A new compaction picking priority that optimizes for write amplification for random updates.

Summary: Introduce a compaction picking priority that picks files who contains the oldest rows to compact. This is a mode that slightly improves write amplification for random update cases.

Test Plan: Add a unit test and run it in valgrind too.

Reviewers: yhchiang, anthony, IslamAbdelRahman, rven, kradhakrishnan, MarkCallaghan, igor

Reviewed By: igor

Subscribers: leveldb, dhruba

Differential Revision: https://reviews.facebook.net/D51459

HISTORY.md

@@ -9,6 +9,7 @@
 
 ### Public API Changes
 * When options.db_write_buffer_size triggers, only the column family with the largest column family size will be flushed, not all the column families.
+* Change names in CompactionPri and add a new one.
 
 ## 4.2.0 (11/9/2015)
 ### New Features

db/db_compaction_test.cc

@@ -1868,6 +1868,53 @@ TEST_P(DBCompactionTestWithParam, ForceBottommostLevelCompaction) {
 
 INSTANTIATE_TEST_CASE_P(DBCompactionTestWithParam, DBCompactionTestWithParam,
                         ::testing::Values(1, 4));
+
+class CompactionPriTest : public DBTestBase,
+                          public testing::WithParamInterface<uint32_t> {
+ public:
+  CompactionPriTest() : DBTestBase("/compaction_pri_test") {
+    compaction_pri_ = GetParam();
+  }
+
+  // Required if inheriting from testing::WithParamInterface<>
+  static void SetUpTestCase() {}
+  static void TearDownTestCase() {}
+
+  uint32_t compaction_pri_;
+};
+
+TEST_P(CompactionPriTest, Test) {
+  Options options;
+  options.write_buffer_size = 16 * 1024;
+  options.compaction_pri = static_cast<CompactionPri>(compaction_pri_);
+  options.hard_pending_compaction_bytes_limit = 256 * 1024;
+  options.max_bytes_for_level_base = 64 * 1024;
+  options.max_bytes_for_level_multiplier = 4;
+  options.compression = kNoCompression;
+  options = CurrentOptions(options);
+  DestroyAndReopen(options);
+
+  Random rnd(301);
+  const int kNKeys = 5000;
+  int keys[kNKeys];
+  for (int i = 0; i < kNKeys; i++) {
+    keys[i] = i;
+  }
+  std::random_shuffle(std::begin(keys), std::end(keys));
+
+  for (int i = 0; i < kNKeys; i++) {
+    ASSERT_OK(Put(Key(keys[i]), RandomString(&rnd, 102)));
+  }
+
+  dbfull()->TEST_WaitForCompact();
+  for (int i = 0; i < kNKeys; i++) {
+    ASSERT_NE("NOT_FOUND", Get(Key(i)));
+  }
+}
+
+INSTANTIATE_TEST_CASE_P(CompactionPriTest, CompactionPriTest,
+                        ::testing::Values(0, 1, 2));
+
 #endif // !defined(ROCKSDB_LITE)
 }  // namespace rocksdb
 

db/version_set.cc

@@ -1386,16 +1386,22 @@ void VersionStorageInfo::UpdateFilesByCompactionPri(
       num = temp.size();
     }
     switch (mutable_cf_options.compaction_pri) {
-      case kCompactionPriByCompensatedSize:
+      case kByCompensatedSize:
         std::partial_sort(temp.begin(), temp.begin() + num, temp.end(),
                           CompareCompensatedSizeDescending);
         break;
-      case kCompactionPriByLargestSeq:
+      case kOldestLargestSeqFirst:
         std::sort(temp.begin(), temp.end(),
                   [this](const Fsize& f1, const Fsize& f2) -> bool {
                     return f1.file->largest_seqno < f2.file->largest_seqno;
                   });
         break;
+      case kOldestSmallestSeqFirst:
+        std::sort(temp.begin(), temp.end(),
+                  [this](const Fsize& f1, const Fsize& f2) -> bool {
+                    return f1.file->smallest_seqno < f2.file->smallest_seqno;
+                  });
+        break;
       default:
         assert(false);
     }

include/rocksdb/options.h

@@ -82,9 +82,14 @@ enum CompactionStyle : char {
 
 enum CompactionPri : char {
   // Slightly Priotize larger files by size compensated by #deletes
-  kCompactionPriByCompensatedSize = 0x0,
-  // First compact files whose data is oldest.
-  kCompactionPriByLargestSeq = 0x1,
+  kByCompensatedSize = 0x0,
+  // First compact files whose data's latest update time is oldest.
+  // Try this if you only update some hot keys in small ranges.
+  kOldestLargestSeqFirst = 0x1,
+  // First compact files whose range hasn't been compacted to the next level
+  // for the longest. If your updates are random across the key space,
+  // write amplification is slightly better with this option.
+  kOldestSmallestSeqFirst = 0x2,
 };
 
 enum class WALRecoveryMode : char {

util/mutable_cf_options.h

@@ -67,7 +67,7 @@ struct MutableCFOptions {
         level0_file_num_compaction_trigger(0),
         level0_slowdown_writes_trigger(0),
         level0_stop_writes_trigger(0),
-        compaction_pri(kCompactionPriByCompensatedSize),
+        compaction_pri(kByCompensatedSize),
         max_grandparent_overlap_factor(0),
         expanded_compaction_factor(0),
         source_compaction_factor(0),

util/options.cc

@@ -110,7 +110,7 @@ ColumnFamilyOptions::ColumnFamilyOptions()
       disable_auto_compactions(false),
       purge_redundant_kvs_while_flush(true),
       compaction_style(kCompactionStyleLevel),
-      compaction_pri(kCompactionPriByCompensatedSize),
+      compaction_pri(kByCompensatedSize),
       verify_checksums_in_compaction(true),
       filter_deletes(false),
       max_sequential_skip_in_iterations(8),