Compaction實(shí)現(xiàn)細(xì)節(jié)

void DBImpl::BackgroundCallCompaction(PrepickedCompaction* prepicked_compaction,
                                      Env::Priority bg_thread_pri) {
  bool made_progress = false;
  JobContext job_context(next_job_id_.fetch_add(1), true);
  TEST_SYNC_POINT("BackgroundCallCompaction:0");
  LogBuffer log_buffer(InfoLogLevel::INFO_LEVEL,
                       immutable_db_options_.info_log.get());
  {
    //加了DB鎖
    //在backgroundcompaction配置好compaction_job前都加鎖（選要compact的文件之類的）
    //compaction_job.Run()會(huì)解鎖，跑完再加鎖
    InstrumentedMutexLock l(&mutex_);

    // This call will unlock/lock the mutex to wait for current running
    // IngestExternalFile() calls to finish.
    WaitForIngestFile();

    num_running_compactions_++;

    std::unique_ptr<std::list<uint64_t>::iterator>
        pending_outputs_inserted_elem(new std::list<uint64_t>::iterator(
            CaptureCurrentFileNumberInPendingOutputs()));

    assert((bg_thread_pri == Env::Priority::BOTTOM &&
            bg_bottom_compaction_scheduled_) ||
           (bg_thread_pri == Env::Priority::LOW && bg_compaction_scheduled_));
    //調(diào)用compaction
    Status s = BackgroundCompaction(&made_progress, &job_context, &log_buffer,
                                    prepicked_compaction, bg_thread_pri);
    //...
    //compaction完后的處理
}

Compaction流程

Status DBImpl::BackgroundCompaction(bool* made_progress,
                                    JobContext* job_context,
                                    LogBuffer* log_buffer,
                                    PrepickedCompaction* prepicked_compaction,
                                    Env::Priority thread_pri) {
    ... //判斷是否是manual compaction以及compaction空間是否夠
      
    auto cfd = PickCompactionFromQueue(&task_token, log_buffer);
    if (cfd == nullptr) {
      // Can't find any executable task from the compaction queue.
      // All tasks have been throttled by compaction thread limiter.
      ++unscheduled_compactions_;
      return Status::Busy();
    }
  
      auto* mutable_cf_options = cfd->GetLatestMutableCFOptions();
    if (!mutable_cf_options->disable_auto_compactions && !cfd->IsDropped()) {
      // NOTE: try to avoid unnecessary copy of MutableCFOptions if
      // compaction is not necessary. Need to make sure mutex is held
      // until we make a copy in the following code
      //挑選需要compaction的file這個(gè)之后分析      
      c.reset(cfd->PickCompaction(*mutable_cf_options, log_buffer));

      //。。。
      //一些處理
    }
  
  
    if (!c) {
    // Nothing to do
    ROCKS_LOG_BUFFER(log_buffer, "Compaction nothing to do");
  } else if (c->deletion_compaction()) {
    // 簡(jiǎn)單的刪除input文件的compaction
    //  ->If true, then the compaction can be done by simply deleting input files.
    // TODO(icanadi) Do we want to honor snapshots here? i.e. not delete old
    // file if there is alive snapshot pointing to it
    ...
  } else if (!trivial_move_disallowed && c->IsTrivialMove()) {
     //沒有merge和split，只需要將文件移動(dòng)到下一level即可
     //Is this a trivial compaction that can be implemented by just
     // moving a single input file to the next level (no merging or splitting)
    ...
  } else if (!is_prepicked && c->output_level() > 0 &&
             c->output_level() ==
                 c->column_family_data()
                     ->current()
                     ->storage_info()
                     ->MaxOutputLevel(
                         immutable_db_options_.allow_ingest_behind) &&
             env_->GetBackgroundThreads(Env::Priority::BOTTOM) > 0) {
    // Forward compactions involving last level to the bottom pool if it exists,
    // such that compactions unlikely to contribute to write stalls can be
    // delayed or deprioritized.
    //移動(dòng)到更bottom的線程去做compaction
    ...
  } else {
    //做compaction
    ...
      
    //構(gòu)建compactionjob
    CompactionJob compaction_job(
        job_context->job_id, c.get(), immutable_db_options_,
        env_options_for_compaction_, versions_.get(), &shutting_down_,
        preserve_deletes_seqnum_.load(), log_buffer, directories_.GetDbDir(),
        GetDataDir(c->column_family_data(), c->output_path_id()), stats_,
        &mutex_, &error_handler_, snapshot_seqs,
        earliest_write_conflict_snapshot, snapshot_checker, table_cache_,
        &event_logger_, c->mutable_cf_options()->paranoid_file_checks,
        c->mutable_cf_options()->report_bg_io_stats, dbname_,
        &compaction_job_stats, thread_pri,
        is_manual ? &manual_compaction_paused_ : nullptr);
    
    //計(jì)算subcompactions
    compaction_job.Prepare();

    NotifyOnCompactionBegin(c->column_family_data(), c.get(), status,
                            compaction_job_stats, job_context->job_id);

    //允許并發(fā)做compaction
    mutex_.Unlock();
    TEST_SYNC_POINT_CALLBACK(
        "DBImpl::BackgroundCompaction:NonTrivial:BeforeRun", nullptr);
    compaction_job.Run();
    TEST_SYNC_POINT("DBImpl::BackgroundCompaction:NonTrivial:AfterRun");
    mutex_.Lock();

    status = compaction_job.Install(*c->mutable_cf_options());
    if (status.ok()) {
      InstallSuperVersionAndScheduleWork(c->column_family_data(),
                                         &job_context->superversion_contexts[0],
                                         *c->mutable_cf_options());
    }
    *made_progress = true;
    TEST_SYNC_POINT_CALLBACK("DBImpl::BackgroundCompaction:AfterCompaction",
                             c->column_family_data());
  }

Prepare用來確定是否分割compaction，執(zhí)行subcompaction

void CompactionJob::Prepare() {
  AutoThreadOperationStageUpdater stage_updater(
      ThreadStatus::STAGE_COMPACTION_PREPARE);

  // Generate file_levels_ for compaction berfore making Iterator
  auto* c = compact_->compaction;
  assert(c->column_family_data() != nullptr);
  assert(c->column_family_data()->current()->storage_info()->NumLevelFiles(
             compact_->compaction->level()) > 0);

  write_hint_ =
      c->column_family_data()->CalculateSSTWriteHint(c->output_level());
  bottommost_level_ = c->bottommost_level();

  //這里面確定是否要做subcompaction
  //判斷條件如果是level，那么startlevel要為0，outlevel要>0（或者手動(dòng)compaction且outlevel>0）
  if (c->ShouldFormSubcompactions()) {
    {
      StopWatch sw(env_, stats_, SUBCOMPACTION_SETUP_TIME);
      //在GenSubcompactionBoundaries會(huì)解析到對(duì)應(yīng)的sub compaction以及邊界
      GenSubcompactionBoundaries();
    }
    assert(sizes_.size() == boundaries_.size() + 1);

    for (size_t i = 0; i <= boundaries_.size(); i++) {
      Slice* start = i == 0 ? nullptr : &boundaries_[i - 1];
      Slice* end = i == boundaries_.size() ? nullptr : &boundaries_[i];
      compact_->sub_compact_states.emplace_back(c, start, end, sizes_[i]);
    }
    RecordInHistogram(stats_, NUM_SUBCOMPACTIONS_SCHEDULED,
                      compact_->sub_compact_states.size());
  } else {
    compact_->sub_compact_states.emplace_back(c, nullptr, nullptr);
  }
}

Compaction的執(zhí)行

Status CompactionJob::Run() {
  AutoThreadOperationStageUpdater stage_updater(
      ThreadStatus::STAGE_COMPACTION_RUN);
  TEST_SYNC_POINT("CompactionJob::Run():Start");
  log_buffer_->FlushBufferToLog();
  LogCompaction();

  const size_t num_threads = compact_->sub_compact_states.size();
  assert(num_threads > 0);
  const uint64_t start_micros = env_->NowMicros();

  //為subcompaction啟動(dòng)線程執(zhí)行（無subcompaction/第一個(gè)subcompaction，就在本線程執(zhí)行）
  // Launch a thread for each of subcompactions 1...num_threads-1
  std::vector<port::Thread> thread_pool;
  thread_pool.reserve(num_threads - 1);
  for (size_t i = 1; i < compact_->sub_compact_states.size(); i++) {
    thread_pool.emplace_back(&CompactionJob::ProcessKeyValueCompaction, this,
                             &compact_->sub_compact_states[i]);
  }

  
  // Always schedule the first subcompaction (whether or not there are also
  // others) in the current thread to be efficient with resources
  ProcessKeyValueCompaction(&compact_->sub_compact_states[0]);

  // Wait for all other threads (if there are any) to finish execution
  for (auto& thread : thread_pool) {
    thread.join();
  }

ProcessKeyValueCompaction

和flushmemtable一樣，也是每個(gè)file的iterator構(gòu)成一個(gè)merge iterator，一個(gè)一個(gè)key順序讀出來構(gòu)建block和index