Reduce overhead of SortFileByOverlappingRatio() (#10161)

Summary:
Currently SortFileByOverlappingRatio() is O(nlogn). It is usually OK but When there are a lot of files in an LSM-tree, SortFileByOverlappingRatio() can take non-trivial amount of time. The problem is severe when the user is loading keys in sorted order, where compaction is only trivial move and this operation becomes the bottleneck and limit the total throughput. This commit makes SortFileByOverlappingRatio() only find the top 50 files based on score. 50 files are usually enough for the parallel compactions needed for the level, and in case it is not enough, we would fall back to random, which should be acceptable.

Pull Request resolved: https://github.com/facebook/rocksdb/pull/10161

Test Plan:
Run a fillseq that generates a lot of files, and observe throughput improved (although stall is not yet eliminated). The command ran:

TEST_TMPDIR=/dev/shm/ ./db_bench_sort --benchmarks=fillseq --compression_type=lz4 --write_buffer_size=5000000 --num=100000000 --value_size=1000

The throughput improved by 11%.

Reviewed By: jay-zhuang

Differential Revision: D37129469

fbshipit-source-id: 492da2ef5bfc7cdd6daa3986b50d2ff91f88542d

HISTORY.md

@@ -64,6 +64,9 @@
 ### Performance Improvements
 * When compiled with folly (Meta-internal integration; experimental in open source build), improve the locking performance (CPU efficiency) of LRUCache by using folly DistributedMutex in place of standard mutex.
 
+### Performance Improvements
+* Rather than doing total sort against all files in a level, SortFileByOverlappingRatio() to only find the top 50 files based on score. This can improve write throughput for the use cases where data is loaded in increasing key order and there are a lot of files in one LSM-tree, where applying compaction results is the bottleneck.
+
 ## 7.3.0 (05/20/2022)
 ### Bug Fixes
 * Fixed a bug where manual flush would block forever even though flush options had wait=false.

db/compaction/compaction_picker_test.cc

@@ -1349,6 +1349,61 @@ TEST_F(CompactionPickerTest, CompactionPriRoundRobin) {
   }
 }
 
+TEST_F(CompactionPickerTest, CompactionPriMinOverlappingManyFiles) {
+  NewVersionStorage(6, kCompactionStyleLevel);
+  ioptions_.compaction_pri = kMinOverlappingRatio;
+  mutable_cf_options_.max_bytes_for_level_base = 15000000;
+  mutable_cf_options_.max_bytes_for_level_multiplier = 10;
+
+  // file 7 and 8 over lap with the same file, but file 8 is smaller so
+  // it will be picked.
+  Add(2, 13U, "010", "011",
+      6100U);  // Overlaps with a large file. Not picked
+  Add(2, 14U, "020", "021",
+      6100U);  // Overlaps with a large file. Not picked
+  Add(2, 15U, "030", "031",
+      6100U);  // Overlaps with a large file. Not picked
+  Add(2, 16U, "040", "041",
+      6100U);  // Overlaps with a large file. Not picked
+  Add(2, 17U, "050", "051",
+      6100U);  // Overlaps with a large file. Not picked
+  Add(2, 18U, "060", "061",
+      6100U);  // Overlaps with a large file. Not picked
+  Add(2, 19U, "070", "071",
+      6100U);  // Overlaps with a large file. Not picked
+  Add(2, 20U, "080", "081",
+      6100U);  // Overlaps with a large file. Not picked
+
+  Add(2, 6U, "150", "167", 60000000U);  // Overlaps with file 26, 27
+  Add(2, 7U, "168", "169", 60000000U);  // Overlaps with file 27
+  Add(2, 8U, "201", "300", 61000000U);  // Overlaps with file 28, but the file
+                                        // itself is larger. Should be picked.
+  Add(2, 9U, "610", "611",
+      6100U);  // Overlaps with a large file. Not picked
+  Add(2, 10U, "620", "621",
+      6100U);  // Overlaps with a large file. Not picked
+  Add(2, 11U, "630", "631",
+      6100U);  // Overlaps with a large file. Not picked
+  Add(2, 12U, "640", "641",
+      6100U);  // Overlaps with a large file. Not picked
+
+  Add(3, 31U, "001", "100", 260000000U);
+  Add(3, 26U, "160", "165", 260000000U);
+  Add(3, 27U, "166", "170", 260000000U);
+  Add(3, 28U, "180", "400", 260000000U);
+  Add(3, 29U, "401", "500", 260000000U);
+  Add(3, 30U, "601", "700", 260000000U);
+  UpdateVersionStorageInfo();
+
+  std::unique_ptr<Compaction> compaction(level_compaction_picker.PickCompaction(
+      cf_name_, mutable_cf_options_, mutable_db_options_, vstorage_.get(),
+      &log_buffer_));
+  ASSERT_TRUE(compaction.get() != nullptr);
+  ASSERT_EQ(1U, compaction->num_input_files(0));
+  // Picking file 8 because overlapping ratio is the biggest.
+  ASSERT_EQ(8U, compaction->input(0, 0)->fd.GetNumber());
+}
+
 // This test exhibits the bug where we don't properly reset parent_index in
 // PickCompaction()
 TEST_F(CompactionPickerTest, ParentIndexResetBug) {

db/version_set.cc

@@ -3189,11 +3189,15 @@ void SortFileByOverlappingRatio(
                                           ttl_boost_score;
   }
 
-  std::sort(temp->begin(), temp->end(),
-            [&](const Fsize& f1, const Fsize& f2) -> bool {
-              return file_to_order[f1.file->fd.GetNumber()] <
-                     file_to_order[f2.file->fd.GetNumber()];
-            });
+  size_t num_to_sort = temp->size() > VersionStorageInfo::kNumberFilesToSort
+                           ? VersionStorageInfo::kNumberFilesToSort
+                           : temp->size();
+
+  std::partial_sort(temp->begin(), temp->begin() + num_to_sort, temp->end(),
+                    [&](const Fsize& f1, const Fsize& f2) -> bool {
+                      return file_to_order[f1.file->fd.GetNumber()] <
+                             file_to_order[f2.file->fd.GetNumber()];
+                    });
 }
 
 void SortFileByRoundRobin(const InternalKeyComparator& icmp,