Determine boundaries of subcompactions

Summary:
Up to this point, the subcompactions that make up a compaction
job have been divided based on the key range of the L1 files, and each
subcompaction has handled the key range of only one file. However
DBOption.max_subcompactions allows the user to designate how many
subcompactions at most to perform. This patch updates the
CompactionJob::GetSubcompactionBoundaries() to determine these
divisions accordingly based on that option and other input/system factors.

The current approach orders the starting and/or ending keys of certain
compaction input files and then generates a histogram to approximate the
size covered by the key range between each consecutive pair of keys. Then
it groups these ranges into groups so that the sizes are approximately equal
to one another. The approach has also been adapted to work for universal
compaction as well instead of just for level-based compaction as it was before.

These subcompactions are then executed in parallel by locally spawning
threads, one for each. The results are then aggregated and the compaction
completed.

Test Plan: make all && make check

Reviewers: yhchiang, anthony, igor, noetzli, sdong

Reviewed By: sdong

Subscribers: MarkCallaghan, dhruba, leveldb

Differential Revision: https://reviews.facebook.net/D43269

db/compaction.cc

@@ -377,4 +377,21 @@ std::unique_ptr<CompactionFilter> Compaction::CreateCompactionFilter() const {
       context);
 }
 
+bool Compaction::IsOutputLevelEmpty() const {
+  return inputs_.back().level != output_level_ || inputs_.back().empty();
+}
+
+bool Compaction::ShouldFormSubcompactions() const {
+  if (mutable_cf_options_.max_subcompactions <= 1 || cfd_ == nullptr) {
+    return false;
+  }
+  if (cfd_->ioptions()->compaction_style == kCompactionStyleLevel) {
+    return start_level_ == 0 && !IsOutputLevelEmpty();
+  } else if (cfd_->ioptions()->compaction_style == kCompactionStyleUniversal) {
+    return number_levels_ > 1 && output_level_ > 0;
+  } else {
+    return false;
+  }
+}
+
 }  // namespace rocksdb

db/compaction.h

@@ -198,16 +198,11 @@ class Compaction {
   // Create a CompactionFilter from compaction_filter_factory
   std::unique_ptr<CompactionFilter> CreateCompactionFilter() const;
 
-  // Should this compaction be broken up into smaller ones run in parallel?
-  bool IsSubCompaction() const {
-    return start_level_ == 0 && output_level_ == 1 &&
-           mutable_cf_options_.max_subcompactions > 1;
-  }
+  // Is the input level corresponding to output_level_ empty?
+  bool IsOutputLevelEmpty() const;
 
-  // If is_sub_compaction == true, how many smaller compactions should execute
-  int NumSubCompactions() const {
-    return mutable_cf_options_.max_subcompactions;
-  }
+  // Should this compaction be broken up into smaller ones run in parallel?
+  bool ShouldFormSubcompactions() const;
 
  private:
   // mark (or clear) all files that are being compacted

db/compaction_job.cc

@@ -15,9 +15,13 @@
 
 #include <inttypes.h>
 #include <algorithm>
+#include <functional>
 #include <vector>
 #include <memory>
 #include <list>
+#include <set>
+#include <thread>
+#include <utility>
 
 #include "db/builder.h"
 #include "db/db_iter.h"
@@ -58,7 +62,7 @@
 namespace rocksdb {
 
 // Maintains state for each sub-compaction
-struct CompactionJob::SubCompactionState {
+struct CompactionJob::SubcompactionState {
   Compaction* compaction;
 
   // The boundaries of the key-range this compaction is interested in. No two
@@ -66,10 +70,10 @@ struct CompactionJob::SubCompactionState {
   // 'start' is inclusive, 'end' is exclusive, and nullptr means unbounded
   Slice *start, *end;
 
-  // The return status of this compaction
+  // The return status of this subcompaction
   Status status;
 
-  // Files produced by compaction
+  // Files produced by this subcompaction
   struct Output {
     uint64_t number;
     uint32_t path_id;
@@ -88,7 +92,7 @@ struct CompactionJob::SubCompactionState {
       // This subcompaction's ouptut could be empty if compaction was aborted
       // before this subcompaction had a chance to generate any output files.
       // When subcompactions are executed sequentially this is more likely and
-      // will be particulalry likely for the last subcompaction to be empty.
+      // will be particulalry likely for the later subcompactions to be empty.
       // Once they are run in parallel however it should be much rarer.
       return nullptr;
     } else {
@@ -96,11 +100,12 @@ struct CompactionJob::SubCompactionState {
     }
   }
 
-  // State during the sub-compaction
+  // State during the subcompaction
   uint64_t total_bytes;
   uint64_t num_input_records;
   uint64_t num_output_records;
   CompactionJobStats compaction_job_stats;
+  uint64_t approx_size;
 
   // "level_ptrs" holds indices that remember which file of an associated
   // level we were last checking during the last call to compaction->
@@ -110,7 +115,8 @@ struct CompactionJob::SubCompactionState {
   // is in or beyond the last file checked during the previous call
   std::vector<size_t> level_ptrs;
 
-  SubCompactionState(Compaction* c, Slice* _start, Slice* _end)
+  SubcompactionState(Compaction* c, Slice* _start, Slice* _end,
+                     uint64_t size = 0)
       : compaction(c),
         start(_start),
         end(_end),
@@ -118,16 +124,15 @@ struct CompactionJob::SubCompactionState {
         builder(nullptr),
         total_bytes(0),
         num_input_records(0),
-        num_output_records(0) {
+        num_output_records(0),
+        approx_size(size) {
     assert(compaction != nullptr);
     level_ptrs = std::vector<size_t>(compaction->number_levels(), 0);
   }
 
-  SubCompactionState(SubCompactionState&& o) {
-    *this = std::move(o);
-  }
+  SubcompactionState(SubcompactionState&& o) { *this = std::move(o); }
 
-  SubCompactionState& operator=(SubCompactionState&& o) {
+  SubcompactionState& operator=(SubcompactionState&& o) {
     compaction = std::move(o.compaction);
     start = std::move(o.start);
     end = std::move(o.end);
@@ -138,14 +143,16 @@ struct CompactionJob::SubCompactionState {
     total_bytes = std::move(o.total_bytes);
     num_input_records = std::move(o.num_input_records);
     num_output_records = std::move(o.num_output_records);
+    compaction_job_stats = std::move(o.compaction_job_stats);
+    approx_size = std::move(o.approx_size);
     level_ptrs = std::move(o.level_ptrs);
     return *this;
   }
 
   // Because member unique_ptrs do not have these.
-  SubCompactionState(const SubCompactionState&) = delete;
+  SubcompactionState(const SubcompactionState&) = delete;
 
-  SubCompactionState& operator=(const SubCompactionState&) = delete;
+  SubcompactionState& operator=(const SubcompactionState&) = delete;
 };
 
 // Maintains state for the entire compaction
@@ -154,7 +161,7 @@ struct CompactionJob::CompactionState {
 
   // REQUIRED: subcompaction states are stored in order of increasing
   // key-range
-  std::vector<CompactionJob::SubCompactionState> sub_compact_states;
+  std::vector<CompactionJob::SubcompactionState> sub_compact_states;
   Status status;
 
   uint64_t total_bytes;
@@ -176,13 +183,11 @@ struct CompactionJob::CompactionState {
   }
 
   Slice SmallestUserKey() {
-    for (size_t i = 0; i < sub_compact_states.size(); i++) {
-      if (!sub_compact_states[i].outputs.empty()) {
-        return sub_compact_states[i].outputs[0].smallest.user_key();
+    for (auto& s : sub_compact_states) {
+      if (!s.outputs.empty()) {
+        return s.outputs[0].smallest.user_key();
       }
     }
-    // TODO(aekmekji): should we exit with an error if it reaches here?
-    assert(0);
     return Slice(nullptr, 0);
   }
 
@@ -193,19 +198,18 @@ struct CompactionJob::CompactionState {
         return sub_compact_states[i].current_output()->largest.user_key();
       }
     }
-    // TODO(aekmekji): should we exit with an error if it reaches here?
-    assert(0);
     return Slice(nullptr, 0);
   }
 };
 
 void CompactionJob::AggregateStatistics() {
-  for (SubCompactionState& sc : compact_->sub_compact_states) {
+  for (SubcompactionState& sc : compact_->sub_compact_states) {
     compact_->total_bytes += sc.total_bytes;
     compact_->num_input_records += sc.num_input_records;
     compact_->num_output_records += sc.num_output_records;
-
-    if (compaction_job_stats_) {
+  }
+  if (compaction_job_stats_) {
+    for (SubcompactionState& sc : compact_->sub_compact_states) {
       compaction_job_stats_->Add(sc.compaction_job_stats);
     }
   }
@@ -325,78 +329,141 @@ void CompactionJob::Prepare() {
     earliest_snapshot_ = existing_snapshots_[0];
   }
 
-  InitializeSubCompactions();
+  if (c->ShouldFormSubcompactions()) {
+    const uint64_t start_micros = env_->NowMicros();
+    GenSubcompactionBoundaries();
+    MeasureTime(stats_, SUBCOMPACTION_SETUP_TIME,
+                env_->NowMicros() - start_micros);
+
+    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]);
+    }
+  } else {
+    compact_->sub_compact_states.emplace_back(c, nullptr, nullptr);
+  }
 }
 
-// For L0-L1 compaction, iterators work in parallel by processing
-// different subsets of the full key range. This function sets up
-// the local states used by each of these subcompactions during
-// their execution
-void CompactionJob::InitializeSubCompactions() {
-  Compaction* c = compact_->compaction;
-  auto& bounds = sub_compaction_boundaries_;
-  if (c->IsSubCompaction()) {
-    auto* cmp = c->column_family_data()->user_comparator();
-    for (size_t which = 0; which < c->num_input_levels(); which++) {
-      if (c->level(which) == 1) {
-        const LevelFilesBrief* flevel = c->input_levels(which);
-        size_t num_files = flevel->num_files;
-
-        if (num_files > 1) {
-          std::vector<Slice> candidates;
-          auto& files = flevel->files;
-          Slice global_min = ExtractUserKey(files[0].smallest_key);
-          Slice global_max = ExtractUserKey(files[num_files - 1].largest_key);
+struct RangeWithSize {
+  Range range;
+  uint64_t size;
 
-          for (size_t i = 1; i < num_files; i++) {
-            // Make sure the smallest key in two consecutive L1 files are
-            // unique before adding the smallest key as a boundary. Also ensure
-            // that the boundary won't lead to an empty subcompaction (happens
-            // if the boundary == the smallest or largest key)
-            Slice s1 = ExtractUserKey(files[i].smallest_key);
-            Slice s2 = i == num_files - 1
-                          ? Slice()
-                          : ExtractUserKey(files[i + 1].smallest_key);
-
-            if ( (i == num_files - 1 && cmp->Compare(s1, global_max) < 0)
-              || (i < num_files - 1 && cmp->Compare(s1, s2) < 0 &&
-                    cmp->Compare(s1, global_min) > 0)) {
-              candidates.emplace_back(s1);
-            }
-          }
+  RangeWithSize(const Slice& a, const Slice& b, uint64_t s = 0)
+      : range(a, b), size(s) {}
+};
+
+bool SliceCompare(const Comparator* cmp, const Slice& a, const Slice& b) {
+  // Returns true if a < b
+  return cmp->Compare(ExtractUserKey(a), ExtractUserKey(b)) < 0;
+}
+
+// Generates a histogram representing potential divisions of key ranges from
+// the input. It adds the starting and/or ending keys of certain input files
+// to the working set and then finds the approximate size of data in between
+// each consecutive pair of slices. Then it divides these ranges into
+// consecutive groups such that each group has a similar size.
+void CompactionJob::GenSubcompactionBoundaries() {
+  auto* c = compact_->compaction;
+  auto* cfd = c->column_family_data();
+  std::set<Slice, std::function<bool(const Slice& a, const Slice& b)> > bounds(
+      std::bind(&SliceCompare, cfd->user_comparator(), std::placeholders::_1,
+                std::placeholders::_2));
+  int start_lvl = c->start_level();
+  int out_lvl = c->output_level();
+
+  // Add the starting and/or ending key of certain input files as a potential
+  // boundary (because we're inserting into a set, it avoids duplicates)
+  for (size_t lvl_idx = 0; lvl_idx < c->num_input_levels(); lvl_idx++) {
+    int lvl = c->level(lvl_idx);
+    if (lvl >= start_lvl && lvl <= out_lvl) {
+      const LevelFilesBrief* flevel = c->input_levels(lvl_idx);
+      size_t num_files = flevel->num_files;
+
+      if (num_files == 0) {
+        break;
+      }
 
-          // Divide the potential L1 file boundaries (those that passed the
-          // checks above) into 'max_subcompactions' groups such that each have
-          // as close to an equal number of files in it as possible
-          // TODO(aekmekji): refine this later to depend on file size
-          size_t files_left = candidates.size();
-          size_t subcompactions_left =
-              static_cast<size_t>(db_options_.max_subcompactions) < files_left
-                  ? db_options_.max_subcompactions
-                  : files_left;
-
-          size_t num_to_include;
-          size_t index = 0;
-
-          while (files_left > 1 && subcompactions_left > 1) {
-            num_to_include = files_left / subcompactions_left;
-            index += num_to_include;
-            sub_compaction_boundaries_.emplace_back(candidates[index]);
-            files_left -= num_to_include;
-            subcompactions_left--;
+      if (lvl == 0) {
+        // For level 0 add the starting and ending key of each file since the
+        // files may have greatly differing key ranges (not range-partitioned)
+        for (size_t i = 0; i < num_files; i++) {
+          bounds.emplace(flevel->files[i].smallest_key);
+          bounds.emplace(flevel->files[i].largest_key);
+        }
+      } else {
+        // For all other levels add the smallest/largest key in the level to
+        // encompass the range covered by that level
+        bounds.emplace(flevel->files[0].smallest_key);
+        bounds.emplace(flevel->files[num_files - 1].largest_key);
+        if (lvl == out_lvl) {
+          // For the last level include the starting keys of all files since
+          // the last level is the largest and probably has the widest key
+          // range. Since it's range partitioned, the ending key of one file
+          // and the starting key of the next are very close (or identical).
+          for (size_t i = 1; i < num_files; i++) {
+            bounds.emplace(flevel->files[i].smallest_key);
           }
         }
-        break;
       }
     }
   }
 
-  // Note: it's necessary for the first iterator sub-range to have
-  // start == nullptr and for the last to have end == nullptr
-  for (size_t i = 0; i <= bounds.size(); i++) {
-    Slice *start = i == 0 ? nullptr : &bounds[i - 1];
-    Slice *end = i == bounds.size() ? nullptr : &bounds[i];
-    compact_->sub_compact_states.emplace_back(compact_->compaction, start, end);
+  // Combine consecutive pairs of boundaries into ranges with an approximate
+  // size of data covered by keys in that range
+  uint64_t sum = 0;
+  std::vector<RangeWithSize> ranges;
+  auto* v = cfd->current();
+  for (auto it = bounds.begin();;) {
+    const Slice a = *it;
+    it++;
+
+    if (it == bounds.end()) {
+      break;
+    }
+
+    const Slice b = *it;
+    uint64_t size = versions_->ApproximateSize(v, a, b, start_lvl, out_lvl + 1);
+    ranges.emplace_back(a, b, size);
+    sum += size;
+  }
+
+  // Group the ranges into subcompactions
+  const double min_file_fill_percent = 4.0 / 5;
+  uint64_t max_output_files = std::ceil(
+      sum / min_file_fill_percent /
+      cfd->GetCurrentMutableCFOptions()->MaxFileSizeForLevel(out_lvl));
+  uint64_t subcompactions =
+      std::min({static_cast<uint64_t>(ranges.size()),
+                static_cast<uint64_t>(db_options_.max_subcompactions),
+                max_output_files});
+
+  double mean = sum * 1.0 / subcompactions;
+
+  if (subcompactions > 1) {
+    // Greedily add ranges to the subcompaction until the sum of the ranges'
+    // sizes becomes >= the expected mean size of a subcompaction
+    sum = 0;
+    for (size_t i = 0; i < ranges.size() - 1; i++) {
+      if (subcompactions == 1) {
+        // If there's only one left to schedule then it goes to the end so no
+        // need to put an end boundary
+        break;
+      }
+      sum += ranges[i].size;
+      if (sum >= mean) {
+        boundaries_.emplace_back(ExtractUserKey(ranges[i].range.limit));
+        sizes_.emplace_back(sum);
+        subcompactions--;
+        sum = 0;
+      }
+    }
+    sizes_.emplace_back(sum + ranges.back().size);
+  } else {
+    // Only one range so its size is the total sum of sizes computed above
+    sizes_.emplace_back(sum);
   }
 }
 
@@ -407,15 +474,35 @@ Status CompactionJob::Run() {
   log_buffer_->FlushBufferToLog();
   LogCompaction();
 
-  // Run each subcompaction sequentially
+  const size_t num_threads = compact_->sub_compact_states.size();
+  assert(num_threads > 0);
   const uint64_t start_micros = env_->NowMicros();
-  for (size_t i = 0; i < compact_->sub_compact_states.size(); i++) {
-    ProcessKeyValueCompaction(&compact_->sub_compact_states[i]);
+
+  // Launch a thread for each of subcompactions 1...num_threads-1
+  std::vector<std::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();
+  }
+
+  if (output_directory_ && !db_options_.disableDataSync) {
+    output_directory_->Fsync();
+  }
+
   compaction_stats_.micros = env_->NowMicros() - start_micros;
   MeasureTime(stats_, COMPACTION_TIME, compaction_stats_.micros);
 
-  // Determine if any of the subcompactions failed
+  // Check if any thread encountered an error during execution
   Status status;
   for (const auto& state : compact_->sub_compact_states) {
     if (!state.status.ok()) {
@@ -485,7 +572,8 @@ Status CompactionJob::Install(const MutableCFOptions& mutable_cf_options,
          << "num_output_files" << compact_->NumOutputFiles()
          << "total_output_size" << compact_->total_bytes
          << "num_input_records" << compact_->num_input_records
-         << "num_output_records" << compact_->num_output_records;
+         << "num_output_records" << compact_->num_output_records
+         << "num_subcompactions" << compact_->sub_compact_states.size();
 
   if (measure_io_stats_ && compaction_job_stats_ != nullptr) {
     stream << "file_write_nanos" << compaction_job_stats_->file_write_nanos;
@@ -507,7 +595,7 @@ Status CompactionJob::Install(const MutableCFOptions& mutable_cf_options,
   return status;
 }
 
-void CompactionJob::ProcessKeyValueCompaction(SubCompactionState* sub_compact) {
+void CompactionJob::ProcessKeyValueCompaction(SubcompactionState* sub_compact) {
   assert(sub_compact != nullptr);
   std::unique_ptr<Iterator> input_ptr(
       versions_->MakeInputIterator(sub_compact->compaction));
@@ -764,10 +852,6 @@ void CompactionJob::ProcessKeyValueCompaction(SubCompactionState* sub_compact) {
   if (status.ok()) {
     status = input->status();
   }
-  if (output_directory_ && !db_options_.disableDataSync) {
-    // TODO(aekmekji): Maybe only call once after all subcompactions complete?
-    output_directory_->Fsync();
-  }
 
   if (measure_io_stats_) {
     sub_compact->compaction_job_stats.file_write_nanos +=
@@ -788,9 +872,9 @@ void CompactionJob::ProcessKeyValueCompaction(SubCompactionState* sub_compact) {
 }
 
 Status CompactionJob::WriteKeyValue(const Slice& key, const Slice& value,
-    const ParsedInternalKey& ikey, const Status& input_status,
-    SubCompactionState* sub_compact) {
-
+                                    const ParsedInternalKey& ikey,
+                                    const Status& input_status,
+                                    SubcompactionState* sub_compact) {
   Slice newkey(key.data(), key.size());
   std::string kstr;
 
@@ -833,6 +917,10 @@ Status CompactionJob::WriteKeyValue(const Slice& key, const Slice& value,
     std::max(sub_compact->current_output()->largest_seqno, seqno);
 
   // Close output file if it is big enough
+  // TODO(aekmekji): determine if file should be closed earlier than this
+  // during subcompactions (i.e. if output size, estimated by input size, is
+  // going to be 1.2MB and max_output_file_size = 1MB, prefer to have 0.6MB
+  // and 0.6MB instead of 1MB and 0.2MB)
   Status status;
   if (sub_compact->builder->FileSize() >=
       sub_compact->compaction->max_output_file_size()) {
@@ -867,8 +955,8 @@ void CompactionJob::RecordDroppedKeys(
   }
 }
 
-Status CompactionJob::FinishCompactionOutputFile(const Status& input_status,
-                                          SubCompactionState* sub_compact) {
+Status CompactionJob::FinishCompactionOutputFile(
+    const Status& input_status, SubcompactionState* sub_compact) {
   AutoThreadOperationStageUpdater stage_updater(
       ThreadStatus::STAGE_COMPACTION_SYNC_FILE);
   assert(sub_compact != nullptr);
@@ -975,9 +1063,9 @@ Status CompactionJob::InstallCompactionResults(
   // Add compaction outputs
   compaction->AddInputDeletions(compact_->compaction->edit());
 
-  for (SubCompactionState& sub_compact : compact_->sub_compact_states) {
+  for (SubcompactionState& sub_compact : compact_->sub_compact_states) {
     for (size_t i = 0; i < sub_compact.outputs.size(); i++) {
-      const SubCompactionState::Output& out = sub_compact.outputs[i];
+      const SubcompactionState::Output& out = sub_compact.outputs[i];
       compaction->edit()->AddFile(compaction->output_level(), out.number,
                                   out.path_id, out.file_size, out.smallest,
                                   out.largest, out.smallest_seqno,
@@ -1028,8 +1116,8 @@ void CompactionJob::RecordCompactionIOStats() {
   IOSTATS_RESET(bytes_written);
 }
 
-Status CompactionJob::OpenCompactionOutputFile(SubCompactionState*
-                                                                sub_compact) {
+Status CompactionJob::OpenCompactionOutputFile(
+    SubcompactionState* sub_compact) {
   assert(sub_compact != nullptr);
   assert(sub_compact->builder == nullptr);
   // no need to lock because VersionSet::next_file_number_ is atomic
@@ -1048,7 +1136,7 @@ Status CompactionJob::OpenCompactionOutputFile(SubCompactionState*
     LogFlush(db_options_.info_log);
     return s;
   }
-  SubCompactionState::Output out;
+  SubcompactionState::Output out;
   out.number = file_number;
   out.path_id = sub_compact->compaction->output_path_id();
   out.smallest.Clear();
@@ -1083,7 +1171,7 @@ Status CompactionJob::OpenCompactionOutputFile(SubCompactionState*
 }
 
 void CompactionJob::CleanupCompaction() {
-  for (SubCompactionState& sub_compact : compact_->sub_compact_states) {
+  for (SubcompactionState& sub_compact : compact_->sub_compact_states) {
     const auto& sub_status = sub_compact.status;
 
     if (sub_compact.builder != nullptr) {
@@ -1094,7 +1182,7 @@ void CompactionJob::CleanupCompaction() {
       assert(!sub_status.ok() || sub_compact.outfile == nullptr);
     }
     for (size_t i = 0; i < sub_compact.outputs.size(); i++) {
-      const SubCompactionState::Output& out = sub_compact.outputs[i];
+      const SubcompactionState::Output& out = sub_compact.outputs[i];
 
       // If this file was inserted into the table cache then remove
       // them here because this compaction was not committed.

db/compaction_job.h

@@ -79,32 +79,31 @@ class CompactionJob {
                  InstrumentedMutex* db_mutex);
 
  private:
-  struct SubCompactionState;
+  struct SubcompactionState;
 
   void AggregateStatistics();
-  // Set up the individual states used by each subcompaction
-  void InitializeSubCompactions();
+  void GenSubcompactionBoundaries();
 
   // update the thread status for starting a compaction.
   void ReportStartedCompaction(Compaction* compaction);
   void AllocateCompactionOutputFileNumbers();
   // Call compaction filter. Then iterate through input and compact the
   // kv-pairs
-  void ProcessKeyValueCompaction(SubCompactionState* sub_compact);
+  void ProcessKeyValueCompaction(SubcompactionState* sub_compact);
 
   Status WriteKeyValue(const Slice& key, const Slice& value,
                        const ParsedInternalKey& ikey,
                        const Status& input_status,
-                       SubCompactionState* sub_compact);
+                       SubcompactionState* sub_compact);
 
   Status FinishCompactionOutputFile(const Status& input_status,
-                                    SubCompactionState* sub_compact);
+                                    SubcompactionState* sub_compact);
   Status InstallCompactionResults(const MutableCFOptions& mutable_cf_options,
                                   InstrumentedMutex* db_mutex);
   SequenceNumber findEarliestVisibleSnapshot(SequenceNumber in,
                                              SequenceNumber* prev_snapshot);
   void RecordCompactionIOStats();
-  Status OpenCompactionOutputFile(SubCompactionState* sub_compact);
+  Status OpenCompactionOutputFile(SubcompactionState* sub_compact);
   void CleanupCompaction();
   void UpdateCompactionJobStats(
     const InternalStats::CompactionStats& stats) const;
@@ -156,7 +155,10 @@ class CompactionJob {
 
   bool paranoid_file_checks_;
   bool measure_io_stats_;
-  std::vector<Slice> sub_compaction_boundaries_;
+  // Stores the Slices that designate the boundaries for each subcompaction
+  std::vector<Slice> boundaries_;
+  // Stores the approx size of keys covered in the range of each subcompaction
+  std::vector<uint64_t> sizes_;
 };
 
 }  // namespace rocksdb

db/compaction_job_stats_test.cc

@@ -758,8 +758,6 @@ TEST_P(CompactionJobStatsTest, CompactionJobStatsTest) {
               num_keys_per_L0_file));
       ASSERT_EQ(stats_checker->NumberOfUnverifiedStats(), 1U);
       Compact(1, smallest_key, largest_key);
-      // TODO(aekmekji): account for whether parallel L0-L1 compaction is
-      // enabled or not. If so then num_L1_files will increase by 1
       if (options.max_subcompactions == 1) {
         --num_L1_files;
       }

db/db_test.cc

@@ -7224,7 +7224,7 @@ TEST_F(DBTest, DynamicLevelCompressionPerLevel2) {
   ASSERT_GT(num_zlib.load(), 0);
 }
 
-TEST_P(DBTestWithParam, DynamicCompactionOptions) {
+TEST_F(DBTest, DynamicCompactionOptions) {
   // minimum write buffer size is enforced at 64KB
   const uint64_t k32KB = 1 << 15;
   const uint64_t k64KB = 1 << 16;
@@ -7250,7 +7250,6 @@ TEST_P(DBTestWithParam, DynamicCompactionOptions) {
   options.target_file_size_multiplier = 1;
   options.max_bytes_for_level_base = k128KB;
   options.max_bytes_for_level_multiplier = 4;
-  options.max_subcompactions = max_subcompactions_;
 
   // Block flush thread and disable compaction thread
   env_->SetBackgroundThreads(1, Env::LOW);

db/version_set.cc

@@ -3002,15 +3002,27 @@ bool VersionSet::ManifestContains(uint64_t manifest_file_num,
   return result;
 }
 
+// TODO(aekmekji): in CompactionJob::GenSubcompactionBoundaries(), this
+// function is called repeatedly with consecutive pairs of slices. For example
+// if the slice list is [a, b, c, d] this function is called with arguments
+// (a,b) then (b,c) then (c,d). Knowing this, an optimization is possible where
+// we avoid doing binary search for the keys b and c twice and instead somehow
+// maintain state of where they first appear in the files.
 uint64_t VersionSet::ApproximateSize(Version* v, const Slice& start,
-                                     const Slice& end) {
+                                     const Slice& end, int start_level,
+                                     int end_level) {
   // pre-condition
   assert(v->cfd_->internal_comparator().Compare(start, end) <= 0);
 
   uint64_t size = 0;
   const auto* vstorage = v->storage_info();
+  end_level = end_level == -1
+                  ? vstorage->num_non_empty_levels()
+                  : std::min(end_level, vstorage->num_non_empty_levels());
 
-  for (int level = 0; level < vstorage->num_non_empty_levels(); level++) {
+  assert(start_level <= end_level);
+
+  for (int level = start_level; level < end_level; level++) {
     const LevelFilesBrief& files_brief = vstorage->LevelFilesBrief(level);
     if (!files_brief.num_files) {
       // empty level, skip exploration
@@ -3142,7 +3154,7 @@ Iterator* VersionSet::MakeInputIterator(Compaction* c) {
   read_options.verify_checksums =
     c->mutable_cf_options()->verify_checksums_in_compaction;
   read_options.fill_cache = false;
-  if (c->IsSubCompaction()) {
+  if (c->ShouldFormSubcompactions()) {
     read_options.total_order_seek = true;
   }
 

db/version_set.h

@@ -646,7 +646,10 @@ class VersionSet {
   void AddLiveFiles(std::vector<FileDescriptor>* live_list);
 
   // Return the approximate size of data to be scanned for range [start, end)
-  uint64_t ApproximateSize(Version* v, const Slice& start, const Slice& end);
+  // in levels [start_level, end_level). If end_level == 0 it will search
+  // through all non-empty levels
+  uint64_t ApproximateSize(Version* v, const Slice& start, const Slice& end,
+                           int start_level = 0, int end_level = -1);
 
   // Return the size of the current manifest file
   uint64_t manifest_file_size() const { return manifest_file_size_; }

include/rocksdb/options.h

@@ -907,12 +907,9 @@ struct DBOptions {
   int max_background_compactions;
 
   // This integer represents the maximum number of threads that will
-  // concurrently perform a level-based compaction from L0 to L1. A value
-  // of 1 means there is no parallelism, and a greater number enables a
-  // multi-threaded version of the L0-L1 compaction that divides the compaction
-  // into multiple, smaller ones that are run simultaneously. This is still
-  // under development and is only available for level-based compaction.
-  // Default: 1
+  // concurrently perform a compaction job by breaking it into multiple,
+  // smaller ones that are run simultaneously.
+  // Default: 1 (i.e. no subcompactions)
   uint32_t max_subcompactions;
 
   // Maximum number of concurrent background memtable flush jobs, submitted to

include/rocksdb/statistics.h

@@ -233,6 +233,7 @@ enum Histograms : uint32_t {
   DB_GET = 0,
   DB_WRITE,
   COMPACTION_TIME,
+  SUBCOMPACTION_SETUP_TIME,
   TABLE_SYNC_MICROS,
   COMPACTION_OUTFILE_SYNC_MICROS,
   WAL_FILE_SYNC_MICROS,
@@ -259,6 +260,7 @@ const std::vector<std::pair<Histograms, std::string>> HistogramsNameMap = {
     {DB_GET, "rocksdb.db.get.micros"},
     {DB_WRITE, "rocksdb.db.write.micros"},
     {COMPACTION_TIME, "rocksdb.compaction.times.micros"},
+    {SUBCOMPACTION_SETUP_TIME, "rocksdb.subcompaction.setup.times.micros"},
     {TABLE_SYNC_MICROS, "rocksdb.table.sync.micros"},
     {COMPACTION_OUTFILE_SYNC_MICROS, "rocksdb.compaction.outfile.sync.micros"},
     {WAL_FILE_SYNC_MICROS, "rocksdb.wal.file.sync.micros"},