Update version to 6.28.2

Summary: Update version to 6.28.2 for bug fix

.circleci/config.yml

@@ -24,6 +24,13 @@ commands:
           command: |
             HOMEBREW_NO_AUTO_UPDATE=1 brew install cmake
 
+  install-jdk8-on-macos:
+    steps:
+      - run:
+          name: Install JDK 8 on macos
+          command: |
+            brew install --cask adoptopenjdk/openjdk/adoptopenjdk8
+
   increase-max-open-files-on-macos:
     steps:
       - run:
@@ -55,7 +62,7 @@ commands:
   pre-steps-macos:
       steps:
         - pre-steps:
-            python-version: "3.6.0"
+            python-version: "3.7.8"
 
   post-steps:
     steps:
@@ -144,8 +151,10 @@ executors:
 jobs:
   build-macos:
     macos:
-      xcode: 11.3.0
+      xcode: 12.5.1
     resource_class: large
+    environment:
+      DISABLE_JEMALLOC: 1 # jemalloc cause env_test hang, disable it for now
     steps:
       - increase-max-open-files-on-macos
       - install-pyenv-on-macos
@@ -156,7 +165,7 @@ jobs:
 
   build-macos-cmake:
     macos:
-      xcode: 11.3.0
+      xcode: 12.5.1
     resource_class: large
     steps:
       - increase-max-open-files-on-macos
@@ -522,14 +531,16 @@ jobs:
 
   build-macos-java:
     macos:
-      xcode: 11.3.0
+      xcode: 12.5.1
     resource_class: medium
     environment:
       JAVA_HOME: /Library/Java/JavaVirtualMachines/adoptopenjdk-8.jdk/Contents/Home
+      DISABLE_JEMALLOC: 1 # jemalloc causes java 8 crash, maybe a known issue
     steps:
       - increase-max-open-files-on-macos
       - install-pyenv-on-macos
       - install-gflags-on-macos
+      - install-jdk8-on-macos
       - pre-steps-macos
       - run:
           name: "Set Java Environment"
@@ -548,7 +559,7 @@ jobs:
 
   build-macos-java-static:
     macos:
-      xcode: 11.3.0
+      xcode: 12.5.1
     resource_class: medium
     environment:
       JAVA_HOME: /Library/Java/JavaVirtualMachines/adoptopenjdk-8.jdk/Contents/Home
@@ -556,6 +567,7 @@ jobs:
       - increase-max-open-files-on-macos
       - install-pyenv-on-macos
       - install-gflags-on-macos
+      - install-jdk8-on-macos
       - install-cmake-on-macos
       - pre-steps-macos
       - run:

HISTORY.md

@@ -1,4 +1,12 @@
 # Rocksdb Change Log
+## 6.28.2 (2022-01-31)
+### Bug Fixes
+* Fixed a major bug in which batched MultiGet could return old values for keys deleted by DeleteRange when memtable Bloom filter is enabled (memtable_prefix_bloom_size_ratio > 0). (The fix includes a substantial MultiGet performance improvement in the unusual case of both memtable_whole_key_filtering and prefix_extractor.)
+
+## 6.28.1 (2022-01-10)
+### Bug Fixes
+* Fixed compilation errors on newer compiler, e.g. clang-12
+
 ## 6.28.0 (2021-12-17)
 ### New Features
 * Introduced 'CommitWithTimestamp' as a new tag. Currently, there is no API for user to trigger a write with this tag to the WAL. This is part of the efforts to support write-commited transactions with user-defined timestamps.

db/db_bloom_filter_test.cc

@@ -1502,6 +1502,63 @@ TEST_F(DBBloomFilterTest, MemtableWholeKeyBloomFilter) {
   ASSERT_EQ(1, get_perf_context()->bloom_memtable_hit_count);
 }
 
+TEST_F(DBBloomFilterTest, MemtableWholeKeyBloomFilterMultiGet) {
+  Options options = CurrentOptions();
+  options.memtable_prefix_bloom_size_ratio = 0.015;
+  options.memtable_whole_key_filtering = true;
+  Reopen(options);
+  std::string key1("AA");
+  std::string key2("BB");
+  std::string key3("CC");
+  std::string key4("DD");
+  std::string key_not("EE");
+  std::string value1("Value1");
+  std::string value2("Value2");
+  std::string value3("Value3");
+  std::string value4("Value4");
+
+  ASSERT_OK(Put(key1, value1, WriteOptions()));
+  ASSERT_OK(Put(key2, value2, WriteOptions()));
+  ASSERT_OK(Flush());
+  ASSERT_OK(Put(key3, value3, WriteOptions()));
+  const Snapshot* snapshot = db_->GetSnapshot();
+  ASSERT_OK(Put(key4, value4, WriteOptions()));
+
+  // Delete key2 and key3
+  ASSERT_OK(
+      db_->DeleteRange(WriteOptions(), db_->DefaultColumnFamily(), "BA", "CZ"));
+
+  // Read without snapshot
+  auto results = MultiGet({key_not, key1, key2, key3, key4});
+  ASSERT_EQ(results[0], "NOT_FOUND");
+  ASSERT_EQ(results[1], value1);
+  ASSERT_EQ(results[2], "NOT_FOUND");
+  ASSERT_EQ(results[3], "NOT_FOUND");
+  ASSERT_EQ(results[4], value4);
+
+  // Also check Get
+  ASSERT_EQ(Get(key1), value1);
+  ASSERT_EQ(Get(key2), "NOT_FOUND");
+  ASSERT_EQ(Get(key3), "NOT_FOUND");
+  ASSERT_EQ(Get(key4), value4);
+
+  // Read with snapshot
+  results = MultiGet({key_not, key1, key2, key3, key4}, snapshot);
+  ASSERT_EQ(results[0], "NOT_FOUND");
+  ASSERT_EQ(results[1], value1);
+  ASSERT_EQ(results[2], value2);
+  ASSERT_EQ(results[3], value3);
+  ASSERT_EQ(results[4], "NOT_FOUND");
+
+  // Also check Get
+  ASSERT_EQ(Get(key1, snapshot), value1);
+  ASSERT_EQ(Get(key2, snapshot), value2);
+  ASSERT_EQ(Get(key3, snapshot), value3);
+  ASSERT_EQ(Get(key4, snapshot), "NOT_FOUND");
+
+  db_->ReleaseSnapshot(snapshot);
+}
+
 TEST_F(DBBloomFilterTest, MemtablePrefixBloomOutOfDomain) {
   constexpr size_t kPrefixSize = 8;
   const std::string kKey = "key";

db/memtable.cc

@@ -33,6 +33,7 @@
 #include "rocksdb/iterator.h"
 #include "rocksdb/merge_operator.h"
 #include "rocksdb/slice_transform.h"
+#include "rocksdb/types.h"
 #include "rocksdb/write_buffer_manager.h"
 #include "table/internal_iterator.h"
 #include "table/iterator_wrapper.h"
@@ -447,11 +448,13 @@ FragmentedRangeTombstoneIterator* MemTable::NewRangeTombstoneIterator(
       is_range_del_table_empty_.load(std::memory_order_relaxed)) {
     return nullptr;
   }
+  return NewRangeTombstoneIteratorInternal(read_options, read_seq);
+}
+
+FragmentedRangeTombstoneIterator* MemTable::NewRangeTombstoneIteratorInternal(
+    const ReadOptions& read_options, SequenceNumber read_seq) {
   auto* unfragmented_iter = new MemTableIterator(
       *this, read_options, nullptr /* arena */, true /* use_range_del_table */);
-  if (unfragmented_iter == nullptr) {
-    return nullptr;
-  }
   auto fragmented_tombstone_list =
       std::make_shared<FragmentedRangeTombstoneList>(
           std::unique_ptr<InternalIterator>(unfragmented_iter),
@@ -960,53 +963,58 @@ void MemTable::MultiGet(const ReadOptions& read_options, MultiGetRange* range,
   }
   PERF_TIMER_GUARD(get_from_memtable_time);
 
+  // For now, memtable Bloom filter is effectively disabled if there are any
+  // range tombstones. This is the simplest way to ensure range tombstones are
+  // handled. TODO: allow Bloom checks where max_covering_tombstone_seq==0
+  bool no_range_del = read_options.ignore_range_deletions ||
+                      is_range_del_table_empty_.load(std::memory_order_relaxed);
   MultiGetRange temp_range(*range, range->begin(), range->end());
-  if (bloom_filter_) {
-    std::array<Slice*, MultiGetContext::MAX_BATCH_SIZE> keys;
-    std::array<bool, MultiGetContext::MAX_BATCH_SIZE> may_match = {{true}};
-    autovector<Slice, MultiGetContext::MAX_BATCH_SIZE> prefixes;
+  if (bloom_filter_ && no_range_del) {
+    bool whole_key =
+        !prefix_extractor_ || moptions_.memtable_whole_key_filtering;
+    std::array<Slice, MultiGetContext::MAX_BATCH_SIZE> bloom_keys;
+    std::array<bool, MultiGetContext::MAX_BATCH_SIZE> may_match;
+    std::array<size_t, MultiGetContext::MAX_BATCH_SIZE> range_indexes;
     int num_keys = 0;
     for (auto iter = temp_range.begin(); iter != temp_range.end(); ++iter) {
-      if (!prefix_extractor_) {
-        keys[num_keys++] = &iter->ukey_without_ts;
+      if (whole_key) {
+        bloom_keys[num_keys] = iter->ukey_without_ts;
+        range_indexes[num_keys++] = iter.index();
       } else if (prefix_extractor_->InDomain(iter->ukey_without_ts)) {
-        prefixes.emplace_back(
-            prefix_extractor_->Transform(iter->ukey_without_ts));
-        keys[num_keys++] = &prefixes.back();
-      }
-    }
-    bloom_filter_->MayContain(num_keys, &keys[0], &may_match[0]);
-    int idx = 0;
-    for (auto iter = temp_range.begin(); iter != temp_range.end(); ++iter) {
-      if (prefix_extractor_ &&
-          !prefix_extractor_->InDomain(iter->ukey_without_ts)) {
+        bloom_keys[num_keys] =
+            prefix_extractor_->Transform(iter->ukey_without_ts);
+        range_indexes[num_keys++] = iter.index();
+      } else {
+        // TODO: consider not counting these as Bloom hits to more closely
+        // match bloom_sst_hit_count
         PERF_COUNTER_ADD(bloom_memtable_hit_count, 1);
-        continue;
       }
-      if (!may_match[idx]) {
-        temp_range.SkipKey(iter);
+    }
+    bloom_filter_->MayContain(num_keys, &bloom_keys[0], &may_match[0]);
+    for (int i = 0; i < num_keys; ++i) {
+      if (!may_match[i]) {
+        temp_range.SkipIndex(range_indexes[i]);
         PERF_COUNTER_ADD(bloom_memtable_miss_count, 1);
       } else {
         PERF_COUNTER_ADD(bloom_memtable_hit_count, 1);
       }
-      idx++;
     }
   }
   for (auto iter = temp_range.begin(); iter != temp_range.end(); ++iter) {
-    SequenceNumber seq = kMaxSequenceNumber;
     bool found_final_value{false};
     bool merge_in_progress = iter->s->IsMergeInProgress();
+    if (!no_range_del) {
       std::unique_ptr<FragmentedRangeTombstoneIterator> range_del_iter(
-        NewRangeTombstoneIterator(
+          NewRangeTombstoneIteratorInternal(
               read_options, GetInternalKeySeqno(iter->lkey->internal_key())));
-    if (range_del_iter != nullptr) {
       iter->max_covering_tombstone_seq = std::max(
           iter->max_covering_tombstone_seq,
           range_del_iter->MaxCoveringTombstoneSeqnum(iter->lkey->user_key()));
     }
+    SequenceNumber dummy_seq;
     GetFromTable(*(iter->lkey), iter->max_covering_tombstone_seq, true,
                  callback, &iter->is_blob_index, iter->value->GetSelf(),
-                 iter->timestamp, iter->s, &(iter->merge_context), &seq,
+                 iter->timestamp, iter->s, &(iter->merge_context), &dummy_seq,
                  &found_final_value, &merge_in_progress);
 
     if (!found_final_value && merge_in_progress) {

db/memtable.h

@@ -600,6 +600,10 @@ class MemTable {
                     std::string* value, std::string* timestamp, Status* s,
                     MergeContext* merge_context, SequenceNumber* seq,
                     bool* found_final_value, bool* merge_in_progress);
+
+  // Always returns non-null and assumes certain pre-checks are done
+  FragmentedRangeTombstoneIterator* NewRangeTombstoneIteratorInternal(
+      const ReadOptions& read_options, SequenceNumber read_seq);
 };
 
 extern const char* EncodeKey(std::string* scratch, const Slice& target);

include/rocksdb/advanced_options.h

@@ -366,10 +366,16 @@ struct AdvancedColumnFamilyOptions {
                                    Slice delta_value,
                                    std::string* merged_value) = nullptr;
 
-  // if prefix_extractor is set and memtable_prefix_bloom_size_ratio is not 0,
-  // create prefix bloom for memtable with the size of
+  // Should really be called `memtable_bloom_size_ratio`. Enables a dynamic
+  // Bloom filter in memtable to optimize many queries that must go beyond
+  // the memtable. The size in bytes of the filter is
   // write_buffer_size * memtable_prefix_bloom_size_ratio.
-  // If it is larger than 0.25, it is sanitized to 0.25.
+  // * If prefix_extractor is set, the filter includes prefixes.
+  // * If memtable_whole_key_filtering, the filter includes whole keys.
+  // * If both, the filter includes both.
+  // * If neither, the feature is disabled.
+  //
+  // If this value is larger than 0.25, it is sanitized to 0.25.
   //
   // Default: 0 (disable)
   //

include/rocksdb/comparator.h

@@ -75,7 +75,7 @@ class Comparator : public Customizable {
   //
   // Names starting with "rocksdb." are reserved and should not be used
   // by any clients of this package.
-  virtual const char* Name() const override = 0;
+  const char* Name() const override = 0;
 
   // Advanced functions: these are used to reduce the space requirements
   // for internal data structures like index blocks.

include/rocksdb/file_checksum.h

@@ -85,7 +85,7 @@ class FileChecksumGenFactory : public Customizable {
       const FileChecksumGenContext& context) = 0;
 
   // Return the name of this FileChecksumGenFactory.
-  virtual const char* Name() const override = 0;
+  const char* Name() const override = 0;
 };
 
 // FileChecksumList stores the checksum information of a list of files (e.g.,

include/rocksdb/memtablerep.h

@@ -311,7 +311,7 @@ class MemTableRepFactory : public Customizable {
     return CreateMemTableRep(key_cmp, allocator, slice_transform, logger);
   }
 
-  virtual const char* Name() const override = 0;
+  const char* Name() const override = 0;
 
   // Return true if the current MemTableRep supports concurrent inserts
   // Default: false

include/rocksdb/sst_partitioner.h

@@ -93,7 +93,7 @@ class SstPartitionerFactory : public Customizable {
       const SstPartitioner::Context& context) const = 0;
 
   // Returns a name that identifies this partitioner factory.
-  virtual const char* Name() const override = 0;
+  const char* Name() const override = 0;
 };
 
 /*

include/rocksdb/table_properties.h

@@ -162,7 +162,7 @@ class TablePropertiesCollectorFactory : public Customizable {
       TablePropertiesCollectorFactory::Context context) = 0;
 
   // The name of the properties collector can be used for debugging purpose.
-  virtual const char* Name() const override = 0;
+  const char* Name() const override = 0;
 
   // Can be overridden by sub-classes to return the Name, followed by
   // configuration info that will // be logged to the info log when the

include/rocksdb/version.h

@@ -11,7 +11,7 @@
 
 #define ROCKSDB_MAJOR 6
 #define ROCKSDB_MINOR 28
-#define ROCKSDB_PATCH 0
+#define ROCKSDB_PATCH 2
 
 // Do not use these. We made the mistake of declaring macros starting with
 // double underscore. Now we have to live with our choice. We'll deprecate these

options/customizable_test.cc

@@ -202,22 +202,31 @@ struct SimpleOptions {
   TestCustomizable* cp = nullptr;
 };
 
+static SimpleOptions dummy_simple_options;
+template <typename T1>
+int offset_of(T1 SimpleOptions::*member) {
+  return static_cast<int>(
+      reinterpret_cast<uintptr_t>(
+          std::addressof(dummy_simple_options.*member)) -
+      reinterpret_cast<uintptr_t>(std::addressof(dummy_simple_options)));
+}
+
 static std::unordered_map<std::string, OptionTypeInfo> simple_option_info = {
 #ifndef ROCKSDB_LITE
     {"bool",
-     {offsetof(struct SimpleOptions, b), OptionType::kBoolean,
+     {offset_of(&SimpleOptions::b), OptionType::kBoolean,
       OptionVerificationType::kNormal, OptionTypeFlags::kNone}},
     {"unique",
      OptionTypeInfo::AsCustomUniquePtr<TestCustomizable>(
-         offsetof(struct SimpleOptions, cu), OptionVerificationType::kNormal,
+         offset_of(&SimpleOptions::cu), OptionVerificationType::kNormal,
          OptionTypeFlags::kAllowNull)},
     {"shared",
      OptionTypeInfo::AsCustomSharedPtr<TestCustomizable>(
-         offsetof(struct SimpleOptions, cs), OptionVerificationType::kNormal,
+         offset_of(&SimpleOptions::cs), OptionVerificationType::kNormal,
          OptionTypeFlags::kAllowNull)},
     {"pointer",
      OptionTypeInfo::AsCustomRawPtr<TestCustomizable>(
-         offsetof(struct SimpleOptions, cp), OptionVerificationType::kNormal,
+         offset_of(&SimpleOptions::cp), OptionVerificationType::kNormal,
          OptionTypeFlags::kAllowNull)},
 #endif  // ROCKSDB_LITE
 };

table/multiget_context.h

@@ -235,9 +235,9 @@ class MultiGetContext {
 
     bool empty() const { return RemainingMask() == 0; }
 
-    void SkipKey(const Iterator& iter) {
-      skip_mask_ |= uint64_t{1} << iter.index_;
-    }
+    void SkipIndex(size_t index) { skip_mask_ |= uint64_t{1} << index; }
+
+    void SkipKey(const Iterator& iter) { SkipIndex(iter.index_); }
 
     bool IsKeySkipped(const Iterator& iter) const {
       return skip_mask_ & (uint64_t{1} << iter.index_);

tools/db_crashtest.py

@@ -87,6 +87,7 @@ default_params = {
     "partition_filters": lambda: random.randint(0, 1),
     "partition_pinning": lambda: random.randint(0, 3),
     "pause_background_one_in": 1000000,
+    "prefix_size" : lambda: random.choice([-1, 1, 5, 7, 8]),
     "prefixpercent": 5,
     "progress_reports": 0,
     "readpercent": 45,
@@ -156,6 +157,8 @@ default_params = {
     "user_timestamp_size": 0,
     "secondary_cache_fault_one_in" : lambda: random.choice([0, 0, 32]),
     "prepopulate_block_cache" : lambda: random.choice([0, 1]),
+    "memtable_prefix_bloom_size_ratio": lambda: random.choice([0.001, 0.01, 0.1, 0.5]),
+    "memtable_whole_key_filtering": lambda: random.randint(0, 1),
 }
 
 _TEST_DIR_ENV_VAR = 'TEST_TMPDIR'
@@ -233,9 +236,6 @@ simple_default_params = {
     "max_background_compactions": 1,
     "max_bytes_for_level_base": 67108864,
     "memtablerep": "skip_list",
-    "prefixpercent": 0,
-    "readpercent": 50,
-    "prefix_size" : -1,
     "target_file_size_base": 16777216,
     "target_file_size_multiplier": 1,
     "test_batches_snapshots": 0,
@@ -386,8 +386,15 @@ def finalize_and_sanitize(src_params):
         dest_params["readpercent"] += dest_params.get("iterpercent", 10)
         dest_params["iterpercent"] = 0
         dest_params["test_batches_snapshots"] = 0
+    if dest_params.get("prefix_size") == -1:
+        dest_params["readpercent"] += dest_params.get("prefixpercent", 20)
+        dest_params["prefixpercent"] = 0
+        dest_params["test_batches_snapshots"] = 0
     if dest_params.get("test_batches_snapshots") == 0:
         dest_params["batch_protection_bytes_per_key"] = 0
+    if (dest_params.get("prefix_size") == -1 and
+        dest_params.get("memtable_whole_key_filtering") == 0):
+        dest_params["memtable_prefix_bloom_size_ratio"] = 0
     return dest_params
 
 def gen_cmd_params(args):

util/dynamic_bloom.h

@@ -65,7 +65,7 @@ class DynamicBloom {
   // Multithreaded access to this function is OK
   bool MayContain(const Slice& key) const;
 
-  void MayContain(int num_keys, Slice** keys, bool* may_match) const;
+  void MayContain(int num_keys, Slice* keys, bool* may_match) const;
 
   // Multithreaded access to this function is OK
   bool MayContainHash(uint32_t hash) const;
@@ -120,12 +120,12 @@ inline bool DynamicBloom::MayContain(const Slice& key) const {
   return (MayContainHash(BloomHash(key)));
 }
 
-inline void DynamicBloom::MayContain(int num_keys, Slice** keys,
+inline void DynamicBloom::MayContain(int num_keys, Slice* keys,
                                      bool* may_match) const {
   std::array<uint32_t, MultiGetContext::MAX_BATCH_SIZE> hashes;
   std::array<size_t, MultiGetContext::MAX_BATCH_SIZE> byte_offsets;
   for (int i = 0; i < num_keys; ++i) {
-    hashes[i] = BloomHash(*keys[i]);
+    hashes[i] = BloomHash(keys[i]);
     size_t a = FastRange32(kLen, hashes[i]);
     PREFETCH(data_ + a, 0, 3);
     byte_offsets[i] = a;