@nxtedition/rocksdb 7.0.11 → 7.0.14

package/binding.cc

@@ -1395,7 +1395,6 @@ NAPI_METHOD(db_get_property) {
   return result;
 }
 
-
 NAPI_METHOD(db_get_latest_sequence) {
   NAPI_ARGV(1);
 

package/deps/rocksdb/rocksdb/CMakeLists.txt

@@ -1301,6 +1301,7 @@ if(WITH_TESTS)
         db/version_set_test.cc
         db/wal_manager_test.cc
         db/wal_edit_test.cc
+        db/wide/db_wide_basic_test.cc
         db/wide/wide_column_serialization_test.cc
         db/write_batch_test.cc
         db/write_callback_test.cc

package/deps/rocksdb/rocksdb/Makefile

@@ -1383,6 +1383,9 @@ db_blob_compaction_test: $(OBJ_DIR)/db/blob/db_blob_compaction_test.o $(TEST_LIB
 db_readonly_with_timestamp_test: $(OBJ_DIR)/db/db_readonly_with_timestamp_test.o $(TEST_LIBRARY) $(LIBRARY)
 	$(AM_LINK)
 
+db_wide_basic_test: $(OBJ_DIR)/db/wide/db_wide_basic_test.o $(TEST_LIBRARY) $(LIBRARY)
+	$(AM_LINK)
+
 db_with_timestamp_basic_test: $(OBJ_DIR)/db/db_with_timestamp_basic_test.o $(TEST_LIBRARY) $(LIBRARY)
 	$(AM_LINK)
 

package/deps/rocksdb/rocksdb/TARGETS

@@ -5240,6 +5240,12 @@ cpp_unittest_wrapper(name="db_wal_test",
             extra_compiler_flags=[])
 
 
+cpp_unittest_wrapper(name="db_wide_basic_test",
+            srcs=["db/wide/db_wide_basic_test.cc"],
+            deps=[":rocksdb_test_lib"],
+            extra_compiler_flags=[])
+
+
 cpp_unittest_wrapper(name="db_with_timestamp_basic_test",
             srcs=["db/db_with_timestamp_basic_test.cc"],
             deps=[":rocksdb_test_lib"],

package/deps/rocksdb/rocksdb/cache/cache_bench_tool.cc

@@ -3,6 +3,7 @@
 //  COPYING file in the root directory) and Apache 2.0 License
 //  (found in the LICENSE.Apache file in the root directory).
 
+#include "cache_key.h"
 #ifdef GFLAGS
 #include <cinttypes>
 #include <cstddef>
@@ -214,7 +215,8 @@ struct KeyGen {
     EncodeFixed64(key_data + 10, key);
     key_data[18] = char{4};
     EncodeFixed64(key_data + 19, key);
-    return Slice(&key_data[off], sizeof(key_data) - off);
+    assert(27 >= kCacheKeySize);
+    return Slice(&key_data[off], kCacheKeySize);
   }
 };
 
@@ -321,8 +323,9 @@ class CacheBench {
     Random64 rnd(1);
     KeyGen keygen;
     for (uint64_t i = 0; i < 2 * FLAGS_cache_size; i += FLAGS_value_bytes) {
-      cache_->Insert(keygen.GetRand(rnd, max_key_, max_log_), createValue(rnd),
-                     &helper1, FLAGS_value_bytes);
+      Status s = cache_->Insert(keygen.GetRand(rnd, max_key_, max_log_),
+                                createValue(rnd), &helper1, FLAGS_value_bytes);
+      assert(s.ok());
     }
   }
 
@@ -542,8 +545,9 @@ class CacheBench {
                              FLAGS_value_bytes);
         } else {
           // do insert
-          cache_->Insert(key, createValue(thread->rnd), &helper2,
-                         FLAGS_value_bytes, &handle);
+          Status s = cache_->Insert(key, createValue(thread->rnd), &helper2,
+                                    FLAGS_value_bytes, &handle);
+          assert(s.ok());
         }
       } else if (random_op < insert_threshold_) {
         if (handle) {
@@ -551,8 +555,9 @@ class CacheBench {
           handle = nullptr;
         }
         // do insert
-        cache_->Insert(key, createValue(thread->rnd), &helper3,
-                       FLAGS_value_bytes, &handle);
+        Status s = cache_->Insert(key, createValue(thread->rnd), &helper3,
+                                  FLAGS_value_bytes, &handle);
+        assert(s.ok());
       } else if (random_op < lookup_threshold_) {
         if (handle) {
           cache_->Release(handle);

package/deps/rocksdb/rocksdb/cache/cache_key.h

@@ -65,6 +65,8 @@ class CacheKey {
   uint64_t offset_etc64_;
 };
 
+constexpr uint8_t kCacheKeySize = static_cast<uint8_t>(sizeof(CacheKey));
+
 // A file-specific generator of cache keys, sometimes referred to as the
 // "base" cache key for a file because all the cache keys for various offsets
 // within the file are computed using simple arithmetic. The basis for the

package/deps/rocksdb/rocksdb/cache/fast_lru_cache.cc

@@ -192,8 +192,7 @@ LRUCacheShard::LRUCacheShard(size_t capacity, size_t estimated_value_size,
     : capacity_(capacity),
       strict_capacity_limit_(strict_capacity_limit),
       table_(
-          CalcHashBits(capacity, estimated_value_size, metadata_charge_policy) +
-          static_cast<uint8_t>(ceil(log2(1.0 / kLoadFactor)))),
+          CalcHashBits(capacity, estimated_value_size, metadata_charge_policy)),
       usage_(0),
       lru_usage_(0) {
   set_metadata_charge_policy(metadata_charge_policy);
@@ -295,16 +294,29 @@ void LRUCacheShard::EvictFromLRU(size_t charge,
   }
 }
 
-uint8_t LRUCacheShard::CalcHashBits(
-    size_t capacity, size_t estimated_value_size,
+size_t LRUCacheShard::CalcEstimatedHandleCharge(
+    size_t estimated_value_size,
     CacheMetadataChargePolicy metadata_charge_policy) {
   LRUHandle h;
   h.CalcTotalCharge(estimated_value_size, metadata_charge_policy);
-  size_t num_entries = capacity / h.total_charge;
+  return h.total_charge;
+}
+
+uint8_t LRUCacheShard::CalcHashBits(
+    size_t capacity, size_t estimated_value_size,
+    CacheMetadataChargePolicy metadata_charge_policy) {
+  size_t handle_charge =
+      CalcEstimatedHandleCharge(estimated_value_size, metadata_charge_policy);
+  size_t num_entries =
+      static_cast<size_t>(capacity / (kLoadFactor * handle_charge));
+
+  // Compute the ceiling of log2(num_entries). If num_entries == 0, return 0.
   uint8_t num_hash_bits = 0;
-  while (num_entries >>= 1) {
+  size_t num_entries_copy = num_entries;
+  while (num_entries_copy >>= 1) {
     ++num_hash_bits;
   }
+  num_hash_bits += size_t{1} << num_hash_bits < num_entries ? 1 : 0;
   return num_hash_bits;
 }
 

package/deps/rocksdb/rocksdb/cache/fast_lru_cache.h

@@ -22,10 +22,14 @@
 #include "util/distributed_mutex.h"
 
 namespace ROCKSDB_NAMESPACE {
+
 namespace fast_lru_cache {
 
-// LRU cache implementation using an open-address hash table.
+// Forward declaration of friend class.
+class FastLRUCacheTest;
 
+// LRU cache implementation using an open-address hash table.
+//
 // Every slot in the hash table is an LRUHandle. Because handles can be
 // referenced externally, we can't discard them immediately once they are
 // deleted (via a delete or an LRU eviction) or replaced by a new version
@@ -51,7 +55,7 @@ namespace fast_lru_cache {
 // - Not R --> R: When an unreferenced element becomes referenced. This can only
 //    happen if the element is V, since references to an element can only be
 //    created when it's visible.
-
+//
 // Internally, the cache uses an open-addressed hash table to index the handles.
 // We use tombstone counters to keep track of displacements.
 // Because of the tombstones and the two possible visibility states of an
@@ -70,9 +74,6 @@ namespace fast_lru_cache {
 // slot. In any case, the slot becomes available. When a handle is inserted
 // into that slot, it becomes a visible element again.
 
-constexpr uint8_t kCacheKeySize =
-    static_cast<uint8_t>(sizeof(ROCKSDB_NAMESPACE::CacheKey));
-
 // The load factor p is a real number in (0, 1) such that at all
 // times at most a fraction p of all slots, without counting tombstones,
 // are occupied by elements. This means that the probability that a
@@ -367,6 +368,8 @@ class ALIGN_AS(CACHE_LINE_SIZE) LRUCacheShard final : public CacheShard {
 
  private:
   friend class LRUCache;
+  friend class FastLRUCacheTest;
+
   void LRU_Remove(LRUHandle* e);
   void LRU_Insert(LRUHandle* e);
 
@@ -376,6 +379,11 @@ class ALIGN_AS(CACHE_LINE_SIZE) LRUCacheShard final : public CacheShard {
   // holding the mutex_.
   void EvictFromLRU(size_t charge, autovector<LRUHandle>* deleted);
 
+  // Returns the charge of a single handle.
+  static size_t CalcEstimatedHandleCharge(
+      size_t estimated_value_size,
+      CacheMetadataChargePolicy metadata_charge_policy);
+
   // Returns the number of bits used to hash an element in the hash
   // table.
   static uint8_t CalcHashBits(size_t capacity, size_t estimated_value_size,

package/deps/rocksdb/rocksdb/cache/lru_cache_test.cc

@@ -206,6 +206,7 @@ TEST_F(LRUCacheTest, EntriesWithPriority) {
   ValidateLRUList({"e", "f", "g", "Z", "d"}, 2);
 }
 
+namespace fast_lru_cache {
 // TODO(guido) Consolidate the following FastLRUCache tests with
 // that of LRUCache.
 class FastLRUCacheTest : public testing::Test {
@@ -238,6 +239,38 @@ class FastLRUCacheTest : public testing::Test {
 
   Status Insert(char key, size_t len) { return Insert(std::string(len, key)); }
 
+  size_t CalcEstimatedHandleChargeWrapper(
+      size_t estimated_value_size,
+      CacheMetadataChargePolicy metadata_charge_policy) {
+    return fast_lru_cache::LRUCacheShard::CalcEstimatedHandleCharge(
+        estimated_value_size, metadata_charge_policy);
+  }
+
+  uint8_t CalcHashBitsWrapper(
+      size_t capacity, size_t estimated_value_size,
+      CacheMetadataChargePolicy metadata_charge_policy) {
+    return fast_lru_cache::LRUCacheShard::CalcHashBits(
+        capacity, estimated_value_size, metadata_charge_policy);
+  }
+
+  // Maximum number of items that a shard can hold.
+  double CalcMaxOccupancy(size_t capacity, size_t estimated_value_size,
+                          CacheMetadataChargePolicy metadata_charge_policy) {
+    size_t handle_charge =
+        fast_lru_cache::LRUCacheShard::CalcEstimatedHandleCharge(
+            estimated_value_size, metadata_charge_policy);
+    return capacity / (fast_lru_cache::kLoadFactor * handle_charge);
+  }
+
+  bool TableSizeIsAppropriate(uint8_t hash_bits, double max_occupancy) {
+    if (hash_bits == 0) {
+      return max_occupancy <= 1;
+    } else {
+      return (1 << hash_bits >= max_occupancy) &&
+             (1 << (hash_bits - 1) <= max_occupancy);
+    }
+  }
+
  private:
   fast_lru_cache::LRUCacheShard* cache_ = nullptr;
 };
@@ -253,6 +286,62 @@ TEST_F(FastLRUCacheTest, ValidateKeySize) {
   EXPECT_NOK(Insert('f', 0));
 }
 
+TEST_F(FastLRUCacheTest, CalcHashBitsTest) {
+  size_t capacity = 1024;
+  size_t estimated_value_size = 1;
+  CacheMetadataChargePolicy metadata_charge_policy = kDontChargeCacheMetadata;
+  double max_occupancy =
+      CalcMaxOccupancy(capacity, estimated_value_size, metadata_charge_policy);
+  uint8_t hash_bits = CalcHashBitsWrapper(capacity, estimated_value_size,
+                                          metadata_charge_policy);
+  EXPECT_TRUE(TableSizeIsAppropriate(hash_bits, max_occupancy));
+
+  capacity = 1024;
+  estimated_value_size = 1;
+  metadata_charge_policy = kFullChargeCacheMetadata;
+  max_occupancy =
+      CalcMaxOccupancy(capacity, estimated_value_size, metadata_charge_policy);
+  hash_bits = CalcHashBitsWrapper(capacity, estimated_value_size,
+                                  metadata_charge_policy);
+  EXPECT_TRUE(TableSizeIsAppropriate(hash_bits, max_occupancy));
+
+  // No elements fit in cache.
+  capacity = 0;
+  estimated_value_size = 1;
+  metadata_charge_policy = kDontChargeCacheMetadata;
+  hash_bits = CalcHashBitsWrapper(capacity, estimated_value_size,
+                                  metadata_charge_policy);
+  EXPECT_TRUE(TableSizeIsAppropriate(hash_bits, 0 /* max_occupancy */));
+
+  // Set the capacity just below a single handle. Because the load factor is <
+  // 100% at least one handle will fit in the table.
+  estimated_value_size = 1;
+  size_t handle_charge = CalcEstimatedHandleChargeWrapper(
+      8192 /* estimated_value_size */, kDontChargeCacheMetadata);
+  capacity = handle_charge - 1;
+  // The load factor should be bounded away from 100%.
+  assert(static_cast<size_t>(capacity / fast_lru_cache::kLoadFactor) >
+         handle_charge);
+  metadata_charge_policy = kDontChargeCacheMetadata;
+  max_occupancy =
+      CalcMaxOccupancy(capacity, estimated_value_size, metadata_charge_policy);
+  hash_bits = CalcHashBitsWrapper(capacity, estimated_value_size,
+                                  metadata_charge_policy);
+  EXPECT_TRUE(TableSizeIsAppropriate(hash_bits, max_occupancy));
+
+  // Large capacity.
+  capacity = 31924172;
+  estimated_value_size = 321;
+  metadata_charge_policy = kFullChargeCacheMetadata;
+  max_occupancy =
+      CalcMaxOccupancy(capacity, estimated_value_size, metadata_charge_policy);
+  hash_bits = CalcHashBitsWrapper(capacity, estimated_value_size,
+                                  metadata_charge_policy);
+  EXPECT_TRUE(TableSizeIsAppropriate(hash_bits, max_occupancy));
+}
+
+}  // namespace fast_lru_cache
+
 class TestSecondaryCache : public SecondaryCache {
  public:
   // Specifies what action to take on a lookup for a particular key

package/deps/rocksdb/rocksdb/db/blob/blob_file_reader.cc

@@ -443,50 +443,40 @@ void BlobFileReader::MultiGetBlob(
   }
 
   assert(s.ok());
+
+  uint64_t total_bytes = 0;
   for (size_t i = 0; i < num_blobs; ++i) {
     auto& req = read_reqs[i];
+    const auto& record_slice = req.result;
+
     assert(statuses[i]);
-    if (req.status.ok() && req.result.size() != req.len) {
+    if (req.status.ok() && record_slice.size() != req.len) {
       req.status = IOStatus::Corruption("Failed to read data from blob file");
     }
+
     *statuses[i] = req.status;
-  }
+    if (!statuses[i]->ok()) {
+      continue;
+    }
 
-  if (read_options.verify_checksums) {
-    for (size_t i = 0; i < num_blobs; ++i) {
-      assert(statuses[i]);
+    // Verify checksums if enabled
+    if (read_options.verify_checksums) {
+      *statuses[i] = VerifyBlob(record_slice, user_keys[i], value_sizes[i]);
       if (!statuses[i]->ok()) {
         continue;
       }
-      const Slice& record_slice = read_reqs[i].result;
-      s = VerifyBlob(record_slice, user_keys[i], value_sizes[i]);
-      if (!s.ok()) {
-        assert(statuses[i]);
-        *statuses[i] = s;
-      }
     }
-  }
 
-  for (size_t i = 0; i < num_blobs; ++i) {
-    assert(statuses[i]);
-    if (!statuses[i]->ok()) {
-      continue;
-    }
-    const Slice& record_slice = read_reqs[i].result;
-    const Slice value_slice(record_slice.data() + adjustments[i],
-                            value_sizes[i]);
-    s = UncompressBlobIfNeeded(value_slice, compression_type_, clock_,
-                               statistics_, values[i]);
-    if (!s.ok()) {
-      *statuses[i] = s;
+    // Uncompress blob if needed
+    Slice value_slice(record_slice.data() + adjustments[i], value_sizes[i]);
+    *statuses[i] = UncompressBlobIfNeeded(value_slice, compression_type_,
+                                          clock_, statistics_, values[i]);
+    if (statuses[i]->ok()) {
+      total_bytes += record_slice.size();
     }
   }
 
   if (bytes_read) {
-    uint64_t total_bytes = 0;
-    for (const auto& req : read_reqs) {
-      total_bytes += req.result.size();
-    }
     *bytes_read = total_bytes;
   }
 }

package/deps/rocksdb/rocksdb/db/blob/blob_source.cc

@@ -9,7 +9,9 @@
 #include <string>
 
 #include "db/blob/blob_file_reader.h"
+#include "db/blob/blob_log_format.h"
 #include "options/cf_options.h"
+#include "table/multiget_context.h"
 
 namespace ROCKSDB_NAMESPACE {
 
@@ -98,9 +100,16 @@ Status BlobSource::GetBlob(const ReadOptions& read_options,
     Slice key = cache_key.AsSlice();
     s = GetBlobFromCache(key, &blob_entry);
     if (s.ok() && blob_entry.GetValue()) {
-      assert(blob_entry.GetValue()->size() == value_size);
+      // For consistency, the size of on-disk (possibly compressed) blob record
+      // is assigned to bytes_read.
       if (bytes_read) {
-        *bytes_read = value_size;
+        uint64_t adjustment =
+            read_options.verify_checksums
+                ? BlobLogRecord::CalculateAdjustmentForRecordHeader(
+                      user_key.size())
+                : 0;
+        assert(offset >= adjustment);
+        *bytes_read = value_size + adjustment;
       }
       value->PinSelf(*blob_entry.GetValue());
       return s;
@@ -152,6 +161,142 @@ Status BlobSource::GetBlob(const ReadOptions& read_options,
   return s;
 }
 
+void BlobSource::MultiGetBlob(
+    const ReadOptions& read_options,
+    const autovector<std::reference_wrapper<const Slice>>& user_keys,
+    uint64_t file_number, uint64_t file_size,
+    const autovector<uint64_t>& offsets,
+    const autovector<uint64_t>& value_sizes, autovector<Status*>& statuses,
+    autovector<PinnableSlice*>& blobs, uint64_t* bytes_read) {
+  size_t num_blobs = user_keys.size();
+  assert(num_blobs > 0);
+  assert(num_blobs <= MultiGetContext::MAX_BATCH_SIZE);
+  assert(num_blobs == offsets.size());
+  assert(num_blobs == value_sizes.size());
+  assert(num_blobs == statuses.size());
+  assert(num_blobs == blobs.size());
+
+#ifndef NDEBUG
+  for (size_t i = 0; i < offsets.size() - 1; ++i) {
+    assert(offsets[i] <= offsets[i + 1]);
+  }
+#endif  // !NDEBUG
+
+  using Mask = uint64_t;
+  Mask cache_hit_mask = 0;
+
+  Status s;
+  uint64_t total_bytes = 0;
+  const OffsetableCacheKey base_cache_key(db_id_, db_session_id_, file_number,
+                                          file_size);
+
+  if (blob_cache_) {
+    size_t cached_blob_count = 0;
+    for (size_t i = 0; i < num_blobs; ++i) {
+      CachableEntry<std::string> blob_entry;
+      const CacheKey cache_key = base_cache_key.WithOffset(offsets[i]);
+      const Slice key = cache_key.AsSlice();
+
+      s = GetBlobFromCache(key, &blob_entry);
+      if (s.ok() && blob_entry.GetValue()) {
+        assert(statuses[i]);
+        *statuses[i] = s;
+        blobs[i]->PinSelf(*blob_entry.GetValue());
+
+        // Update the counter for the number of valid blobs read from the cache.
+        ++cached_blob_count;
+        // For consistency, the size of each on-disk (possibly compressed) blob
+        // record is accumulated to total_bytes.
+        uint64_t adjustment =
+            read_options.verify_checksums
+                ? BlobLogRecord::CalculateAdjustmentForRecordHeader(
+                      user_keys[i].get().size())
+                : 0;
+        assert(offsets[i] >= adjustment);
+        total_bytes += value_sizes[i] + adjustment;
+        cache_hit_mask |= (Mask{1} << i);  // cache hit
+      }
+    }
+
+    // All blobs were read from the cache.
+    if (cached_blob_count == num_blobs) {
+      if (bytes_read) {
+        *bytes_read = total_bytes;
+      }
+      return;
+    }
+  }
+
+  const bool no_io = read_options.read_tier == kBlockCacheTier;
+  if (no_io) {
+    for (size_t i = 0; i < num_blobs; ++i) {
+      if (!(cache_hit_mask & (Mask{1} << i))) {
+        assert(statuses[i]);
+        *statuses[i] =
+            Status::Incomplete("Cannot read blob(s): no disk I/O allowed");
+      }
+    }
+    return;
+  }
+
+  {
+    // Find the rest of blobs from the file since I/O is allowed.
+    autovector<std::reference_wrapper<const Slice>> _user_keys;
+    autovector<uint64_t> _offsets;
+    autovector<uint64_t> _value_sizes;
+    autovector<Status*> _statuses;
+    autovector<PinnableSlice*> _blobs;
+    uint64_t _bytes_read = 0;
+
+    for (size_t i = 0; i < num_blobs; ++i) {
+      if (!(cache_hit_mask & (Mask{1} << i))) {
+        _user_keys.emplace_back(user_keys[i]);
+        _offsets.push_back(offsets[i]);
+        _value_sizes.push_back(value_sizes[i]);
+        _statuses.push_back(statuses[i]);
+        _blobs.push_back(blobs[i]);
+      }
+    }
+
+    CacheHandleGuard<BlobFileReader> blob_file_reader;
+    s = blob_file_cache_->GetBlobFileReader(file_number, &blob_file_reader);
+    if (!s.ok()) {
+      for (size_t i = 0; i < _blobs.size(); ++i) {
+        assert(_statuses[i]);
+        *_statuses[i] = s;
+      }
+      return;
+    }
+
+    assert(blob_file_reader.GetValue());
+
+    blob_file_reader.GetValue()->MultiGetBlob(read_options, _user_keys,
+                                              _offsets, _value_sizes, _statuses,
+                                              _blobs, &_bytes_read);
+
+    if (read_options.fill_cache) {
+      // If filling cache is allowed and a cache is configured, try to put
+      // the blob(s) to the cache.
+      for (size_t i = 0; i < _blobs.size(); ++i) {
+        if (_statuses[i]->ok()) {
+          CachableEntry<std::string> blob_entry;
+          const CacheKey cache_key = base_cache_key.WithOffset(_offsets[i]);
+          const Slice key = cache_key.AsSlice();
+          s = PutBlobIntoCache(key, &blob_entry, _blobs[i]);
+          if (!s.ok()) {
+            *_statuses[i] = s;
+          }
+        }
+      }
+    }
+
+    total_bytes += _bytes_read;
+    if (bytes_read) {
+      *bytes_read = total_bytes;
+    }
+  }
+}
+
 bool BlobSource::TEST_BlobInCache(uint64_t file_number, uint64_t file_size,
                                   uint64_t offset) const {
   const CacheKey cache_key = GetCacheKey(file_number, file_size, offset);

package/deps/rocksdb/rocksdb/db/blob/blob_source.h

@@ -13,6 +13,7 @@
 #include "rocksdb/cache.h"
 #include "rocksdb/rocksdb_namespace.h"
 #include "table/block_based/cachable_entry.h"
+#include "util/autovector.h"
 
 namespace ROCKSDB_NAMESPACE {
 
@@ -36,12 +37,41 @@ class BlobSource {
 
   ~BlobSource();
 
+  // Read a blob from the underlying cache or storage.
+  //
+  // If successful, returns ok and sets "*value" to the newly retrieved
+  // uncompressed blob. If there was an error while fetching the blob, sets
+  // "*value" to empty and returns a non-ok status.
+  //
+  // Note: For consistency, whether the blob is found in the cache or on disk,
+  // sets "*bytes_read" to the size of on-disk (possibly compressed) blob
+  // record.
   Status GetBlob(const ReadOptions& read_options, const Slice& user_key,
                  uint64_t file_number, uint64_t offset, uint64_t file_size,
                  uint64_t value_size, CompressionType compression_type,
                  FilePrefetchBuffer* prefetch_buffer, PinnableSlice* value,
                  uint64_t* bytes_read);
 
+  // Read multiple blobs from the underlying cache or storage.
+  //
+  // If successful, returns ok and sets the elements of blobs to the newly
+  // retrieved uncompressed blobs. If there was an error while fetching one of
+  // blobs, sets its corresponding "blobs[i]" to empty and sets "statuses[i]" to
+  // a non-ok status.
+  //
+  // Note:
+  //  - Offsets must be sorted in ascending order by caller.
+  //  - For consistency, whether the blob is found in the cache or on disk, sets
+  //  "*bytes_read" to the total size of on-disk (possibly compressed) blob
+  //  records.
+  void MultiGetBlob(
+      const ReadOptions& read_options,
+      const autovector<std::reference_wrapper<const Slice>>& user_keys,
+      uint64_t file_number, uint64_t file_size,
+      const autovector<uint64_t>& offsets,
+      const autovector<uint64_t>& value_sizes, autovector<Status*>& statuses,
+      autovector<PinnableSlice*>& blobs, uint64_t* bytes_read);
+
   inline Status GetBlobFileReader(
       uint64_t blob_file_number,
       CacheHandleGuard<BlobFileReader>* blob_file_reader) {