Always allow L0->L1 trivial move during manual compaction (#11375)

Summary:
during manual compaction (CompactRange()), L0->L1 trivial move is disabled when only L0 overlaps with compacting key range (introduced in https://github.com/facebook/rocksdb/issues/7368 to enforce kForce* contract). This can cause large memory usage due to compaction readahead when number of L0 files is large. This PR allows L0->L1 trivial move in this case, and will do a L1 -> L1 intra-level compaction when needed (`bottommost_level_compaction` is kForce*). In brief, consider a DB with only L0 file, and user calls CompactRange(kForce, nullptr, nullptr),
- before this PR, RocksDB does a L0 -> L1 compaction (disallow trivial move),
- after this PR, RocksDB does a L0 -> L1 compaction (allow trivial move), and a L1 -> L1 compaction.
Users can use kForceOptimized to avoid this extra L1->L1 compaction overhead when L0s are overlapping and cannot be trivial moved.

This PR also fixed a bug (see previous discussion in https://github.com/facebook/rocksdb/issues/11041) where `final_output_level` of a manual compaction can be miscalculated when `level_compaction_dynamic_level_bytes=true`. This bug could cause incorrect level being moved when CompactRangeOptions::change_level is specified.

Pull Request resolved: https://github.com/facebook/rocksdb/pull/11375

Test Plan: - Added new unit tests to test that L0 -> L1 compaction allows trivial move and L1 -> L1 compaction is done when needed.

Reviewed By: ajkr

Differential Revision: D44943518

Pulled By: cbi42

fbshipit-source-id: e9fb770d17b163c18a623e1d1bd6b81159192708

HISTORY.md

@@ -9,6 +9,7 @@
 * For level compaction with `level_compaction_dynamic_level_bytes=true`, RocksDB now trivially moves levels down to fill LSM starting from bottommost level during DB open. See more in comments for option `level_compaction_dynamic_level_bytes` (#11321).
 * User-provided `ReadOptions` take effect for more reads of non-`CacheEntryRole::kDataBlock` blocks.
 * For level compaction with `level_compaction_dynamic_level_bytes=true`, RocksDB now drains unnecessary levels through background compaction automatically (#11340). This together with #11321 makes it automatic to migrate other compaction settings to level compaction with `level_compaction_dynamic_level_bytes=true`. In addition, a live DB that becomes smaller will now have unnecessary levels drained which can help to reduce read and space amp.
+* If `CompactRange()` is called with `CompactRangeOptions::bottommost_level_compaction=kForce*` to compact from L0 to L1, RocksDB now will try to do trivial move from L0 to L1 and then do an intra L1 compaction, instead of a L0 to L1 compaction with trivial move disabled (#11375)).
 
 ### Bug Fixes
 * In the DB::VerifyFileChecksums API, ensure that file system reads of SST files are equal to the readahead_size in ReadOptions, if specified. Previously, each read was 2x the readahead_size.

db/db_compaction_filter_test.cc

@@ -742,7 +742,7 @@ TEST_F(DBTestCompactionFilter, CompactionFilterContextCfId) {
   ASSERT_TRUE(filter->compaction_filter_created());
 }
 
-// Compaction filters aplies to all records, regardless snapshots.
+// Compaction filters applies to all records, regardless snapshots.
 TEST_F(DBTestCompactionFilter, CompactionFilterIgnoreSnapshot) {
   std::string five = std::to_string(5);
   Options options = CurrentOptions();

db/db_compaction_test.cc

@@ -136,11 +136,12 @@ class DBCompactionTestWithParam
 
 class DBCompactionTestWithBottommostParam
     : public DBTestBase,
-      public testing::WithParamInterface<BottommostLevelCompaction> {
+      public testing::WithParamInterface<
+          std::tuple<BottommostLevelCompaction, bool>> {
  public:
   DBCompactionTestWithBottommostParam()
       : DBTestBase("db_compaction_test", /*env_do_fsync=*/true) {
-    bottommost_level_compaction_ = GetParam();
+    bottommost_level_compaction_ = std::get<0>(GetParam());
   }
 
   BottommostLevelCompaction bottommost_level_compaction_;
@@ -7339,10 +7340,63 @@ TEST_P(DBCompactionTestL0FilesMisorderCorruptionWithParam,
   Destroy(options_);
 }
 
+TEST_F(DBCompactionTest, SingleLevelUniveresal) {
+  // Tests that manual compaction works with single level universal compaction.
+  Options options = CurrentOptions();
+  options.compaction_style = kCompactionStyleUniversal;
+  options.disable_auto_compactions = true;
+  options.num_levels = 1;
+  DestroyAndReopen(options);
+
+  Random rnd(31);
+  for (int i = 0; i < 10; ++i) {
+    for (int j = 0; j < 50; ++j) {
+      ASSERT_OK(Put(Key(i * 100 + j), rnd.RandomString(50)));
+    }
+    ASSERT_OK(Flush());
+  }
+  ASSERT_EQ(NumTableFilesAtLevel(0), 10);
+  ASSERT_OK(db_->CompactRange(CompactRangeOptions(), nullptr, nullptr));
+  ASSERT_EQ(NumTableFilesAtLevel(0), 1);
+}
+
+TEST_F(DBCompactionTest, SingleOverlappingNonL0BottommostManualCompaction) {
+  // Tests that manual compact will rewrite bottommost level
+  // when there is only a single non-L0 level that overlaps with
+  // manual compaction range.
+  constexpr int kSstNum = 10;
+  Options options = CurrentOptions();
+  options.disable_auto_compactions = true;
+  options.num_levels = 7;
+  for (auto b : {BottommostLevelCompaction::kForce,
+                 BottommostLevelCompaction::kForceOptimized}) {
+    DestroyAndReopen(options);
+
+    // Generate some sst files on level 0 with sequence keys (no overlap)
+    for (int i = 0; i < kSstNum; i++) {
+      for (int j = 1; j < UCHAR_MAX; j++) {
+        auto key = std::string(kSstNum, '\0');
+        key[kSstNum - i] += static_cast<char>(j);
+        ASSERT_OK(Put(key, std::string(i % 1000, 'A')));
+      }
+      ASSERT_OK(Flush());
+    }
+    MoveFilesToLevel(4);
+    ASSERT_EQ(NumTableFilesAtLevel(4), kSstNum);
+    CompactRangeOptions cro;
+    cro.bottommost_level_compaction = b;
+    ASSERT_OK(db_->CompactRange(cro, nullptr, nullptr));
+    ASSERT_EQ(NumTableFilesAtLevel(4), 1);
+  }
+}
+
 TEST_P(DBCompactionTestWithBottommostParam, SequenceKeysManualCompaction) {
   constexpr int kSstNum = 10;
   Options options = CurrentOptions();
   options.disable_auto_compactions = true;
+  options.num_levels = 7;
+  const bool dynamic_level = std::get<1>(GetParam());
+  options.level_compaction_dynamic_level_bytes = dynamic_level;
   DestroyAndReopen(options);
 
   // Generate some sst files on level 0 with sequence keys (no overlap)
@@ -7360,25 +7414,42 @@ TEST_P(DBCompactionTestWithBottommostParam, SequenceKeysManualCompaction) {
 
   auto cro = CompactRangeOptions();
   cro.bottommost_level_compaction = bottommost_level_compaction_;
+  bool trivial_moved = false;
+  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->SetCallBack(
+      "DBImpl::BackgroundCompaction:TrivialMove",
+      [&](void* /*arg*/) { trivial_moved = true; });
+  ROCKSDB_NAMESPACE::SyncPoint::GetInstance()->EnableProcessing();
+  // All bottommost_level_compaction options should allow l0 -> l1 trivial move.
   ASSERT_OK(db_->CompactRange(cro, nullptr, nullptr));
+  ASSERT_TRUE(trivial_moved);
   if (bottommost_level_compaction_ == BottommostLevelCompaction::kForce ||
       bottommost_level_compaction_ ==
           BottommostLevelCompaction::kForceOptimized) {
-    // Real compaction to compact all sst files from level 0 to 1 file on level
+    // bottommost level should go through intra-level compaction
-    // 1
+    // and has only 1 file
-    ASSERT_EQ("0,1", FilesPerLevel(0));
+    if (dynamic_level) {
+      ASSERT_EQ("0,0,0,0,0,0,1", FilesPerLevel(0));
+    } else {
+      ASSERT_EQ("0,1", FilesPerLevel(0));
+    }
   } else {
-    // Just trivial move from level 0 -> 1
+    // Just trivial move from level 0 -> 1/base
-    ASSERT_EQ("0," + std::to_string(kSstNum), FilesPerLevel(0));
+    if (dynamic_level) {
+      ASSERT_EQ("0,0,0,0,0,0," + std::to_string(kSstNum), FilesPerLevel(0));
+    } else {
+      ASSERT_EQ("0," + std::to_string(kSstNum), FilesPerLevel(0));
+    }
   }
 }
 
 INSTANTIATE_TEST_CASE_P(
     DBCompactionTestWithBottommostParam, DBCompactionTestWithBottommostParam,
-    ::testing::Values(BottommostLevelCompaction::kSkip,
+    ::testing::Combine(
-                      BottommostLevelCompaction::kIfHaveCompactionFilter,
+        ::testing::Values(BottommostLevelCompaction::kSkip,
-                      BottommostLevelCompaction::kForce,
+                          BottommostLevelCompaction::kIfHaveCompactionFilter,
-                      BottommostLevelCompaction::kForceOptimized));
+                          BottommostLevelCompaction::kForce,
+                          BottommostLevelCompaction::kForceOptimized),
+        ::testing::Bool()));
 
 TEST_F(DBCompactionTest, UpdateLevelSubCompactionTest) {
   Options options = CurrentOptions();

db/db_impl/db_impl.h

@@ -734,13 +734,17 @@ class DBImpl : public DB {
   // max_file_num_to_ignore allows bottom level compaction to filter out newly
   // compacted SST files. Setting max_file_num_to_ignore to kMaxUint64 will
   // disable the filtering
+  // If `final_output_level` is not nullptr, it is set to manual compaction's
+  // output level if returned status is OK, and it may or may not be set to
+  // manual compaction's output level if returned status is not OK.
   Status RunManualCompaction(ColumnFamilyData* cfd, int input_level,
                              int output_level,
                              const CompactRangeOptions& compact_range_options,
                              const Slice* begin, const Slice* end,
                              bool exclusive, bool disallow_trivial_move,
                              uint64_t max_file_num_to_ignore,
-                             const std::string& trim_ts);
+                             const std::string& trim_ts,
+                             int* final_output_level = nullptr);
 
   // Return an internal iterator over the current state of the database.
   // The keys of this iterator are internal keys (see format.h).

db/db_impl/db_impl_compaction_flush.cc

@@ -1054,8 +1054,8 @@ Status DBImpl::CompactRangeInternal(const CompactRangeOptions& options,
     }
     s = RunManualCompaction(cfd, ColumnFamilyData::kCompactAllLevels,
                             final_output_level, options, begin, end, exclusive,
-                            false, std::numeric_limits<uint64_t>::max(),
+                            false /* disable_trivial_move */,
-                            trim_ts);
+                            std::numeric_limits<uint64_t>::max(), trim_ts);
   } else {
     int first_overlapped_level = kInvalidLevel;
     int max_overlapped_level = kInvalidLevel;
@@ -1142,74 +1142,83 @@ Status DBImpl::CompactRangeInternal(const CompactRangeOptions& options,
       CleanupSuperVersion(super_version);
     }
     if (s.ok() && first_overlapped_level != kInvalidLevel) {
-      // max_file_num_to_ignore can be used to filter out newly created SST
+      if (cfd->ioptions()->compaction_style == kCompactionStyleUniversal ||
-      // files, useful for bottom level compaction in a manual compaction
+          cfd->ioptions()->compaction_style == kCompactionStyleFIFO) {
-      uint64_t max_file_num_to_ignore = std::numeric_limits<uint64_t>::max();
+        assert(first_overlapped_level == 0);
-      uint64_t next_file_number = versions_->current_next_file_number();
+        s = RunManualCompaction(
-      final_output_level = max_overlapped_level;
+            cfd, first_overlapped_level, first_overlapped_level, options, begin,
-      int output_level;
+            end, exclusive, true /* disallow_trivial_move */,
-      for (int level = first_overlapped_level; level <= max_overlapped_level;
+            std::numeric_limits<uint64_t>::max() /* max_file_num_to_ignore */,
-           level++) {
+            trim_ts);
-        bool disallow_trivial_move = false;
+        final_output_level = max_overlapped_level;
-        // in case the compaction is universal or if we're compacting the
+      } else {
-        // bottom-most level, the output level will be the same as input one.
+        assert(cfd->ioptions()->compaction_style == kCompactionStyleLevel);
-        // level 0 can never be the bottommost level (i.e. if all files are in
+        uint64_t next_file_number = versions_->current_next_file_number();
-        // level 0, we will compact to level 1)
+        // Start compaction from `first_overlapped_level`, one level down at a
-        if (cfd->ioptions()->compaction_style == kCompactionStyleUniversal ||
+        // time, until output level >= max_overlapped_level.
-            cfd->ioptions()->compaction_style == kCompactionStyleFIFO) {
+        // When max_overlapped_level == 0, we will still compact from L0 -> L1
-          output_level = level;
+        // (or LBase), and followed by a bottommost level intra-level compaction
-        } else if (level == max_overlapped_level && level > 0) {
+        // at L1 (or LBase), if applicable.
-          if (options.bottommost_level_compaction ==
+        int level = first_overlapped_level;
-              BottommostLevelCompaction::kSkip) {
+        final_output_level = level;
-            // Skip bottommost level compaction
+        int output_level, base_level;
-            continue;
+        while (level < max_overlapped_level || level == 0) {
-          } else if (options.bottommost_level_compaction ==
-                         BottommostLevelCompaction::kIfHaveCompactionFilter &&
-                     cfd->ioptions()->compaction_filter == nullptr &&
-                     cfd->ioptions()->compaction_filter_factory == nullptr) {
-            // Skip bottommost level compaction since we don't have a compaction
-            // filter
-            continue;
-          }
-          output_level = level;
-          // update max_file_num_to_ignore only for bottom level compaction
-          // because data in newly compacted files in middle levels may still
-          // need to be pushed down
-          max_file_num_to_ignore = next_file_number;
-        } else {
           output_level = level + 1;
-          if (cfd->ioptions()->compaction_style == kCompactionStyleLevel &&
+          if (cfd->ioptions()->level_compaction_dynamic_level_bytes &&
-              cfd->ioptions()->level_compaction_dynamic_level_bytes &&
               level == 0) {
             output_level = ColumnFamilyData::kCompactToBaseLevel;
           }
-          // if it's a BottommostLevel compaction and `kForce*` compaction is
+          // Use max value for `max_file_num_to_ignore` to always compact
-          // set, disallow trivial move
+          // files down.
-          if (level == max_overlapped_level &&
+          s = RunManualCompaction(
-              (options.bottommost_level_compaction ==
+              cfd, level, output_level, options, begin, end, exclusive,
-                   BottommostLevelCompaction::kForce ||
+              !trim_ts.empty() /* disallow_trivial_move */,
-               options.bottommost_level_compaction ==
+              std::numeric_limits<uint64_t>::max() /* max_file_num_to_ignore */,
-                   BottommostLevelCompaction::kForceOptimized)) {
+              trim_ts,
-            disallow_trivial_move = true;
+              output_level == ColumnFamilyData::kCompactToBaseLevel
+                  ? &base_level
+                  : nullptr);
+          if (!s.ok()) {
+            break;
           }
+          if (output_level == ColumnFamilyData::kCompactToBaseLevel) {
+            assert(base_level > 0);
+            level = base_level;
+          } else {
+            ++level;
+          }
+          final_output_level = level;
+          TEST_SYNC_POINT("DBImpl::RunManualCompaction()::1");
+          TEST_SYNC_POINT("DBImpl::RunManualCompaction()::2");
         }
-        // trim_ts need real compaction to remove latest record
+        if (s.ok()) {
-        if (!trim_ts.empty()) {
+          assert(final_output_level > 0);
-          disallow_trivial_move = true;
+          // bottommost level intra-level compaction
-        }
+          // TODO(cbi): this preserves earlier behavior where if
-        s = RunManualCompaction(cfd, level, output_level, options, begin, end,
+          //  max_overlapped_level = 0 and bottommost_level_compaction is
-                                exclusive, disallow_trivial_move,
+          //  kIfHaveCompactionFilter, we only do a L0 -> LBase compaction
-                                max_file_num_to_ignore, trim_ts);
+          //  and do not do intra-LBase compaction even when user configures
-        if (!s.ok()) {
+          //  compaction filter. We may want to still do a LBase -> LBase
-          break;
+          //  compaction in case there is some file in LBase that did not go
-        }
+          //  through L0 -> LBase compaction, and hence did not go through
-        if (output_level == ColumnFamilyData::kCompactToBaseLevel) {
+          //  compaction filter.
-          final_output_level = cfd->NumberLevels() - 1;
+          if ((options.bottommost_level_compaction ==
-        } else if (output_level > final_output_level) {
+                   BottommostLevelCompaction::kIfHaveCompactionFilter &&
-          final_output_level = output_level;
+               max_overlapped_level != 0 &&
+               (cfd->ioptions()->compaction_filter != nullptr ||
+                cfd->ioptions()->compaction_filter_factory != nullptr)) ||
+              options.bottommost_level_compaction ==
+                  BottommostLevelCompaction::kForceOptimized ||
+              options.bottommost_level_compaction ==
+                  BottommostLevelCompaction::kForce) {
+            // Use `next_file_number` as `max_file_num_to_ignore` to avoid
+            // rewriting newly compacted files when it is kForceOptimized.
+            s = RunManualCompaction(
+                cfd, final_output_level, final_output_level, options, begin,
+                end, exclusive, !trim_ts.empty() /* disallow_trivial_move */,
+                next_file_number /* max_file_num_to_ignore */, trim_ts);
+          }
         }
-        TEST_SYNC_POINT("DBImpl::RunManualCompaction()::1");
-        TEST_SYNC_POINT("DBImpl::RunManualCompaction()::2");
       }
     }
   }
@@ -1853,7 +1862,8 @@ Status DBImpl::RunManualCompaction(
     ColumnFamilyData* cfd, int input_level, int output_level,
     const CompactRangeOptions& compact_range_options, const Slice* begin,
     const Slice* end, bool exclusive, bool disallow_trivial_move,
-    uint64_t max_file_num_to_ignore, const std::string& trim_ts) {
+    uint64_t max_file_num_to_ignore, const std::string& trim_ts,
+    int* final_output_level) {
   assert(input_level == ColumnFamilyData::kCompactAllLevels ||
          input_level >= 0);
 
@@ -2004,6 +2014,15 @@ Status DBImpl::RunManualCompaction(
     } else if (!scheduled) {
       if (compaction == nullptr) {
         manual.done = true;
+        if (final_output_level) {
+          // No compaction needed or there is a conflicting compaction.
+          // Still set `final_output_level` to the level where we would
+          // have compacted to.
+          *final_output_level = output_level;
+          if (output_level == ColumnFamilyData::kCompactToBaseLevel) {
+            *final_output_level = cfd->current()->storage_info()->base_level();
+          }
+        }
         bg_cv_.SignalAll();
         continue;
       }
@@ -2037,6 +2056,9 @@ Status DBImpl::RunManualCompaction(
       }
       scheduled = true;
       TEST_SYNC_POINT("DBImpl::RunManualCompaction:Scheduled");
+      if (final_output_level) {
+        *final_output_level = compaction->output_level();
+      }
     }
   }
 

db/db_sst_test.cc

@@ -810,9 +810,10 @@ TEST_F(DBSSTTest, RateLimitedWALDelete) {
   // We created 4 sst files in L0
   ASSERT_EQ("4", FilesPerLevel(0));
 
-  // Compaction will move the 4 files in L0 to trash and create 1 L1 file
+  // Compaction will move the 4 files in L0 to trash and create 1 L1 file.
+  // Use kForceOptimized to not rewrite the new L1 file.
   CompactRangeOptions cro;
-  cro.bottommost_level_compaction = BottommostLevelCompaction::kForce;
+  cro.bottommost_level_compaction = BottommostLevelCompaction::kForceOptimized;
   ASSERT_OK(db_->CompactRange(cro, nullptr, nullptr));
   ASSERT_OK(dbfull()->TEST_WaitForCompact(true));
   ASSERT_EQ("0,1", FilesPerLevel(0));