(), false /* skip_filters */,
            kDefaultColumnFamilyName, unknown_level),
        0 /* column_family_id */, file_writer.get());
  } else {
    s = DB::Open(opts, dbname, &db);
    ASSERT_OK(s);
    ASSERT_TRUE(db != nullptr);
  }
  // Populate slightly more than 1M keys
  for (int i = 0; i < num_keys1; i++) {
    for (int j = 0; j < num_keys2; j++) {
      std::string key = MakeKey(i * 2, j, through_db);
      if (!through_db) {
        tb->Add(key, key);
      } else {
        db->Put(wo, key, key);
      }
    }
  }
  if (!through_db) {
    tb->Finish();
    file_writer->Close();
  } else {
    db->Flush(FlushOptions());
  }

  std::unique_ptr<TableReader> table_reader;
  if (!through_db) {
    std::unique_ptr<RandomAccessFile> raf;
    s = env->NewRandomAccessFile(file_name, &raf, env_options);
    if (!s.ok()) {
      fprintf(stderr, "Create File Error: %s\n", s.ToString().c_str());
      exit(1);
    }
    uint64_t file_size;
    env->GetFileSize(file_name, &file_size);
    std::unique_ptr<RandomAccessFileReader> file_reader(
        new RandomAccessFileReader(NewLegacyRandomAccessFileWrapper(raf),
                                   file_name));
    s = opts.table_factory->NewTableReader(
        TableReaderOptions(ioptions, moptions.prefix_extractor.get(),
                           env_options, ikc),
        std::move(file_reader), file_size, &table_reader);
    if (!s.ok()) {
      fprintf(stderr, "Open Table Error: %s\n", s.ToString().c_str());
      exit(1);
    }
  }

  Random rnd(301);
  std::string result;
  HistogramImpl hist;

  for (int it = 0; it < num_iter; it++) {
    for (int i = 0; i < num_keys1; i++) {
      for (int j = 0; j < num_keys2; j++) {
        int r1 = rnd.Uniform(num_keys1) * 2;
        int r2 = rnd.Uniform(num_keys2);
        if (if_query_empty_keys) {
          r1++;
          r2 = num_keys2 * 2 - r2;
        }

        if (!for_iterator) {
          // Query one existing key;
          std::string key = MakeKey(r1, r2, through_db);
          uint64_t start_time = Now(env, measured_by_nanosecond);
          if (!through_db) {
            PinnableSlice value;
            MergeContext merge_context;
            SequenceNumber max_covering_tombstone_seq = 0;
            GetContext get_context(ioptions.user_comparator,
                                   ioptions.merge_operator, ioptions.info_log,
                                   ioptions.statistics, GetContext::kNotFound,
                                   Slice(key), &value, nullptr, &merge_context,
                                   true, &max_covering_tombstone_seq, env);
            s = table_reader->Get(read_options, key, &get_context, nullptr);
          } else {
            s = db->Get(read_options, key, &result);
          }
          hist.Add(Now(env, measured_by_nanosecond) - start_time);
        } else {
          int r2_len;
          if (if_query_empty_keys) {
            r2_len = 0;
          } else {
            r2_len = rnd.Uniform(num_keys2) + 1;
            if (r2_len + r2 > num_keys2) {
              r2_len = num_keys2 - r2;
            }
          }
          std::string start_key = MakeKey(r1, r2, through_db);
          std::string end_key = MakeKey(r1, r2 + r2_len, through_db);
          uint64_t total_time = 0;
          uint64_t start_time = Now(env, measured_by_nanosecond);
          Iterator* iter = nullptr;
          InternalIterator* iiter = nullptr;
          if (!through_db) {
            iiter = table_reader->NewIterator(
                read_options, /*prefix_extractor=*/nullptr, /*arena=*/nullptr,
                /*skip_filters=*/false, TableReaderCaller::kUncategorized);
          } else {
            iter = db->NewIterator(read_options);
          }
          int count = 0;
          for (through_db ? iter->Seek(start_key) : iiter->Seek(start_key);
               through_db ? iter->Valid() : iiter->Valid();
               through_db ? iter->Next() : iiter->Next()) {
            if (if_query_empty_keys) {
              break;
            }
            // verify key;
            total_time += Now(env, measured_by_nanosecond) - start_time;
            assert(Slice(MakeKey(r1, r2 + count, through_db)) ==
                   (through_db ? iter->key() : iiter->key()));
            start_time = Now(env, measured_by_nanosecond);
            if (++count >= r2_len) {
              break;
            }
          }
          if (count != r2_len) {
            fprintf(
                stderr, "Iterator cannot iterate expected number of entries. "
                "Expected %d but got %d\n", r2_len, count);
            assert(false);
          }
          delete iter;
          total_time += Now(env, measured_by_nanosecond) - start_time;
          hist.Add(total_time);
        }
      }
    }
  }

  fprintf(
      stderr,
      "==================================================="
      "====================================================\n"
      "InMemoryTableSimpleBenchmark: %20s   num_key1:  %5d   "
      "num_key2: %5d  %10s\n"
      "==================================================="
      "===================================================="
      "\nHistogram (unit: %s): \n%s",
      opts.table_factory->Name(), num_keys1, num_keys2,
      for_iterator ? "iterator" : (if_query_empty_keys ? "empty" : "non_empty"),
      measured_by_nanosecond ? "nanosecond" : "microsecond",
      hist.ToString().c_str());
  if (!through_db) {
    env->DeleteFile(file_name);
  } else {
    delete db;
    db = nullptr;
    DestroyDB(dbname, opts);
  }
}
}  // namespace
}  // namespace ROCKSDB_NAMESPACE

DEFINE_bool(query_empty, false, "query non-existing keys instead of existing "
            "ones.");
DEFINE_int32(num_keys1, 4096, "number of distinguish prefix of keys");
DEFINE_int32(num_keys2, 512, "number of distinguish keys for each prefix");
DEFINE_int32(iter, 3, "query non-existing keys instead of existing ones");
DEFINE_int32(prefix_len, 16, "Prefix length used for iterators and indexes");
DEFINE_bool(iterator, false, "For test iterator");
DEFINE_bool(through_db, false, "If enable, a DB instance will be created and "
            "the query will be against DB. Otherwise, will be directly against "
            "a table reader.");
DEFINE_bool(mmap_read, true, "Whether use mmap read");
DEFINE_string(table_factory, "block_based",
              "Table factory to use: `block_based` (default), `plain_table` or "
              "`cuckoo_hash`.");
DEFINE_string(time_unit, "microsecond",
              "The time unit used for measuring performance. User can specify "
              "`microsecond` (default) or `nanosecond`");

int main(int argc, char** argv) {
  SetUsageMessage(std::string("\nUSAGE:\n") + std::string(argv[0]) +
                  " [OPTIONS]...");
  ParseCommandLineFlags(&argc, &argv, true);

  std::shared_ptr<ROCKSDB_NAMESPACE::TableFactory> tf;
  ROCKSDB_NAMESPACE::Options options;
  if (FLAGS_prefix_len < 16) {
    options.prefix_extractor.reset(
        ROCKSDB_NAMESPACE::NewFixedPrefixTransform(FLAGS_prefix_len));
  }
  ROCKSDB_NAMESPACE::ReadOptions ro;
  ROCKSDB_NAMESPACE::EnvOptions env_options;
  options.create_if_missing = true;
  options.compression = ROCKSDB_NAMESPACE::CompressionType::kNoCompression;

  if (FLAGS_table_factory == "cuckoo_hash") {
#ifndef ROCKSDB_LITE
    options.allow_mmap_reads = FLAGS_mmap_read;
    env_options.use_mmap_reads = FLAGS_mmap_read;
    ROCKSDB_NAMESPACE::CuckooTableOptions table_options;
    table_options.hash_table_ratio = 0.75;
    tf.reset(ROCKSDB_NAMESPACE::NewCuckooTableFactory(table_options));
#else
    fprintf(stderr, "Plain table is not supported in lite mode\n");
    exit(1);
#endif  // ROCKSDB_LITE
  } else if (FLAGS_table_factory == "plain_table") {
#ifndef ROCKSDB_LITE
    options.allow_mmap_reads = FLAGS_mmap_read;
    env_options.use_mmap_reads = FLAGS_mmap_read;

    ROCKSDB_NAMESPACE::PlainTableOptions plain_table_options;
    plain_table_options.user_key_len = 16;
    plain_table_options.bloom_bits_per_key = (FLAGS_prefix_len == 16) ? 0 : 8;
    plain_table_options.hash_table_ratio = 0.75;

    tf.reset(new ROCKSDB_NAMESPACE::PlainTableFactory(plain_table_options));
    options.prefix_extractor.reset(
        ROCKSDB_NAMESPACE::NewFixedPrefixTransform(FLAGS_prefix_len));
#else
    fprintf(stderr, "Cuckoo table is not supported in lite mode\n");
    exit(1);
#endif  // ROCKSDB_LITE
  } else if (FLAGS_table_factory == "block_based") {
    tf.reset(new ROCKSDB_NAMESPACE::BlockBasedTableFactory());
  } else {
    fprintf(stderr, "Invalid table type %s\n", FLAGS_table_factory.c_str());
  }

  if (tf) {
    // if user provides invalid options, just fall back to microsecond.
    bool measured_by_nanosecond = FLAGS_time_unit == "nanosecond";

    options.table_factory = tf;
    ROCKSDB_NAMESPACE::TableReaderBenchmark(
        options, env_options, ro, FLAGS_num_keys1, FLAGS_num_keys2, FLAGS_iter,
        FLAGS_prefix_len, FLAGS_query_empty, FLAGS_iterator, FLAGS_through_db,
        measured_by_nanosecond);
  } else {
    return 1;
  }

  return 0;
}

#endif  // GFLAGS
back to top