rocksdb/db/db_compaction_test.cc

//  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.
//
// 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.

#include "port/stack_trace.h"
#include "rocksdb/experimental.h"
#include "util/db_test_util.h"
#include "util/sync_point.h"
namespace rocksdb {

class DBCompactionTest : public DBTestBase {
 public:
  DBCompactionTest() : DBTestBase("/db_compaction_test") {}
};

namespace {
class OnFileDeletionListener : public EventListener {
 public:
  OnFileDeletionListener() :
      matched_count_(0),
      expected_file_name_("") {}

  void SetExpectedFileName(
      const std::string file_name) {
    expected_file_name_ = file_name;
  }

  void VerifyMatchedCount(size_t expected_value) {
    ASSERT_EQ(matched_count_, expected_value);
  }

  void OnTableFileDeleted(
      const TableFileDeletionInfo& info) override {
    if (expected_file_name_ != "") {
      ASSERT_EQ(expected_file_name_, info.file_path);
      expected_file_name_ = "";
      matched_count_++;
    }
  }

 private:
  size_t matched_count_;
  std::string expected_file_name_;
};

class SleepingBackgroundTask {
 public:
  SleepingBackgroundTask()
      : bg_cv_(&mutex_), should_sleep_(true), done_with_sleep_(false) {}
  void DoSleep() {
    MutexLock l(&mutex_);
    while (should_sleep_) {
      bg_cv_.Wait();
    }
    done_with_sleep_ = true;
    bg_cv_.SignalAll();
  }
  void WakeUp() {
    MutexLock l(&mutex_);
    should_sleep_ = false;
    bg_cv_.SignalAll();
  }
  void WaitUntilDone() {
    MutexLock l(&mutex_);
    while (!done_with_sleep_) {
      bg_cv_.Wait();
    }
  }
  bool WokenUp() {
    MutexLock l(&mutex_);
    return should_sleep_ == false;
  }

  void Reset() {
    MutexLock l(&mutex_);
    should_sleep_ = true;
    done_with_sleep_ = false;
  }

  static void DoSleepTask(void* arg) {
    reinterpret_cast<SleepingBackgroundTask*>(arg)->DoSleep();
  }

 private:
  port::Mutex mutex_;
  port::CondVar bg_cv_;  // Signalled when background work finishes
  bool should_sleep_;
  bool done_with_sleep_;
};

static const int kCDTValueSize = 1000;
static const int kCDTKeysPerBuffer = 4;
static const int kCDTNumLevels = 8;
Options DeletionTriggerOptions() {
  Options options;
  options.compression = kNoCompression;
  options.write_buffer_size = kCDTKeysPerBuffer * (kCDTValueSize + 24);
  options.min_write_buffer_number_to_merge = 1;
  options.max_write_buffer_number_to_maintain = 0;
  options.num_levels = kCDTNumLevels;
  options.level0_file_num_compaction_trigger = 1;
  options.target_file_size_base = options.write_buffer_size * 2;
  options.target_file_size_multiplier = 2;
  options.max_bytes_for_level_base =
      options.target_file_size_base * options.target_file_size_multiplier;
  options.max_bytes_for_level_multiplier = 2;
  options.disable_auto_compactions = false;
  return options;
}

bool HaveOverlappingKeyRanges(
    const Comparator* c,
    const SstFileMetaData& a, const SstFileMetaData& b) {
  if (c->Compare(a.smallestkey, b.smallestkey) >= 0) {
    if (c->Compare(a.smallestkey, b.largestkey) <= 0) {
      // b.smallestkey <= a.smallestkey <= b.largestkey
      return true;
    }
  } else if (c->Compare(a.largestkey, b.smallestkey) >= 0) {
    // a.smallestkey < b.smallestkey <= a.largestkey
    return true;
  }
  if (c->Compare(a.largestkey, b.largestkey) <= 0) {
    if (c->Compare(a.largestkey, b.smallestkey) >= 0) {
      // b.smallestkey <= a.largestkey <= b.largestkey
      return true;
    }
  } else if (c->Compare(a.smallestkey, b.largestkey) <= 0) {
    // a.smallestkey <= b.largestkey < a.largestkey
    return true;
  }
  return false;
}

// Identifies all files between level "min_level" and "max_level"
// which has overlapping key range with "input_file_meta".
void GetOverlappingFileNumbersForLevelCompaction(
    const ColumnFamilyMetaData& cf_meta,
    const Comparator* comparator,
    int min_level, int max_level,
    const SstFileMetaData* input_file_meta,
    std::set<std::string>* overlapping_file_names) {
  std::set<const SstFileMetaData*> overlapping_files;
  overlapping_files.insert(input_file_meta);
  for (int m = min_level; m <= max_level; ++m) {
    for (auto& file : cf_meta.levels[m].files) {
      for (auto* included_file : overlapping_files) {
        if (HaveOverlappingKeyRanges(
                comparator, *included_file, file)) {
          overlapping_files.insert(&file);
          overlapping_file_names->insert(file.name);
          break;
        }
      }
    }
  }
}

void VerifyCompactionResult(
    const ColumnFamilyMetaData& cf_meta,
    const std::set<std::string>& overlapping_file_numbers) {
#ifndef NDEBUG
  for (auto& level : cf_meta.levels) {
    for (auto& file : level.files) {
      assert(overlapping_file_numbers.find(file.name) ==
             overlapping_file_numbers.end());
    }
  }
#endif
}

const SstFileMetaData* PickFileRandomly(
    const ColumnFamilyMetaData& cf_meta,
    Random* rand,
    int* level = nullptr) {
  auto file_id = rand->Uniform(static_cast<int>(
      cf_meta.file_count)) + 1;
  for (auto& level_meta : cf_meta.levels) {
    if (file_id <= level_meta.files.size()) {
      if (level != nullptr) {
        *level = level_meta.level;
      }
      auto result = rand->Uniform(file_id);
      return &(level_meta.files[result]);
    }
    file_id -= level_meta.files.size();
  }
  assert(false);
  return nullptr;
}
}  // anonymous namespace

TEST_F(DBCompactionTest, CompactionDeletionTrigger) {
  for (int tid = 0; tid < 2; ++tid) {
    uint64_t db_size[2];
    Options options = CurrentOptions(DeletionTriggerOptions());

    if (tid == 1) {
      // second pass with universal compaction
      options.compaction_style = kCompactionStyleUniversal;
      options.num_levels = 1;
    }

    DestroyAndReopen(options);
    Random rnd(301);

    const int kTestSize = kCDTKeysPerBuffer * 512;
    std::vector<std::string> values;
    for (int k = 0; k < kTestSize; ++k) {
      values.push_back(RandomString(&rnd, kCDTValueSize));
      ASSERT_OK(Put(Key(k), values[k]));
    }
    dbfull()->TEST_WaitForFlushMemTable();
    dbfull()->TEST_WaitForCompact();
    db_size[0] = Size(Key(0), Key(kTestSize - 1));

    for (int k = 0; k < kTestSize; ++k) {
      ASSERT_OK(Delete(Key(k)));
    }
    dbfull()->TEST_WaitForFlushMemTable();
    dbfull()->TEST_WaitForCompact();
    db_size[1] = Size(Key(0), Key(kTestSize - 1));

    // must have much smaller db size.
    ASSERT_GT(db_size[0] / 3, db_size[1]);
  }
}

TEST_F(DBCompactionTest, CompactionDeletionTriggerReopen) {
  for (int tid = 0; tid < 2; ++tid) {
    uint64_t db_size[3];
    Options options = CurrentOptions(DeletionTriggerOptions());

    if (tid == 1) {
      // second pass with universal compaction
      options.compaction_style = kCompactionStyleUniversal;
      options.num_levels = 1;
    }

    DestroyAndReopen(options);
    Random rnd(301);

    // round 1 --- insert key/value pairs.
    const int kTestSize = kCDTKeysPerBuffer * 512;
    std::vector<std::string> values;
    for (int k = 0; k < kTestSize; ++k) {
      values.push_back(RandomString(&rnd, kCDTValueSize));
      ASSERT_OK(Put(Key(k), values[k]));
    }
    dbfull()->TEST_WaitForFlushMemTable();
    dbfull()->TEST_WaitForCompact();
    db_size[0] = Size(Key(0), Key(kTestSize - 1));
    Close();

    // round 2 --- disable auto-compactions and issue deletions.
    options.create_if_missing = false;
    options.disable_auto_compactions = true;
    Reopen(options);

    for (int k = 0; k < kTestSize; ++k) {
      ASSERT_OK(Delete(Key(k)));
    }
    db_size[1] = Size(Key(0), Key(kTestSize - 1));
    Close();
    // as auto_compaction is off, we shouldn't see too much reduce
    // in db size.
    ASSERT_LT(db_size[0] / 3, db_size[1]);

    // round 3 --- reopen db with auto_compaction on and see if
    // deletion compensation still work.
    options.disable_auto_compactions = false;
    Reopen(options);
    // insert relatively small amount of data to trigger auto compaction.
    for (int k = 0; k < kTestSize / 10; ++k) {
      ASSERT_OK(Put(Key(k), values[k]));
    }
    dbfull()->TEST_WaitForFlushMemTable();
    dbfull()->TEST_WaitForCompact();
    db_size[2] = Size(Key(0), Key(kTestSize - 1));
    // this time we're expecting significant drop in size.
    ASSERT_GT(db_size[0] / 3, db_size[2]);
  }
}

TEST_F(DBCompactionTest, CompactionTrigger) {
  Options options;
  options.write_buffer_size = 100 << 10;  // 100KB
  options.num_levels = 3;
  options.level0_file_num_compaction_trigger = 3;
  options = CurrentOptions(options);
  CreateAndReopenWithCF({"pikachu"}, options);

  Random rnd(301);

  for (int num = 0; num < options.level0_file_num_compaction_trigger - 1;
       num++) {
    std::vector<std::string> values;
    // Write 120KB (12 values, each 10K)
    for (int i = 0; i < 12; i++) {
      values.push_back(RandomString(&rnd, 10000));
      ASSERT_OK(Put(1, Key(i), values[i]));
    }
    dbfull()->TEST_WaitForFlushMemTable(handles_[1]);
    ASSERT_EQ(NumTableFilesAtLevel(0, 1), num + 1);
  }

  // generate one more file in level-0, and should trigger level-0 compaction
  std::vector<std::string> values;
  for (int i = 0; i < 12; i++) {
    values.push_back(RandomString(&rnd, 10000));
    ASSERT_OK(Put(1, Key(i), values[i]));
  }
  dbfull()->TEST_WaitForCompact();

  ASSERT_EQ(NumTableFilesAtLevel(0, 1), 0);
  ASSERT_EQ(NumTableFilesAtLevel(1, 1), 1);
}

TEST_F(DBCompactionTest, CompactionsGenerateMultipleFiles) {
  Options options;
  options.write_buffer_size = 100000000;        // Large write buffer
  options = CurrentOptions(options);
  CreateAndReopenWithCF({"pikachu"}, options);

  Random rnd(301);

  // Write 8MB (80 values, each 100K)
  ASSERT_EQ(NumTableFilesAtLevel(0, 1), 0);
  std::vector<std::string> values;
  for (int i = 0; i < 80; i++) {
    values.push_back(RandomString(&rnd, 100000));
    ASSERT_OK(Put(1, Key(i), values[i]));
  }

  // Reopening moves updates to level-0
  ReopenWithColumnFamilies({"default", "pikachu"}, options);
  dbfull()->TEST_CompactRange(0, nullptr, nullptr, handles_[1],
                              true /* disallow trivial move */);

  ASSERT_EQ(NumTableFilesAtLevel(0, 1), 0);
  ASSERT_GT(NumTableFilesAtLevel(1, 1), 1);
  for (int i = 0; i < 80; i++) {
    ASSERT_EQ(Get(1, Key(i)), values[i]);
  }
}

TEST_F(DBCompactionTest, MinorCompactionsHappen) {
  do {
    Options options;
    options.write_buffer_size = 10000;
    options = CurrentOptions(options);
    CreateAndReopenWithCF({"pikachu"}, options);

    const int N = 500;

    int starting_num_tables = TotalTableFiles(1);
    for (int i = 0; i < N; i++) {
      ASSERT_OK(Put(1, Key(i), Key(i) + std::string(1000, 'v')));
    }
    int ending_num_tables = TotalTableFiles(1);
    ASSERT_GT(ending_num_tables, starting_num_tables);

    for (int i = 0; i < N; i++) {
      ASSERT_EQ(Key(i) + std::string(1000, 'v'), Get(1, Key(i)));
    }

    ReopenWithColumnFamilies({"default", "pikachu"}, options);

    for (int i = 0; i < N; i++) {
      ASSERT_EQ(Key(i) + std::string(1000, 'v'), Get(1, Key(i)));
    }
  } while (ChangeCompactOptions());
}

// Check that writes done during a memtable compaction are recovered
// if the database is shutdown during the memtable compaction.
TEST_F(DBCompactionTest, RecoverDuringMemtableCompaction) {
  do {
    Options options;
    options.env = env_;
    options.write_buffer_size = 1000000;
    options = CurrentOptions(options);
    CreateAndReopenWithCF({"pikachu"}, options);

    // Trigger a long memtable compaction and reopen the database during it
    ASSERT_OK(Put(1, "foo", "v1"));  // Goes to 1st log file
    ASSERT_OK(Put(1, "big1", std::string(10000000, 'x')));  // Fills memtable
    ASSERT_OK(Put(1, "big2", std::string(1000, 'y')));  // Triggers compaction
    ASSERT_OK(Put(1, "bar", "v2"));                     // Goes to new log file

    ReopenWithColumnFamilies({"default", "pikachu"}, options);
    ASSERT_EQ("v1", Get(1, "foo"));
    ASSERT_EQ("v2", Get(1, "bar"));
    ASSERT_EQ(std::string(10000000, 'x'), Get(1, "big1"));
    ASSERT_EQ(std::string(1000, 'y'), Get(1, "big2"));
  } while (ChangeOptions());
}

TEST_F(DBCompactionTest, TrivialMoveOneFile) {
  int32_t trivial_move = 0;
  rocksdb::SyncPoint::GetInstance()->SetCallBack(
      "DBImpl::BackgroundCompaction:TrivialMove",
      [&](void* arg) { trivial_move++; });
  rocksdb::SyncPoint::GetInstance()->EnableProcessing();

  Options options;
  options.write_buffer_size = 100000000;
  options = CurrentOptions(options);
  DestroyAndReopen(options);

  int32_t num_keys = 80;
  int32_t value_size = 100 * 1024;  // 100 KB

  Random rnd(301);
  std::vector<std::string> values;
  for (int i = 0; i < num_keys; i++) {
    values.push_back(RandomString(&rnd, value_size));
    ASSERT_OK(Put(Key(i), values[i]));
  }

  // Reopening moves updates to L0
  Reopen(options);
  ASSERT_EQ(NumTableFilesAtLevel(0, 0), 1);  // 1 file in L0
  ASSERT_EQ(NumTableFilesAtLevel(1, 0), 0);  // 0 files in L1

  std::vector<LiveFileMetaData> metadata;
  db_->GetLiveFilesMetaData(&metadata);
  ASSERT_EQ(metadata.size(), 1U);
  LiveFileMetaData level0_file = metadata[0];  // L0 file meta

  // Compaction will initiate a trivial move from L0 to L1
  dbfull()->CompactRange(CompactRangeOptions(), nullptr, nullptr);

  // File moved From L0 to L1
  ASSERT_EQ(NumTableFilesAtLevel(0, 0), 0);  // 0 files in L0
  ASSERT_EQ(NumTableFilesAtLevel(1, 0), 1);  // 1 file in L1

  metadata.clear();
  db_->GetLiveFilesMetaData(&metadata);
  ASSERT_EQ(metadata.size(), 1U);
  ASSERT_EQ(metadata[0].name /* level1_file.name */, level0_file.name);
  ASSERT_EQ(metadata[0].size /* level1_file.size */, level0_file.size);

  for (int i = 0; i < num_keys; i++) {
    ASSERT_EQ(Get(Key(i)), values[i]);
  }

  ASSERT_EQ(trivial_move, 1);
  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
}

TEST_F(DBCompactionTest, TrivialMoveNonOverlappingFiles) {
  int32_t trivial_move = 0;
  int32_t non_trivial_move = 0;
  rocksdb::SyncPoint::GetInstance()->SetCallBack(
      "DBImpl::BackgroundCompaction:TrivialMove",
      [&](void* arg) { trivial_move++; });
  rocksdb::SyncPoint::GetInstance()->SetCallBack(
      "DBImpl::BackgroundCompaction:NonTrivial",
      [&](void* arg) { non_trivial_move++; });
  rocksdb::SyncPoint::GetInstance()->EnableProcessing();

  Options options = CurrentOptions();
  options.disable_auto_compactions = true;
  options.write_buffer_size = 10 * 1024 * 1024;

  DestroyAndReopen(options);
  // non overlapping ranges
  std::vector<std::pair<int32_t, int32_t>> ranges = {
    {100, 199},
    {300, 399},
    {0, 99},
    {200, 299},
    {600, 699},
    {400, 499},
    {500, 550},
    {551, 599},
  };
  int32_t value_size = 10 * 1024;  // 10 KB

  Random rnd(301);
  std::map<int32_t, std::string> values;
  for (uint32_t i = 0; i < ranges.size(); i++) {
    for (int32_t j = ranges[i].first; j <= ranges[i].second; j++) {
      values[j] = RandomString(&rnd, value_size);
      ASSERT_OK(Put(Key(j), values[j]));
    }
    ASSERT_OK(Flush());
  }

  int32_t level0_files = NumTableFilesAtLevel(0, 0);
  ASSERT_EQ(level0_files, ranges.size());    // Multiple files in L0
  ASSERT_EQ(NumTableFilesAtLevel(1, 0), 0);  // No files in L1

  // Since data is non-overlapping we expect compaction to initiate
  // a trivial move
  db_->CompactRange(CompactRangeOptions(), nullptr, nullptr);
  // We expect that all the files were trivially moved from L0 to L1
  ASSERT_EQ(NumTableFilesAtLevel(0, 0), 0);
  ASSERT_EQ(NumTableFilesAtLevel(1, 0) /* level1_files */, level0_files);

  for (uint32_t i = 0; i < ranges.size(); i++) {
    for (int32_t j = ranges[i].first; j <= ranges[i].second; j++) {
      ASSERT_EQ(Get(Key(j)), values[j]);
    }
  }

  ASSERT_EQ(trivial_move, 1);
  ASSERT_EQ(non_trivial_move, 0);

  trivial_move = 0;
  non_trivial_move = 0;
  values.clear();
  DestroyAndReopen(options);
  // Same ranges as above but overlapping
  ranges = {
    {100, 199},
    {300, 399},
    {0, 99},
    {200, 299},
    {600, 699},
    {400, 499},
    {500, 560},  // this range overlap with the next one
    {551, 599},
  };
  for (uint32_t i = 0; i < ranges.size(); i++) {
    for (int32_t j = ranges[i].first; j <= ranges[i].second; j++) {
      values[j] = RandomString(&rnd, value_size);
      ASSERT_OK(Put(Key(j), values[j]));
    }
    ASSERT_OK(Flush());
  }

  db_->CompactRange(CompactRangeOptions(), nullptr, nullptr);

  for (uint32_t i = 0; i < ranges.size(); i++) {
    for (int32_t j = ranges[i].first; j <= ranges[i].second; j++) {
      ASSERT_EQ(Get(Key(j)), values[j]);
    }
  }
  ASSERT_EQ(trivial_move, 0);
  ASSERT_EQ(non_trivial_move, 1);

  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
}

TEST_F(DBCompactionTest, TrivialMoveTargetLevel) {
  int32_t trivial_move = 0;
  int32_t non_trivial_move = 0;
  rocksdb::SyncPoint::GetInstance()->SetCallBack(
      "DBImpl::BackgroundCompaction:TrivialMove",
      [&](void* arg) { trivial_move++; });
  rocksdb::SyncPoint::GetInstance()->SetCallBack(
      "DBImpl::BackgroundCompaction:NonTrivial",
      [&](void* arg) { non_trivial_move++; });
  rocksdb::SyncPoint::GetInstance()->EnableProcessing();

  Options options = CurrentOptions();
  options.disable_auto_compactions = true;
  options.write_buffer_size = 10 * 1024 * 1024;
  options.num_levels = 7;

  DestroyAndReopen(options);
  int32_t value_size = 10 * 1024;  // 10 KB

  // Add 2 non-overlapping files
  Random rnd(301);
  std::map<int32_t, std::string> values;

  // file 1 [0 => 300]
  for (int32_t i = 0; i <= 300; i++) {
    values[i] = RandomString(&rnd, value_size);
    ASSERT_OK(Put(Key(i), values[i]));
  }
  ASSERT_OK(Flush());

  // file 2 [600 => 700]
  for (int32_t i = 600; i <= 700; i++) {
    values[i] = RandomString(&rnd, value_size);
    ASSERT_OK(Put(Key(i), values[i]));
  }
  ASSERT_OK(Flush());

  // 2 files in L0
  ASSERT_EQ("2", FilesPerLevel(0));
  CompactRangeOptions compact_options;
  compact_options.change_level = true;
  compact_options.target_level = 6;
  ASSERT_OK(db_->CompactRange(compact_options, nullptr, nullptr));
  // 2 files in L6
  ASSERT_EQ("0,0,0,0,0,0,2", FilesPerLevel(0));

  ASSERT_EQ(trivial_move, 1);
  ASSERT_EQ(non_trivial_move, 0);

  for (int32_t i = 0; i <= 300; i++) {
    ASSERT_EQ(Get(Key(i)), values[i]);
  }
  for (int32_t i = 600; i <= 700; i++) {
    ASSERT_EQ(Get(Key(i)), values[i]);
  }
}

TEST_F(DBCompactionTest, TrivialMoveToLastLevelWithFiles) {
  int32_t trivial_move = 0;
  int32_t non_trivial_move = 0;
  rocksdb::SyncPoint::GetInstance()->SetCallBack(
      "DBImpl::BackgroundCompaction:TrivialMove",
      [&](void* arg) { trivial_move++; });
  rocksdb::SyncPoint::GetInstance()->SetCallBack(
      "DBImpl::BackgroundCompaction:NonTrivial",
      [&](void* arg) { non_trivial_move++; });
  rocksdb::SyncPoint::GetInstance()->EnableProcessing();

  Options options;
  options.write_buffer_size = 100000000;
  options = CurrentOptions(options);
  DestroyAndReopen(options);

  int32_t value_size = 10 * 1024;  // 10 KB

  Random rnd(301);
  std::vector<std::string> values;
  // File with keys [ 0 => 99 ]
  for (int i = 0; i < 100; i++) {
    values.push_back(RandomString(&rnd, value_size));
    ASSERT_OK(Put(Key(i), values[i]));
  }
  ASSERT_OK(Flush());

  ASSERT_EQ("1", FilesPerLevel(0));
  // Compaction will do L0=>L1 (trivial move) then move L1 files to L3
  CompactRangeOptions compact_options;
  compact_options.change_level = true;
  compact_options.target_level = 3;
  ASSERT_OK(db_->CompactRange(compact_options, nullptr, nullptr));
  ASSERT_EQ("0,0,0,1", FilesPerLevel(0));
  ASSERT_EQ(trivial_move, 1);
  ASSERT_EQ(non_trivial_move, 0);

  // File with keys [ 100 => 199 ]
  for (int i = 100; i < 200; i++) {
    values.push_back(RandomString(&rnd, value_size));
    ASSERT_OK(Put(Key(i), values[i]));
  }
  ASSERT_OK(Flush());

  ASSERT_EQ("1,0,0,1", FilesPerLevel(0));
  // Compaction will do L0=>L1 L1=>L2 L2=>L3 (3 trivial moves)
  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
  ASSERT_EQ("0,0,0,2", FilesPerLevel(0));
  ASSERT_EQ(trivial_move, 4);
  ASSERT_EQ(non_trivial_move, 0);

  for (int i = 0; i < 200; i++) {
    ASSERT_EQ(Get(Key(i)), values[i]);
  }

  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
}

TEST_F(DBCompactionTest, LevelCompactionThirdPath) {
  Options options = CurrentOptions();
  options.db_paths.emplace_back(dbname_, 500 * 1024);
  options.db_paths.emplace_back(dbname_ + "_2", 4 * 1024 * 1024);
  options.db_paths.emplace_back(dbname_ + "_3", 1024 * 1024 * 1024);
  options.compaction_style = kCompactionStyleLevel;
  options.write_buffer_size = 100 << 10;  // 100KB
  options.level0_file_num_compaction_trigger = 2;
  options.num_levels = 4;
  options.max_bytes_for_level_base = 400 * 1024;
  //  options = CurrentOptions(options);

  std::vector<std::string> filenames;
  env_->GetChildren(options.db_paths[1].path, &filenames);
  // Delete archival files.
  for (size_t i = 0; i < filenames.size(); ++i) {
    env_->DeleteFile(options.db_paths[1].path + "/" + filenames[i]);
  }
  env_->DeleteDir(options.db_paths[1].path);
  Reopen(options);

  Random rnd(301);
  int key_idx = 0;

  // First three 110KB files are not going to second path.
  // After that, (100K, 200K)
  for (int num = 0; num < 3; num++) {
    GenerateNewFile(&rnd, &key_idx);
  }

  // Another 110KB triggers a compaction to 400K file to fill up first path
  GenerateNewFile(&rnd, &key_idx);
  ASSERT_EQ(3, GetSstFileCount(options.db_paths[1].path));

  // (1, 4)
  GenerateNewFile(&rnd, &key_idx);
  ASSERT_EQ("1,4", FilesPerLevel(0));
  ASSERT_EQ(4, GetSstFileCount(options.db_paths[1].path));
  ASSERT_EQ(1, GetSstFileCount(dbname_));

  // (1, 4, 1)
  GenerateNewFile(&rnd, &key_idx);
  ASSERT_EQ("1,4,1", FilesPerLevel(0));
  ASSERT_EQ(1, GetSstFileCount(options.db_paths[2].path));
  ASSERT_EQ(4, GetSstFileCount(options.db_paths[1].path));
  ASSERT_EQ(1, GetSstFileCount(dbname_));

  // (1, 4, 2)
  GenerateNewFile(&rnd, &key_idx);
  ASSERT_EQ("1,4,2", FilesPerLevel(0));
  ASSERT_EQ(2, GetSstFileCount(options.db_paths[2].path));
  ASSERT_EQ(4, GetSstFileCount(options.db_paths[1].path));
  ASSERT_EQ(1, GetSstFileCount(dbname_));

  // (1, 4, 3)
  GenerateNewFile(&rnd, &key_idx);
  ASSERT_EQ("1,4,3", FilesPerLevel(0));
  ASSERT_EQ(3, GetSstFileCount(options.db_paths[2].path));
  ASSERT_EQ(4, GetSstFileCount(options.db_paths[1].path));
  ASSERT_EQ(1, GetSstFileCount(dbname_));

  // (1, 4, 4)
  GenerateNewFile(&rnd, &key_idx);
  ASSERT_EQ("1,4,4", FilesPerLevel(0));
  ASSERT_EQ(4, GetSstFileCount(options.db_paths[2].path));
  ASSERT_EQ(4, GetSstFileCount(options.db_paths[1].path));
  ASSERT_EQ(1, GetSstFileCount(dbname_));

  // (1, 4, 5)
  GenerateNewFile(&rnd, &key_idx);
  ASSERT_EQ("1,4,5", FilesPerLevel(0));
  ASSERT_EQ(5, GetSstFileCount(options.db_paths[2].path));
  ASSERT_EQ(4, GetSstFileCount(options.db_paths[1].path));
  ASSERT_EQ(1, GetSstFileCount(dbname_));

  // (1, 4, 6)
  GenerateNewFile(&rnd, &key_idx);
  ASSERT_EQ("1,4,6", FilesPerLevel(0));
  ASSERT_EQ(6, GetSstFileCount(options.db_paths[2].path));
  ASSERT_EQ(4, GetSstFileCount(options.db_paths[1].path));
  ASSERT_EQ(1, GetSstFileCount(dbname_));

  // (1, 4, 7)
  GenerateNewFile(&rnd, &key_idx);
  ASSERT_EQ("1,4,7", FilesPerLevel(0));
  ASSERT_EQ(7, GetSstFileCount(options.db_paths[2].path));
  ASSERT_EQ(4, GetSstFileCount(options.db_paths[1].path));
  ASSERT_EQ(1, GetSstFileCount(dbname_));

  // (1, 4, 8)
  GenerateNewFile(&rnd, &key_idx);
  ASSERT_EQ("1,4,8", FilesPerLevel(0));
  ASSERT_EQ(8, GetSstFileCount(options.db_paths[2].path));
  ASSERT_EQ(4, GetSstFileCount(options.db_paths[1].path));
  ASSERT_EQ(1, GetSstFileCount(dbname_));

  for (int i = 0; i < key_idx; i++) {
    auto v = Get(Key(i));
    ASSERT_NE(v, "NOT_FOUND");
    ASSERT_TRUE(v.size() == 1 || v.size() == 10000);
  }

  Reopen(options);

  for (int i = 0; i < key_idx; i++) {
    auto v = Get(Key(i));
    ASSERT_NE(v, "NOT_FOUND");
    ASSERT_TRUE(v.size() == 1 || v.size() == 10000);
  }

  Destroy(options);
}

TEST_F(DBCompactionTest, LevelCompactionPathUse) {
  Options options = CurrentOptions();
  options.db_paths.emplace_back(dbname_, 500 * 1024);
  options.db_paths.emplace_back(dbname_ + "_2", 4 * 1024 * 1024);
  options.db_paths.emplace_back(dbname_ + "_3", 1024 * 1024 * 1024);
  options.compaction_style = kCompactionStyleLevel;
  options.write_buffer_size = 100 << 10;  // 100KB
  options.level0_file_num_compaction_trigger = 2;
  options.num_levels = 4;
  options.max_bytes_for_level_base = 400 * 1024;
  //  options = CurrentOptions(options);

  std::vector<std::string> filenames;
  env_->GetChildren(options.db_paths[1].path, &filenames);
  // Delete archival files.
  for (size_t i = 0; i < filenames.size(); ++i) {
    env_->DeleteFile(options.db_paths[1].path + "/" + filenames[i]);
  }
  env_->DeleteDir(options.db_paths[1].path);
  Reopen(options);

  Random rnd(301);
  int key_idx = 0;

  // Always gets compacted into 1 Level1 file,
  // 0/1 Level 0 file
  for (int num = 0; num < 3; num++) {
    key_idx = 0;
    GenerateNewFile(&rnd, &key_idx);
  }

  key_idx = 0;
  GenerateNewFile(&rnd, &key_idx);
  ASSERT_EQ(1, GetSstFileCount(options.db_paths[1].path));

  key_idx = 0;
  GenerateNewFile(&rnd, &key_idx);
  ASSERT_EQ("1,1", FilesPerLevel(0));
  ASSERT_EQ(1, GetSstFileCount(options.db_paths[1].path));
  ASSERT_EQ(1, GetSstFileCount(dbname_));

  key_idx = 0;
  GenerateNewFile(&rnd, &key_idx);
  ASSERT_EQ("0,1", FilesPerLevel(0));
  ASSERT_EQ(0, GetSstFileCount(options.db_paths[2].path));
  ASSERT_EQ(1, GetSstFileCount(options.db_paths[1].path));
  ASSERT_EQ(0, GetSstFileCount(dbname_));

  key_idx = 0;
  GenerateNewFile(&rnd, &key_idx);
  ASSERT_EQ("1,1", FilesPerLevel(0));
  ASSERT_EQ(0, GetSstFileCount(options.db_paths[2].path));
  ASSERT_EQ(1, GetSstFileCount(options.db_paths[1].path));
  ASSERT_EQ(1, GetSstFileCount(dbname_));

  key_idx = 0;
  GenerateNewFile(&rnd, &key_idx);
  ASSERT_EQ("0,1", FilesPerLevel(0));
  ASSERT_EQ(0, GetSstFileCount(options.db_paths[2].path));
  ASSERT_EQ(1, GetSstFileCount(options.db_paths[1].path));
  ASSERT_EQ(0, GetSstFileCount(dbname_));

  key_idx = 0;
  GenerateNewFile(&rnd, &key_idx);
  ASSERT_EQ("1,1", FilesPerLevel(0));
  ASSERT_EQ(0, GetSstFileCount(options.db_paths[2].path));
  ASSERT_EQ(1, GetSstFileCount(options.db_paths[1].path));
  ASSERT_EQ(1, GetSstFileCount(dbname_));

  key_idx = 0;
  GenerateNewFile(&rnd, &key_idx);
  ASSERT_EQ("0,1", FilesPerLevel(0));
  ASSERT_EQ(0, GetSstFileCount(options.db_paths[2].path));
  ASSERT_EQ(1, GetSstFileCount(options.db_paths[1].path));
  ASSERT_EQ(0, GetSstFileCount(dbname_));

  key_idx = 0;
  GenerateNewFile(&rnd, &key_idx);
  ASSERT_EQ("1,1", FilesPerLevel(0));
  ASSERT_EQ(0, GetSstFileCount(options.db_paths[2].path));
  ASSERT_EQ(1, GetSstFileCount(options.db_paths[1].path));
  ASSERT_EQ(1, GetSstFileCount(dbname_));

  key_idx = 0;
  GenerateNewFile(&rnd, &key_idx);
  ASSERT_EQ("0,1", FilesPerLevel(0));
  ASSERT_EQ(0, GetSstFileCount(options.db_paths[2].path));
  ASSERT_EQ(1, GetSstFileCount(options.db_paths[1].path));
  ASSERT_EQ(0, GetSstFileCount(dbname_));

  key_idx = 0;
  GenerateNewFile(&rnd, &key_idx);
  ASSERT_EQ("1,1", FilesPerLevel(0));
  ASSERT_EQ(0, GetSstFileCount(options.db_paths[2].path));
  ASSERT_EQ(1, GetSstFileCount(options.db_paths[1].path));
  ASSERT_EQ(1, GetSstFileCount(dbname_));

  for (int i = 0; i < key_idx; i++) {
    auto v = Get(Key(i));
    ASSERT_NE(v, "NOT_FOUND");
    ASSERT_TRUE(v.size() == 1 || v.size() == 10000);
  }

  Reopen(options);

  for (int i = 0; i < key_idx; i++) {
    auto v = Get(Key(i));
    ASSERT_NE(v, "NOT_FOUND");
    ASSERT_TRUE(v.size() == 1 || v.size() == 10000);
  }

  Destroy(options);
}

TEST_F(DBCompactionTest, ConvertCompactionStyle) {
  Random rnd(301);
  int max_key_level_insert = 200;
  int max_key_universal_insert = 600;

  // Stage 1: generate a db with level compaction
  Options options;
  options.write_buffer_size = 100 << 10;  // 100KB
  options.num_levels = 4;
  options.level0_file_num_compaction_trigger = 3;
  options.max_bytes_for_level_base = 500 << 10;  // 500KB
  options.max_bytes_for_level_multiplier = 1;
  options.target_file_size_base = 200 << 10;  // 200KB
  options.target_file_size_multiplier = 1;
  options = CurrentOptions(options);
  CreateAndReopenWithCF({"pikachu"}, options);

  for (int i = 0; i <= max_key_level_insert; i++) {
    // each value is 10K
    ASSERT_OK(Put(1, Key(i), RandomString(&rnd, 10000)));
  }
  ASSERT_OK(Flush(1));
  dbfull()->TEST_WaitForCompact();

  ASSERT_GT(TotalTableFiles(1, 4), 1);
  int non_level0_num_files = 0;
  for (int i = 1; i < options.num_levels; i++) {
    non_level0_num_files += NumTableFilesAtLevel(i, 1);
  }
  ASSERT_GT(non_level0_num_files, 0);

  // Stage 2: reopen with universal compaction - should fail
  options = CurrentOptions();
  options.compaction_style = kCompactionStyleUniversal;
  options.num_levels = 1;
  options = CurrentOptions(options);
  Status s = TryReopenWithColumnFamilies({"default", "pikachu"}, options);
  ASSERT_TRUE(s.IsInvalidArgument());

  // Stage 3: compact into a single file and move the file to level 0
  options = CurrentOptions();
  options.disable_auto_compactions = true;
  options.target_file_size_base = INT_MAX;
  options.target_file_size_multiplier = 1;
  options.max_bytes_for_level_base = INT_MAX;
  options.max_bytes_for_level_multiplier = 1;
  options.num_levels = 4;
  options = CurrentOptions(options);
  ReopenWithColumnFamilies({"default", "pikachu"}, options);

  CompactRangeOptions compact_options;
  compact_options.change_level = true;
  compact_options.target_level = 0;
  compact_options.bottommost_level_compaction =
      BottommostLevelCompaction::kForce;
  dbfull()->CompactRange(compact_options, handles_[1], nullptr, nullptr);

  // Only 1 file in L0
  ASSERT_EQ("1", FilesPerLevel(1));

  // Stage 4: re-open in universal compaction style and do some db operations
  options = CurrentOptions();
  options.compaction_style = kCompactionStyleUniversal;
  options.num_levels = 4;
  options.write_buffer_size = 100 << 10;  // 100KB
  options.level0_file_num_compaction_trigger = 3;
  options = CurrentOptions(options);
  ReopenWithColumnFamilies({"default", "pikachu"}, options);

  options.num_levels = 1;
  ReopenWithColumnFamilies({"default", "pikachu"}, options);

  for (int i = max_key_level_insert / 2; i <= max_key_universal_insert; i++) {
    ASSERT_OK(Put(1, Key(i), RandomString(&rnd, 10000)));
  }
  dbfull()->Flush(FlushOptions());
  ASSERT_OK(Flush(1));
  dbfull()->TEST_WaitForCompact();

  for (int i = 1; i < options.num_levels; i++) {
    ASSERT_EQ(NumTableFilesAtLevel(i, 1), 0);
  }

  // verify keys inserted in both level compaction style and universal
  // compaction style
  std::string keys_in_db;
  Iterator* iter = dbfull()->NewIterator(ReadOptions(), handles_[1]);
  for (iter->SeekToFirst(); iter->Valid(); iter->Next()) {
    keys_in_db.append(iter->key().ToString());
    keys_in_db.push_back(',');
  }
  delete iter;

  std::string expected_keys;
  for (int i = 0; i <= max_key_universal_insert; i++) {
    expected_keys.append(Key(i));
    expected_keys.push_back(',');
  }

  ASSERT_EQ(keys_in_db, expected_keys);
}

TEST_F(DBCompactionTest, L0_CompactionBug_Issue44_a) {
  do {
    CreateAndReopenWithCF({"pikachu"}, CurrentOptions());
    ASSERT_OK(Put(1, "b", "v"));
    ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
    ASSERT_OK(Delete(1, "b"));
    ASSERT_OK(Delete(1, "a"));
    ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
    ASSERT_OK(Delete(1, "a"));
    ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
    ASSERT_OK(Put(1, "a", "v"));
    ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
    ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
    ASSERT_EQ("(a->v)", Contents(1));
    env_->SleepForMicroseconds(1000000);  // Wait for compaction to finish
    ASSERT_EQ("(a->v)", Contents(1));
  } while (ChangeCompactOptions());
}

TEST_F(DBCompactionTest, L0_CompactionBug_Issue44_b) {
  do {
    CreateAndReopenWithCF({"pikachu"}, CurrentOptions());
    Put(1, "", "");
    ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
    Delete(1, "e");
    Put(1, "", "");
    ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
    Put(1, "c", "cv");
    ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
    Put(1, "", "");
    ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
    Put(1, "", "");
    env_->SleepForMicroseconds(1000000);  // Wait for compaction to finish
    ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
    Put(1, "d", "dv");
    ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
    Put(1, "", "");
    ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
    Delete(1, "d");
    Delete(1, "b");
    ReopenWithColumnFamilies({"default", "pikachu"}, CurrentOptions());
    ASSERT_EQ("(->)(c->cv)", Contents(1));
    env_->SleepForMicroseconds(1000000);  // Wait for compaction to finish
    ASSERT_EQ("(->)(c->cv)", Contents(1));
  } while (ChangeCompactOptions());
}

TEST_F(DBCompactionTest, ManualCompaction) {
  Options options = CurrentOptions();
  CreateAndReopenWithCF({"pikachu"}, options);

  // iter - 0 with 7 levels
  // iter - 1 with 3 levels
  for (int iter = 0; iter < 2; ++iter) {
    MakeTables(3, "p", "q", 1);
    ASSERT_EQ("1,1,1", FilesPerLevel(1));

    // Compaction range falls before files
    Compact(1, "", "c");
    ASSERT_EQ("1,1,1", FilesPerLevel(1));

    // Compaction range falls after files
    Compact(1, "r", "z");
    ASSERT_EQ("1,1,1", FilesPerLevel(1));

    // Compaction range overlaps files
    Compact(1, "p1", "p9");
    ASSERT_EQ("0,0,1", FilesPerLevel(1));

    // Populate a different range
    MakeTables(3, "c", "e", 1);
    ASSERT_EQ("1,1,2", FilesPerLevel(1));

    // Compact just the new range
    Compact(1, "b", "f");
    ASSERT_EQ("0,0,2", FilesPerLevel(1));

    // Compact all
    MakeTables(1, "a", "z", 1);
    ASSERT_EQ("1,0,2", FilesPerLevel(1));
    db_->CompactRange(CompactRangeOptions(), handles_[1], nullptr, nullptr);
    ASSERT_EQ("0,0,1", FilesPerLevel(1));

    if (iter == 0) {
      options = CurrentOptions();
      options.max_background_flushes = 0;
      options.num_levels = 3;
      options.create_if_missing = true;
      DestroyAndReopen(options);
      CreateAndReopenWithCF({"pikachu"}, options);
    }
  }
}


TEST_F(DBCompactionTest, ManualLevelCompactionOutputPathId) {
  Options options = CurrentOptions();
  options.db_paths.emplace_back(dbname_ + "_2", 2 * 10485760);
  options.db_paths.emplace_back(dbname_ + "_3", 100 * 10485760);
  options.db_paths.emplace_back(dbname_ + "_4", 120 * 10485760);
  CreateAndReopenWithCF({"pikachu"}, options);

  // iter - 0 with 7 levels
  // iter - 1 with 3 levels
  for (int iter = 0; iter < 2; ++iter) {
    for (int i = 0; i < 3; ++i) {
      ASSERT_OK(Put(1, "p", "begin"));
      ASSERT_OK(Put(1, "q", "end"));
      ASSERT_OK(Flush(1));
    }
    ASSERT_EQ("3", FilesPerLevel(1));
    ASSERT_EQ(3, GetSstFileCount(options.db_paths[0].path));
    ASSERT_EQ(0, GetSstFileCount(dbname_));

    // Compaction range falls before files
    Compact(1, "", "c");
    ASSERT_EQ("3", FilesPerLevel(1));

    // Compaction range falls after files
    Compact(1, "r", "z");
    ASSERT_EQ("3", FilesPerLevel(1));

    // Compaction range overlaps files
    Compact(1, "p1", "p9", 1);
    ASSERT_EQ("0,1", FilesPerLevel(1));
    ASSERT_EQ(1, GetSstFileCount(options.db_paths[1].path));
    ASSERT_EQ(0, GetSstFileCount(options.db_paths[0].path));
    ASSERT_EQ(0, GetSstFileCount(dbname_));

    // Populate a different range
    for (int i = 0; i < 3; ++i) {
      ASSERT_OK(Put(1, "c", "begin"));
      ASSERT_OK(Put(1, "e", "end"));
      ASSERT_OK(Flush(1));
    }
    ASSERT_EQ("3,1", FilesPerLevel(1));

    // Compact just the new range
    Compact(1, "b", "f", 1);
    ASSERT_EQ("0,2", FilesPerLevel(1));
    ASSERT_EQ(2, GetSstFileCount(options.db_paths[1].path));
    ASSERT_EQ(0, GetSstFileCount(options.db_paths[0].path));
    ASSERT_EQ(0, GetSstFileCount(dbname_));

    // Compact all
    ASSERT_OK(Put(1, "a", "begin"));
    ASSERT_OK(Put(1, "z", "end"));
    ASSERT_OK(Flush(1));
    ASSERT_EQ("1,2", FilesPerLevel(1));
    ASSERT_EQ(2, GetSstFileCount(options.db_paths[1].path));
    ASSERT_EQ(1, GetSstFileCount(options.db_paths[0].path));
    CompactRangeOptions compact_options;
    compact_options.target_path_id = 1;
    db_->CompactRange(compact_options, handles_[1], nullptr, nullptr);
    ASSERT_EQ("0,1", FilesPerLevel(1));
    ASSERT_EQ(1, GetSstFileCount(options.db_paths[1].path));
    ASSERT_EQ(0, GetSstFileCount(options.db_paths[0].path));
    ASSERT_EQ(0, GetSstFileCount(dbname_));

    if (iter == 0) {
      DestroyAndReopen(options);
      options = CurrentOptions();
      options.db_paths.emplace_back(dbname_ + "_2", 2 * 10485760);
      options.db_paths.emplace_back(dbname_ + "_3", 100 * 10485760);
      options.db_paths.emplace_back(dbname_ + "_4", 120 * 10485760);
      options.max_background_flushes = 1;
      options.num_levels = 3;
      options.create_if_missing = true;
      CreateAndReopenWithCF({"pikachu"}, options);
    }
  }
}

TEST_F(DBCompactionTest, FilesDeletedAfterCompaction) {
  do {
    CreateAndReopenWithCF({"pikachu"}, CurrentOptions());
    ASSERT_OK(Put(1, "foo", "v2"));
    Compact(1, "a", "z");
    const size_t num_files = CountLiveFiles();
    for (int i = 0; i < 10; i++) {
      ASSERT_OK(Put(1, "foo", "v2"));
      Compact(1, "a", "z");
    }
    ASSERT_EQ(CountLiveFiles(), num_files);
  } while (ChangeCompactOptions());
}

// Check level comapction with compact files
TEST_F(DBCompactionTest, CompactFilesOnLevelCompaction) {
  const int kTestKeySize = 16;
  const int kTestValueSize = 984;
  const int kEntrySize = kTestKeySize + kTestValueSize;
  const int kEntriesPerBuffer = 100;
  Options options;
  options.create_if_missing = true;
  options.write_buffer_size = kEntrySize * kEntriesPerBuffer;
  options.compaction_style = kCompactionStyleLevel;
  options.target_file_size_base = options.write_buffer_size;
  options.max_bytes_for_level_base = options.target_file_size_base * 2;
  options.level0_stop_writes_trigger = 2;
  options.max_bytes_for_level_multiplier = 2;
  options.compression = kNoCompression;
  options = CurrentOptions(options);
  CreateAndReopenWithCF({"pikachu"}, options);

  Random rnd(301);
  for (int key = 64 * kEntriesPerBuffer; key >= 0; --key) {
    ASSERT_OK(Put(1, ToString(key), RandomString(&rnd, kTestValueSize)));
  }
  dbfull()->TEST_WaitForFlushMemTable(handles_[1]);
  dbfull()->TEST_WaitForCompact();

  ColumnFamilyMetaData cf_meta;
  dbfull()->GetColumnFamilyMetaData(handles_[1], &cf_meta);
  int output_level = static_cast<int>(cf_meta.levels.size()) - 1;
  for (int file_picked = 5; file_picked > 0; --file_picked) {
    std::set<std::string> overlapping_file_names;
    std::vector<std::string> compaction_input_file_names;
    for (int f = 0; f < file_picked; ++f) {
      int level;
      auto file_meta = PickFileRandomly(cf_meta, &rnd, &level);
      compaction_input_file_names.push_back(file_meta->name);
      GetOverlappingFileNumbersForLevelCompaction(
          cf_meta, options.comparator, level, output_level,
          file_meta, &overlapping_file_names);
    }

    ASSERT_OK(dbfull()->CompactFiles(
        CompactionOptions(), handles_[1],
        compaction_input_file_names,
        output_level));

    // Make sure all overlapping files do not exist after compaction
    dbfull()->GetColumnFamilyMetaData(handles_[1], &cf_meta);
    VerifyCompactionResult(cf_meta, overlapping_file_names);
  }

  // make sure all key-values are still there.
  for (int key = 64 * kEntriesPerBuffer; key >= 0; --key) {
    ASSERT_NE(Get(1, ToString(key)), "NOT_FOUND");
  }
}

TEST_F(DBCompactionTest, PartialCompactionFailure) {
  Options options;
  const int kKeySize = 16;
  const int kKvSize = 1000;
  const int kKeysPerBuffer = 100;
  const int kNumL1Files = 5;
  options.create_if_missing = true;
  options.write_buffer_size = kKeysPerBuffer * kKvSize;
  options.max_write_buffer_number = 2;
  options.target_file_size_base =
      options.write_buffer_size *
      (options.max_write_buffer_number - 1);
  options.level0_file_num_compaction_trigger = kNumL1Files;
  options.max_bytes_for_level_base =
      options.level0_file_num_compaction_trigger *
      options.target_file_size_base;
  options.max_bytes_for_level_multiplier = 2;
  options.compression = kNoCompression;

  env_->SetBackgroundThreads(1, Env::HIGH);
  env_->SetBackgroundThreads(1, Env::LOW);
  // stop the compaction thread until we simulate the file creation failure.
  SleepingBackgroundTask sleeping_task_low;
  env_->Schedule(&SleepingBackgroundTask::DoSleepTask, &sleeping_task_low,
                 Env::Priority::LOW);

  options.env = env_;

  DestroyAndReopen(options);

  const int kNumInsertedKeys =
      options.level0_file_num_compaction_trigger *
      (options.max_write_buffer_number - 1) *
      kKeysPerBuffer;

  Random rnd(301);
  std::vector<std::string> keys;
  std::vector<std::string> values;
  for (int k = 0; k < kNumInsertedKeys; ++k) {
    keys.emplace_back(RandomString(&rnd, kKeySize));
    values.emplace_back(RandomString(&rnd, kKvSize - kKeySize));
    ASSERT_OK(Put(Slice(keys[k]), Slice(values[k])));
    dbfull()->TEST_WaitForFlushMemTable();
  }

  dbfull()->TEST_FlushMemTable(true);
  // Make sure the number of L0 files can trigger compaction.
  ASSERT_GE(NumTableFilesAtLevel(0),
            options.level0_file_num_compaction_trigger);

  auto previous_num_level0_files = NumTableFilesAtLevel(0);

  // Fail the first file creation.
  env_->non_writable_count_ = 1;
  sleeping_task_low.WakeUp();
  sleeping_task_low.WaitUntilDone();

  // Expect compaction to fail here as one file will fail its
  // creation.
  ASSERT_TRUE(!dbfull()->TEST_WaitForCompact().ok());

  // Verify L0 -> L1 compaction does fail.
  ASSERT_EQ(NumTableFilesAtLevel(1), 0);

  // Verify all L0 files are still there.
  ASSERT_EQ(NumTableFilesAtLevel(0), previous_num_level0_files);

  // All key-values must exist after compaction fails.
  for (int k = 0; k < kNumInsertedKeys; ++k) {
    ASSERT_EQ(values[k], Get(keys[k]));
  }

  env_->non_writable_count_ = 0;

  // Make sure RocksDB will not get into corrupted state.
  Reopen(options);

  // Verify again after reopen.
  for (int k = 0; k < kNumInsertedKeys; ++k) {
    ASSERT_EQ(values[k], Get(keys[k]));
  }
}

TEST_F(DBCompactionTest, DeleteMovedFileAfterCompaction) {
  // iter 1 -- delete_obsolete_files_period_micros == 0
  for (int iter = 0; iter < 2; ++iter) {
    // This test triggers move compaction and verifies that the file is not
    // deleted when it's part of move compaction
    Options options = CurrentOptions();
    options.env = env_;
    if (iter == 1) {
      options.delete_obsolete_files_period_micros = 0;
    }
    options.create_if_missing = true;
    options.level0_file_num_compaction_trigger =
        2;  // trigger compaction when we have 2 files
    OnFileDeletionListener* listener = new OnFileDeletionListener();
    options.listeners.emplace_back(listener);
    DestroyAndReopen(options);

    Random rnd(301);
    // Create two 1MB sst files
    for (int i = 0; i < 2; ++i) {
      // Create 1MB sst file
      for (int j = 0; j < 100; ++j) {
        ASSERT_OK(Put(Key(i * 50 + j), RandomString(&rnd, 10 * 1024)));
      }
      ASSERT_OK(Flush());
    }
    // this should execute L0->L1
    dbfull()->TEST_WaitForCompact();
    ASSERT_EQ("0,1", FilesPerLevel(0));

    // block compactions
    SleepingBackgroundTask sleeping_task;
    env_->Schedule(&SleepingBackgroundTask::DoSleepTask, &sleeping_task,
                   Env::Priority::LOW);

    options.max_bytes_for_level_base = 1024 * 1024;  // 1 MB
    Reopen(options);
    std::unique_ptr<Iterator> iterator(db_->NewIterator(ReadOptions()));
    ASSERT_EQ("0,1", FilesPerLevel(0));
    // let compactions go
    sleeping_task.WakeUp();
    sleeping_task.WaitUntilDone();

    // this should execute L1->L2 (move)
    dbfull()->TEST_WaitForCompact();

    ASSERT_EQ("0,0,1", FilesPerLevel(0));

    std::vector<LiveFileMetaData> metadata;
    db_->GetLiveFilesMetaData(&metadata);
    ASSERT_EQ(metadata.size(), 1U);
    auto moved_file_name = metadata[0].name;

    // Create two more 1MB sst files
    for (int i = 0; i < 2; ++i) {
      // Create 1MB sst file
      for (int j = 0; j < 100; ++j) {
        ASSERT_OK(Put(Key(i * 50 + j + 100), RandomString(&rnd, 10 * 1024)));
      }
      ASSERT_OK(Flush());
    }
    // this should execute both L0->L1 and L1->L2 (merge with previous file)
    dbfull()->TEST_WaitForCompact();

    ASSERT_EQ("0,0,2", FilesPerLevel(0));

    // iterator is holding the file
    ASSERT_OK(env_->FileExists(dbname_ + moved_file_name));

    listener->SetExpectedFileName(dbname_ + moved_file_name);
    iterator.reset();

    // this file should have been compacted away
    ASSERT_NOK(env_->FileExists(dbname_ + moved_file_name));
    listener->VerifyMatchedCount(1);
  }
}

TEST_F(DBCompactionTest, CompressLevelCompaction) {
  if (!Zlib_Supported()) {
    return;
  }
  Options options = CurrentOptions();
  options.compaction_style = kCompactionStyleLevel;
  options.write_buffer_size = 100 << 10;  // 100KB
  options.level0_file_num_compaction_trigger = 2;
  options.num_levels = 4;
  options.max_bytes_for_level_base = 400 * 1024;
  // First two levels have no compression, so that a trivial move between
  // them will be allowed. Level 2 has Zlib compression so that a trivial
  // move to level 3 will not be allowed
  options.compression_per_level = {kNoCompression, kNoCompression,
                                   kZlibCompression};
  int matches = 0, didnt_match = 0, trivial_move = 0, non_trivial = 0;

  rocksdb::SyncPoint::GetInstance()->SetCallBack(
      "Compaction::InputCompressionMatchesOutput:Matches",
      [&](void* arg) { matches++; });
  rocksdb::SyncPoint::GetInstance()->SetCallBack(
      "Compaction::InputCompressionMatchesOutput:DidntMatch",
      [&](void* arg) { didnt_match++; });
  rocksdb::SyncPoint::GetInstance()->SetCallBack(
      "DBImpl::BackgroundCompaction:NonTrivial",
      [&](void* arg) { non_trivial++; });
  rocksdb::SyncPoint::GetInstance()->SetCallBack(
      "DBImpl::BackgroundCompaction:TrivialMove",
      [&](void* arg) { trivial_move++; });
  rocksdb::SyncPoint::GetInstance()->EnableProcessing();

  Reopen(options);

  Random rnd(301);
  int key_idx = 0;

  // First three 110KB files are going to level 0
  // After that, (100K, 200K)
  for (int num = 0; num < 3; num++) {
    GenerateNewFile(&rnd, &key_idx);
  }

  // Another 110KB triggers a compaction to 400K file to fill up level 0
  GenerateNewFile(&rnd, &key_idx);
  ASSERT_EQ(4, GetSstFileCount(dbname_));

  // (1, 4)
  GenerateNewFile(&rnd, &key_idx);
  ASSERT_EQ("1,4", FilesPerLevel(0));

  // (1, 4, 1)
  GenerateNewFile(&rnd, &key_idx);
  ASSERT_EQ("1,4,1", FilesPerLevel(0));

  // (1, 4, 2)
  GenerateNewFile(&rnd, &key_idx);
  ASSERT_EQ("1,4,2", FilesPerLevel(0));

  // (1, 4, 3)
  GenerateNewFile(&rnd, &key_idx);
  ASSERT_EQ("1,4,3", FilesPerLevel(0));

  // (1, 4, 4)
  GenerateNewFile(&rnd, &key_idx);
  ASSERT_EQ("1,4,4", FilesPerLevel(0));

  // (1, 4, 5)
  GenerateNewFile(&rnd, &key_idx);
  ASSERT_EQ("1,4,5", FilesPerLevel(0));

  // (1, 4, 6)
  GenerateNewFile(&rnd, &key_idx);
  ASSERT_EQ("1,4,6", FilesPerLevel(0));

  // (1, 4, 7)
  GenerateNewFile(&rnd, &key_idx);
  ASSERT_EQ("1,4,7", FilesPerLevel(0));

  // (1, 4, 8)
  GenerateNewFile(&rnd, &key_idx);
  ASSERT_EQ("1,4,8", FilesPerLevel(0));

  ASSERT_EQ(matches, 12);
  // Currently, the test relies on the number of calls to
  // InputCompressionMatchesOutput() per compaction.
  const int kCallsToInputCompressionMatch = 2;
  ASSERT_EQ(didnt_match, 8 * kCallsToInputCompressionMatch);
  ASSERT_EQ(trivial_move, 12);
  ASSERT_EQ(non_trivial, 8);

  rocksdb::SyncPoint::GetInstance()->DisableProcessing();

  for (int i = 0; i < key_idx; i++) {
    auto v = Get(Key(i));
    ASSERT_NE(v, "NOT_FOUND");
    ASSERT_TRUE(v.size() == 1 || v.size() == 10000);
  }

  Reopen(options);

  for (int i = 0; i < key_idx; i++) {
    auto v = Get(Key(i));
    ASSERT_NE(v, "NOT_FOUND");
    ASSERT_TRUE(v.size() == 1 || v.size() == 10000);
  }

  Destroy(options);
}

// SYNC_POINT is not supported in released Windows mode.
#if !(defined NDEBUG) || !defined(OS_WIN)
// This tests for a bug that could cause two level0 compactions running
// concurrently
TEST_F(DBCompactionTest, SuggestCompactRangeNoTwoLevel0Compactions) {
  Options options = CurrentOptions();
  options.compaction_style = kCompactionStyleLevel;
  options.write_buffer_size = 110 << 10;
  options.level0_file_num_compaction_trigger = 4;
  options.num_levels = 4;
  options.compression = kNoCompression;
  options.max_bytes_for_level_base = 450 << 10;
  options.target_file_size_base = 98 << 10;
  options.max_write_buffer_number = 2;
  options.max_background_compactions = 2;

  DestroyAndReopen(options);

  // fill up the DB
  Random rnd(301);
  for (int num = 0; num < 10; num++) {
    GenerateNewRandomFile(&rnd);
  }
  db_->CompactRange(CompactRangeOptions(), nullptr, nullptr);

  rocksdb::SyncPoint::GetInstance()->LoadDependency(
      {{"CompactionJob::Run():Start",
        "DBCompactionTest::SuggestCompactRangeNoTwoLevel0Compactions:1"},
       {"DBCompactionTest::SuggestCompactRangeNoTwoLevel0Compactions:2",
        "CompactionJob::Run():End"}});

  rocksdb::SyncPoint::GetInstance()->EnableProcessing();

  // trigger L0 compaction
  for (int num = 0; num < options.level0_file_num_compaction_trigger + 1;
       num++) {
    GenerateNewRandomFile(&rnd, /* nowait */ true);
    ASSERT_OK(Flush());
  }

  TEST_SYNC_POINT(
      "DBCompactionTest::SuggestCompactRangeNoTwoLevel0Compactions:1");

  GenerateNewRandomFile(&rnd, /* nowait */ true);
  dbfull()->TEST_WaitForFlushMemTable();
  ASSERT_OK(experimental::SuggestCompactRange(db_, nullptr, nullptr));
  for (int num = 0; num < options.level0_file_num_compaction_trigger + 1;
       num++) {
    GenerateNewRandomFile(&rnd, /* nowait */ true);
    ASSERT_OK(Flush());
  }

  TEST_SYNC_POINT(
      "DBCompactionTest::SuggestCompactRangeNoTwoLevel0Compactions:2");
  dbfull()->TEST_WaitForCompact();
}

#endif  // !(defined NDEBUG) || !defined(OS_WIN)

TEST_F(DBCompactionTest, ForceBottommostLevelCompaction) {
  int32_t trivial_move = 0;
  int32_t non_trivial_move = 0;
  rocksdb::SyncPoint::GetInstance()->SetCallBack(
      "DBImpl::BackgroundCompaction:TrivialMove",
      [&](void* arg) { trivial_move++; });
  rocksdb::SyncPoint::GetInstance()->SetCallBack(
      "DBImpl::BackgroundCompaction:NonTrivial",
      [&](void* arg) { non_trivial_move++; });
  rocksdb::SyncPoint::GetInstance()->EnableProcessing();

  Options options;
  options.write_buffer_size = 100000000;
  options = CurrentOptions(options);
  DestroyAndReopen(options);

  int32_t value_size = 10 * 1024;  // 10 KB

  Random rnd(301);
  std::vector<std::string> values;
  // File with keys [ 0 => 99 ]
  for (int i = 0; i < 100; i++) {
    values.push_back(RandomString(&rnd, value_size));
    ASSERT_OK(Put(Key(i), values[i]));
  }
  ASSERT_OK(Flush());

  ASSERT_EQ("1", FilesPerLevel(0));
  // Compaction will do L0=>L1 (trivial move) then move L1 files to L3
  CompactRangeOptions compact_options;
  compact_options.change_level = true;
  compact_options.target_level = 3;
  ASSERT_OK(db_->CompactRange(compact_options, nullptr, nullptr));
  ASSERT_EQ("0,0,0,1", FilesPerLevel(0));
  ASSERT_EQ(trivial_move, 1);
  ASSERT_EQ(non_trivial_move, 0);

  // File with keys [ 100 => 199 ]
  for (int i = 100; i < 200; i++) {
    values.push_back(RandomString(&rnd, value_size));
    ASSERT_OK(Put(Key(i), values[i]));
  }
  ASSERT_OK(Flush());

  ASSERT_EQ("1,0,0,1", FilesPerLevel(0));
  // Compaction will do L0=>L1 L1=>L2 L2=>L3 (3 trivial moves)
  // then compacte the bottommost level L3=>L3 (non trivial move)
  compact_options = CompactRangeOptions();
  compact_options.bottommost_level_compaction =
      BottommostLevelCompaction::kForce;
  ASSERT_OK(db_->CompactRange(compact_options, nullptr, nullptr));
  ASSERT_EQ("0,0,0,1", FilesPerLevel(0));
  ASSERT_EQ(trivial_move, 4);
  ASSERT_EQ(non_trivial_move, 1);

  // File with keys [ 200 => 299 ]
  for (int i = 200; i < 300; i++) {
    values.push_back(RandomString(&rnd, value_size));
    ASSERT_OK(Put(Key(i), values[i]));
  }
  ASSERT_OK(Flush());

  ASSERT_EQ("1,0,0,1", FilesPerLevel(0));
  trivial_move = 0;
  non_trivial_move = 0;
  compact_options = CompactRangeOptions();
  compact_options.bottommost_level_compaction =
      BottommostLevelCompaction::kSkip;
  // Compaction will do L0=>L1 L1=>L2 L2=>L3 (3 trivial moves)
  // and will skip bottommost level compaction
  ASSERT_OK(db_->CompactRange(compact_options, nullptr, nullptr));
  ASSERT_EQ("0,0,0,2", FilesPerLevel(0));
  ASSERT_EQ(trivial_move, 3);
  ASSERT_EQ(non_trivial_move, 0);

  for (int i = 0; i < 300; i++) {
    ASSERT_EQ(Get(Key(i)), values[i]);
  }

  rocksdb::SyncPoint::GetInstance()->DisableProcessing();
}

}  // namespace rocksdb

int main(int argc, char** argv) {
#if !(defined NDEBUG) || !defined(OS_WIN)
  rocksdb::port::InstallStackTraceHandler();
  ::testing::InitGoogleTest(&argc, argv);
  return RUN_ALL_TESTS();
#else
  return 0;
#endif
}