Allow more intra-L0 compaction when L0 is small (#12214)

Summary:
introduce a new option `intra_l0_compaction_size` to allow more intra-L0 compaction when total L0 size is under a threshold. This option applies only to leveled compaction. It is enabled by default and set to `max_bytes_for_level_base / max_bytes_for_level_multiplier` only for atomic_flush users. When atomic_flush=true, it is more likely that some CF's total L0 size is small when it's eligible for compaction. This option aims to reduce write amplification in this case.

Pull Request resolved: facebook/rocksdb#12214

Test Plan:
- new unit test
- benchmark:
```
TEST_TMPDIR=/dev/shm ./db_bench --benchmarks=fillrandom --write_buffer_size=51200 --max_bytes_for_level_base=5242880 --level0_file_num_compaction_trigger=4 --statistics=1

main:
fillrandom   :     234.499 micros/op 4264 ops/sec 234.499 seconds 1000000 operations;    0.5 MB/s
rocksdb.compact.read.bytes COUNT : 1490756235
rocksdb.compact.write.bytes COUNT : 1469056734
rocksdb.flush.write.bytes COUNT : 71099011

branch:
fillrandom   :     128.494 micros/op 7782 ops/sec 128.494 seconds 1000000 operations;    0.9 MB/s
rocksdb.compact.read.bytes COUNT : 807474156
rocksdb.compact.write.bytes COUNT : 781977610
rocksdb.flush.write.bytes COUNT : 71098785
```

Reviewed By: ajkr

Differential Revision: D52637771

Pulled By: cbi42

fbshipit-source-id: 4f2c7925d0c3a718635c948ea0d4981ed9fabec3
(cherry picked from commit 4b684e96b71cd6f1d50e29fae8b55c323ccd2869)

db/compaction/compaction_picker_level.cc

@@ -106,6 +106,15 @@ class LevelCompactionBuilder {
   // otherwise, returns false.
   bool PickIntraL0Compaction();
 
+  // When total L0 size is small compared to Lbase, try to pick intra-L0
+  // compaction starting from the newest L0 file. This helps to prevent
+  // L0->Lbase compaction with large write-amp.
+  //
+  // Returns true iff an intra-L0 compaction is picked.
+  // `start_level_inputs_` and `output_level_` will be updated accordingly if
+  // a compaction is picked.
+  bool PickSizeBasedIntraL0Compaction();
+
   // Return true if TrivialMove is extended. `start_index` is the index of
   // the initial file picked, which should already be in `start_level_inputs_`.
   bool TryExtendNonL0TrivialMove(int start_index,
@@ -758,6 +767,9 @@ bool LevelCompactionBuilder::PickFileToCompact() {
   // being compacted at level 0.
   if (start_level_ == 0 &&
       !compaction_picker_->level0_compactions_in_progress()->empty()) {
+    if (PickSizeBasedIntraL0Compaction()) {
+      return true;
+    }
     TEST_SYNC_POINT("LevelCompactionPicker::PickCompactionBySize:0");
     return false;
   }
@@ -770,6 +782,9 @@ bool LevelCompactionBuilder::PickFileToCompact() {
   if (TryPickL0TrivialMove()) {
     return true;
   }
+  if (start_level_ == 0 && PickSizeBasedIntraL0Compaction()) {
+    return true;
+  }
 
   const std::vector<FileMetaData*>& level_files =
       vstorage_->LevelFiles(start_level_);
@@ -874,6 +889,56 @@ bool LevelCompactionBuilder::PickIntraL0Compaction() {
                                mutable_cf_options_.max_compaction_bytes,
                                &start_level_inputs_);
 }
+
+bool LevelCompactionBuilder::PickSizeBasedIntraL0Compaction() {
+  assert(start_level_ == 0);
+  int base_level = vstorage_->base_level();
+  if (base_level <= 0) {
+    return false;
+  }
+  const std::vector<FileMetaData*>& l0_files =
+      vstorage_->LevelFiles(/*level=*/0);
+  size_t min_num_file =
+      std::max(2, mutable_cf_options_.level0_file_num_compaction_trigger);
+  if (l0_files.size() < min_num_file) {
+    return false;
+  }
+  uint64_t l0_size = 0;
+  for (const auto& file : l0_files) {
+    l0_size += file->fd.GetFileSize();
+  }
+  const uint64_t min_lbase_size =
+      l0_size * static_cast<uint64_t>(std::max(
+                    10.0, mutable_cf_options_.max_bytes_for_level_multiplier));
+  assert(min_lbase_size >= l0_size);
+  const std::vector<FileMetaData*>& lbase_files =
+      vstorage_->LevelFiles(/*level=*/base_level);
+  uint64_t lbase_size = 0;
+  for (const auto& file : lbase_files) {
+    lbase_size += file->fd.GetFileSize();
+    if (lbase_size > min_lbase_size) {
+      break;
+    }
+  }
+  if (lbase_size <= min_lbase_size) {
+    return false;
+  }
+
+  start_level_inputs_.clear();
+  start_level_inputs_.level = 0;
+  for (const auto& file : l0_files) {
+    if (file->being_compacted) {
+      break;
+    }
+    start_level_inputs_.files.push_back(file);
+  }
+  if (start_level_inputs_.files.size() < min_num_file) {
+    start_level_inputs_.clear();
+    return false;
+  }
+  output_level_ = 0;
+  return true /* picked an intra-L0 compaction */;
+}
 }  // namespace
 
 Compaction* LevelCompactionPicker::PickCompaction(

db/compaction/compaction_picker_test.cc

@@ -4152,6 +4152,58 @@ TEST_P(PerKeyPlacementCompactionPickerTest,
 INSTANTIATE_TEST_CASE_P(PerKeyPlacementCompactionPickerTest,
                         PerKeyPlacementCompactionPickerTest, ::testing::Bool());
 
+TEST_F(CompactionPickerTest, IntraL0WhenL0IsSmall) {
+  mutable_cf_options_.level0_file_num_compaction_trigger = 4;
+  mutable_cf_options_.max_bytes_for_level_multiplier = 10;
+  for (const uint64_t lbase_size_multiplier : {1, 10, 11, 40}) {
+    SCOPED_TRACE("lbase_size_multiplier=" +
+                 std::to_string(lbase_size_multiplier));
+    NewVersionStorage(6, kCompactionStyleLevel);
+    // When L0 size is <= Lbase size / max_bytes_for_level_multiplier,
+    // intra-L0 compaction is picked. Otherwise, L0->L1
+    // compaction is picked.
+    Add(/*level=*/0, /*file_number=*/1U, /*smallest=*/"100",
+        /*largest=*/"200", /*file_size=*/1000, /*path_id=*/0,
+        /*smallest_seq=*/10, /*largest_seq=*/11,
+        /*compensated_file_size=*/1000);
+    Add(/*level=*/0, /*file_number=*/2U, /*smallest=*/"100",
+        /*largest=*/"100", /*file_size=*/1000, /*path_id=*/0,
+        /*smallest_seq=*/20, /*largest_seq=*/21,
+        /*compensated_file_size=*/1000);
+    Add(/*level=*/0, /*file_number=*/3U, /*smallest=*/"100",
+        /*largest=*/"200", /*file_size=*/1000, /*path_id=*/0,
+        /*smallest_seq=*/30, /*largest_seq=*/31,
+        /*compensated_file_size=*/1000);
+    Add(/*level=*/0, /*file_number=*/4U, /*smallest=*/"100",
+        /*largest=*/"200", /*file_size=*/1000, /*path_id=*/0,
+        /*smallest_seq=*/40, /*largest_seq=*/41,
+        /*compensated_file_size=*/1000);
+    const uint64_t l0_size = 4000;
+    const uint64_t lbase_size = l0_size * lbase_size_multiplier;
+    Add(/*level=*/1, /*file_number=*/5U, /*smallest=*/"100",
+        /*largest=*/"200", /*file_size=*/lbase_size, /*path_id=*/0,
+        /*smallest_seq=*/0, /*largest_seq=*/0,
+        /*compensated_file_size=*/lbase_size);
+    UpdateVersionStorageInfo();
+
+    LevelCompactionPicker compaction_picker(ioptions_, &icmp_);
+    std::unique_ptr<Compaction> compaction(compaction_picker.PickCompaction(
+        cf_name_, mutable_cf_options_, mutable_db_options_, vstorage_.get(),
+        &log_buffer_));
+    ASSERT_TRUE(compaction.get() != nullptr);
+    ASSERT_EQ(CompactionReason::kLevelL0FilesNum,
+              compaction->compaction_reason());
+    ASSERT_EQ(4U, compaction->num_input_files(0));
+    if (lbase_size_multiplier >
+        mutable_cf_options_.max_bytes_for_level_multiplier) {
+      ASSERT_EQ(1U, compaction->num_input_levels());
+      ASSERT_EQ(0, compaction->output_level());
+    } else {
+      ASSERT_EQ(2U, compaction->num_input_levels());
+      ASSERT_EQ(1, compaction->output_level());
+    }
+  }
+}
 
 }  // namespace ROCKSDB_NAMESPACE
 

unreleased_history/behavior_changes/more-intra-l0-compaction.md

@@ -0,0 +1 @@
+* For leveled compaction, RocksDB will try to do intra-L0 compaction if the total L0 size is small compared to Lbase (#12214). Users with atomic_flush=true are more likely to see the impact of this change.