Log blob file space amp and expose it via the rocksdb.blob-stats DB property (#9538)

Summary:
Extend the periodic statistics in the info log with the total amount of garbage
in blob files and the space amplification pertaining to blob files, where the
latter is defined as `total_blob_file_size / (total_blob_file_size - total_blob_garbage_size)`.
Also expose the space amp via the `rocksdb.blob-stats` DB property.

Pull Request resolved: https://github.com/facebook/rocksdb/pull/9538

Test Plan: `make check`

Reviewed By: riversand963

Differential Revision: D34126855

Pulled By: ltamasi

fbshipit-source-id: 3153e7a0fe0eca440322db273f4deaabaccc51b2

HISTORY.md

@@ -68,6 +68,7 @@
 ## New Features
 * Introduced an option `BlockBasedTableBuilder::detect_filter_construct_corruption` for detecting corruption during Bloom Filter (format_version >= 5) and Ribbon Filter construction.
 * Improved the SstDumpTool to read the comparator from table properties and use it to read the SST File.
+* Extended the column family statistics in the info log so the total amount of garbage in the blob files and the blob file space amplification factor are also logged. Also exposed the blob file space amp via the `rocksdb.blob-stats` DB property.
 
 ## 6.29.0 (01/21/2022)
 Note: The next release will be major release 7.0. See https://github.com/facebook/rocksdb/issues/9390 for more info.

db/blob/db_blob_basic_test.cc

@@ -694,18 +694,21 @@ TEST_F(DBBlobBasicTest, Properties) {
   constexpr char key3[] = "key3";
   constexpr size_t key3_size = sizeof(key3) - 1;
 
-  constexpr char blob[] = "0000000000";
+  constexpr char blob[] = "00000000000000";
   constexpr size_t blob_size = sizeof(blob) - 1;
 
+  constexpr char longer_blob[] = "00000000000000000000";
+  constexpr size_t longer_blob_size = sizeof(longer_blob) - 1;
+
   ASSERT_OK(Put(key1, blob));
-  ASSERT_OK(Put(key2, blob));
+  ASSERT_OK(Put(key2, longer_blob));
   ASSERT_OK(Flush());
 
   constexpr size_t first_blob_file_expected_size =
       BlobLogHeader::kSize +
       BlobLogRecord::CalculateAdjustmentForRecordHeader(key1_size) + blob_size +
-      BlobLogRecord::CalculateAdjustmentForRecordHeader(key2_size) + blob_size +
+      BlobLogRecord::CalculateAdjustmentForRecordHeader(key2_size) +
-      BlobLogFooter::kSize;
+      longer_blob_size + BlobLogFooter::kSize;
 
   ASSERT_OK(Put(key3, blob));
   ASSERT_OK(Flush());
@@ -747,7 +750,12 @@ TEST_F(DBBlobBasicTest, Properties) {
   ASSERT_OK(db_->CompactRange(CompactRangeOptions(), begin, end));
 
   constexpr size_t expected_garbage_size =
-      BlobLogRecord::CalculateAdjustmentForRecordHeader(key2_size) + blob_size;
+      BlobLogRecord::CalculateAdjustmentForRecordHeader(key2_size) +
+      longer_blob_size;
+
+  constexpr double expected_space_amp =
+      static_cast<double>(total_expected_size) /
+      (total_expected_size - expected_garbage_size);
 
   // Blob file stats
   std::string blob_stats;
@@ -757,7 +765,7 @@ TEST_F(DBBlobBasicTest, Properties) {
   oss << "Number of blob files: 2\nTotal size of blob files: "
       << total_expected_size
       << "\nTotal size of garbage in blob files: " << expected_garbage_size
-      << '\n';
+      << "\nBlob file space amplification: " << expected_space_amp << '\n';
 
   ASSERT_EQ(blob_stats, oss.str());
 }

db/internal_stats.cc

@@ -781,24 +781,14 @@ bool InternalStats::HandleBlobStats(std::string* value, Slice /*suffix*/) {
   const auto* vstorage = current->storage_info();
   assert(vstorage);
 
-  const auto& blob_files = vstorage->GetBlobFiles();
+  const auto blob_st = vstorage->GetBlobStats();
-
-  uint64_t total_file_size = 0;
-  uint64_t total_garbage_size = 0;
-
-  for (const auto& meta : blob_files) {
-    assert(meta);
-
-    total_file_size += meta->GetBlobFileSize();
-    total_garbage_size += meta->GetGarbageBlobBytes();
-  }
 
   std::ostringstream oss;
 
-  oss << "Number of blob files: " << blob_files.size()
+  oss << "Number of blob files: " << vstorage->GetBlobFiles().size()
-      << "\nTotal size of blob files: " << total_file_size
+      << "\nTotal size of blob files: " << blob_st.total_file_size
-      << "\nTotal size of garbage in blob files: " << total_garbage_size
+      << "\nTotal size of garbage in blob files: " << blob_st.total_garbage_size
-      << '\n';
+      << "\nBlob file space amplification: " << blob_st.space_amp << '\n';
 
   value->append(oss.str());
 
@@ -824,7 +814,7 @@ bool InternalStats::HandleLiveBlobFileSize(uint64_t* value, DBImpl* /*db*/,
   const auto* vstorage = current->storage_info();
   assert(vstorage);
 
-  *value = vstorage->GetTotalBlobFileSize();
+  *value = vstorage->GetBlobStats().total_file_size;
 
   return true;
 }
@@ -1681,10 +1671,13 @@ void InternalStats::DumpCFStatsNoFileHistogram(std::string* value) {
     }
   }
 
+  const auto blob_st = vstorage->GetBlobStats();
+
   snprintf(buf, sizeof(buf),
-           "\nBlob file count: %" ROCKSDB_PRIszt ", total size: %.1f GB\n\n",
+           "\nBlob file count: %" ROCKSDB_PRIszt
-           vstorage->GetBlobFiles().size(),
+           ", total size: %.1f GB, garbage size: %.1f GB, space amp: %.1f\n\n",
-           vstorage->GetTotalBlobFileSize() / kGB);
+           vstorage->GetBlobFiles().size(), blob_st.total_file_size / kGB,
+           blob_st.total_garbage_size / kGB, blob_st.space_amp);
   value->append(buf);
 
   uint64_t now_micros = clock_->NowMicros();

db/version_set.h

@@ -356,16 +356,30 @@ class VersionStorageInfo {
   }
 
   // REQUIRES: This version has been saved (see VersionBuilder::SaveTo)
-  uint64_t GetTotalBlobFileSize() const {
+  struct BlobStats {
-    uint64_t total_blob_bytes = 0;
+    uint64_t total_file_size = 0;
+    uint64_t total_garbage_size = 0;
+    double space_amp = 0.0;
+  };
+
+  BlobStats GetBlobStats() const {
+    uint64_t total_file_size = 0;
+    uint64_t total_garbage_size = 0;
 
     for (const auto& meta : blob_files_) {
       assert(meta);
 
-      total_blob_bytes += meta->GetBlobFileSize();
+      total_file_size += meta->GetBlobFileSize();
+      total_garbage_size += meta->GetGarbageBlobBytes();
+    }
+
+    double space_amp = 0.0;
+    if (total_file_size > total_garbage_size) {
+      space_amp = static_cast<double>(total_file_size) /
+                  (total_file_size - total_garbage_size);
     }
 
-    return total_blob_bytes;
+    return BlobStats{total_file_size, total_garbage_size, space_amp};
   }
 
   const ROCKSDB_NAMESPACE::LevelFilesBrief& LevelFilesBrief(int level) const {