make hash_link_list Node's key space consecutively followed at the end

Summary: per sdong's request, this will help processor prefetch on n->key case.

Test Plan: make all check

Reviewers: sdong, haobo, igor

Reviewed By: sdong

CC: leveldb

Differential Revision: https://reviews.facebook.net/D17415

db/db_bench.cc

@@ -1162,8 +1162,8 @@ class Benchmark {
         fresh_db = true;
         if (num_threads > 1) {
           fprintf(stderr, "filluniquerandom multithreaded not supported"
-                           " set --threads=1");
+                           ", use 1 thread");
-          exit(1);
+          num_threads = 1;
         }
         method = &Benchmark::WriteUniqueRandom;
       } else if (name == Slice("overwrite")) {

db/memtable.cc

@@ -142,6 +142,11 @@ Slice MemTableRep::UserKey(const char* key) const {
   return Slice(slice.data(), slice.size() - 8);
 }
 
+KeyHandle MemTableRep::Allocate(const size_t len, char** buf) {
+  *buf = arena_->Allocate(len);
+  return static_cast<KeyHandle>(*buf);
+}
+
 // Encode a suitable internal key target for "target" and return it.
 // Uses *scratch as scratch space, and the returned pointer will point
 // into this scratch space.
@@ -243,7 +248,9 @@ void MemTable::Add(SequenceNumber s, ValueType type,
   const size_t encoded_len =
       VarintLength(internal_key_size) + internal_key_size +
       VarintLength(val_size) + val_size;
-  char* buf = arena_.Allocate(encoded_len);
+  char* buf = nullptr;
+  KeyHandle handle = table_->Allocate(encoded_len, &buf);
+  assert(buf != nullptr);
   char* p = EncodeVarint32(buf, internal_key_size);
   memcpy(p, key.data(), key_size);
   p += key_size;
@@ -252,7 +259,7 @@ void MemTable::Add(SequenceNumber s, ValueType type,
   p = EncodeVarint32(p, val_size);
   memcpy(p, value.data(), val_size);
   assert((unsigned)(p + val_size - buf) == (unsigned)encoded_len);
-  table_->Insert(buf);
+  table_->Insert(handle);
 
   if (prefix_bloom_) {
     assert(prefix_extractor_);

include/rocksdb/memtablerep.h

@@ -45,6 +45,8 @@ class LookupKey;
 class Slice;
 class SliceTransform;
 
+typedef void* KeyHandle;
+
 class MemTableRep {
  public:
   // KeyComparator provides a means to compare keys, which are internal keys
@@ -62,11 +64,19 @@ class MemTableRep {
     virtual ~KeyComparator() { }
   };
 
+  explicit MemTableRep(Arena* arena) : arena_(arena) {}
+
+  // Allocate a buf of len size for storing key. The idea is that a specific
+  // memtable representation knows its underlying data structure better. By
+  // allowing it to allocate memory, it can possibly put correlated stuff
+  // in consecutive memory area to make processor prefetching more efficient.
+  virtual KeyHandle Allocate(const size_t len, char** buf);
+
   // Insert key into the collection. (The caller will pack key and value into a
-  // single buffer and pass that in as the parameter to Insert)
+  // single buffer and pass that in as the parameter to Insert).
   // REQUIRES: nothing that compares equal to key is currently in the
   // collection.
-  virtual void Insert(const char* key) = 0;
+  virtual void Insert(KeyHandle handle) = 0;
 
   // Returns true iff an entry that compares equal to key is in the collection.
   virtual bool Contains(const char* key) const = 0;
@@ -153,6 +163,8 @@ class MemTableRep {
   // When *key is an internal key concatenated with the value, returns the
   // user key.
   virtual Slice UserKey(const char* key) const;
+
+  Arena* arena_;
 };
 
 // This is the base class for all factories that are used by RocksDB to create

util/hash_linklist_rep.cc

@@ -22,12 +22,6 @@ namespace {
 typedef const char* Key;
 
 struct Node {
-  explicit Node(const Key& k) :
-      key(k) {
-  }
-
-  Key const key;
-
   // Accessors/mutators for links.  Wrapped in methods so we can
   // add the appropriate barriers as necessary.
   Node* Next() {
@@ -40,17 +34,19 @@ struct Node {
     // pointer observes a fully initialized version of the inserted node.
     next_.Release_Store(x);
   }
-
   // No-barrier variants that can be safely used in a few locations.
   Node* NoBarrier_Next() {
     return reinterpret_cast<Node*>(next_.NoBarrier_Load());
   }
+
   void NoBarrier_SetNext(Node* x) {
     next_.NoBarrier_Store(x);
   }
 
-private:
+ private:
   port::AtomicPointer next_;
+ public:
+  char key[0];
 };
 
 class HashLinkListRep : public MemTableRep {
@@ -58,7 +54,9 @@ class HashLinkListRep : public MemTableRep {
   HashLinkListRep(const MemTableRep::KeyComparator& compare, Arena* arena,
                   const SliceTransform* transform, size_t bucket_size);
 
-  virtual void Insert(const char* key) override;
+  virtual KeyHandle Allocate(const size_t len, char** buf) override;
+
+  virtual void Insert(KeyHandle handle) override;
 
   virtual bool Contains(const char* key) const override;
 
@@ -93,8 +91,6 @@ class HashLinkListRep : public MemTableRep {
   const SliceTransform* transform_;
 
   const MemTableRep::KeyComparator& compare_;
-  // immutable after construction
-  Arena* const arena_;
 
   bool BucketContains(Node* head, const Slice& key) const;
 
@@ -114,11 +110,6 @@ class HashLinkListRep : public MemTableRep {
     return GetBucket(GetHash(slice));
   }
 
-  Node* NewNode(const Key& key) {
-    char* mem = arena_->AllocateAligned(sizeof(Node));
-    return new (mem) Node(key);
-  }
-
   bool Equal(const Slice& a, const Key& b) const {
     return (compare_(b, a) == 0);
   }
@@ -318,10 +309,10 @@ class HashLinkListRep : public MemTableRep {
 HashLinkListRep::HashLinkListRep(const MemTableRep::KeyComparator& compare,
                                  Arena* arena, const SliceTransform* transform,
                                  size_t bucket_size)
-  : bucket_size_(bucket_size),
+  : MemTableRep(arena),
+    bucket_size_(bucket_size),
     transform_(transform),
-    compare_(compare),
+    compare_(compare) {
-    arena_(arena) {
   char* mem = arena_->AllocateAligned(
       sizeof(port::AtomicPointer) * bucket_size);
 
@@ -335,15 +326,22 @@ HashLinkListRep::HashLinkListRep(const MemTableRep::KeyComparator& compare,
 HashLinkListRep::~HashLinkListRep() {
 }
 
-void HashLinkListRep::Insert(const char* key) {
+KeyHandle HashLinkListRep::Allocate(const size_t len, char** buf) {
-  assert(!Contains(key));
+  char* mem = arena_->AllocateAligned(sizeof(Node) + len);
-  Slice internal_key = GetLengthPrefixedSlice(key);
+  Node* x = new (mem) Node();
+  *buf = x->key;
+  return static_cast<void*>(x);
+}
+
+void HashLinkListRep::Insert(KeyHandle handle) {
+  Node* x = static_cast<Node*>(handle);
+  assert(!Contains(x->key));
+  Slice internal_key = GetLengthPrefixedSlice(x->key);
   auto transformed = GetPrefix(internal_key);
   auto& bucket = buckets_[GetHash(transformed)];
   Node* head = static_cast<Node*>(bucket.Acquire_Load());
 
   if (!head) {
-    Node* x = NewNode(key);
     // NoBarrier_SetNext() suffices since we will add a barrier when
     // we publish a pointer to "x" in prev[i].
     x->NoBarrier_SetNext(nullptr);
@@ -372,9 +370,7 @@ void HashLinkListRep::Insert(const char* key) {
   }
 
   // Our data structure does not allow duplicate insertion
-  assert(cur == nullptr || !Equal(key, cur->key));
+  assert(cur == nullptr || !Equal(x->key, cur->key));
-
-  Node* x = NewNode(key);
 
   // NoBarrier_SetNext() suffices since we will add a barrier when
   // we publish a pointer to "x" in prev[i].

util/hash_skiplist_rep.cc

@@ -25,7 +25,7 @@ class HashSkipListRep : public MemTableRep {
                   const SliceTransform* transform, size_t bucket_size,
                   int32_t skiplist_height, int32_t skiplist_branching_factor);
 
-  virtual void Insert(const char* key) override;
+  virtual void Insert(KeyHandle handle) override;
 
   virtual bool Contains(const char* key) const override;
 
@@ -225,7 +225,8 @@ HashSkipListRep::HashSkipListRep(const MemTableRep::KeyComparator& compare,
                                  Arena* arena, const SliceTransform* transform,
                                  size_t bucket_size, int32_t skiplist_height,
                                  int32_t skiplist_branching_factor)
-    : bucket_size_(bucket_size),
+    : MemTableRep(arena),
+      bucket_size_(bucket_size),
       skiplist_height_(skiplist_height),
       skiplist_branching_factor_(skiplist_branching_factor),
       transform_(transform),
@@ -255,7 +256,8 @@ HashSkipListRep::Bucket* HashSkipListRep::GetInitializedBucket(
   return bucket;
 }
 
-void HashSkipListRep::Insert(const char* key) {
+void HashSkipListRep::Insert(KeyHandle handle) {
+  auto* key = static_cast<char*>(handle);
   assert(!Contains(key));
   auto transformed = transform_->Transform(UserKey(key));
   auto bucket = GetInitializedBucket(transformed);

util/skiplistrep.cc

@@ -13,13 +13,13 @@ class SkipListRep : public MemTableRep {
   SkipList<const char*, const MemTableRep::KeyComparator&> skip_list_;
 public:
   explicit SkipListRep(const MemTableRep::KeyComparator& compare, Arena* arena)
-    : skip_list_(compare, arena) {
+    : MemTableRep(arena), skip_list_(compare, arena) {
   }
 
   // Insert key into the list.
   // REQUIRES: nothing that compares equal to key is currently in the list.
-  virtual void Insert(const char* key) override {
+  virtual void Insert(KeyHandle handle) override {
-    skip_list_.Insert(key);
+    skip_list_.Insert(static_cast<char*>(handle));
   }
 
   // Returns true iff an entry that compares equal to key is in the list.

util/vectorrep.cc

@@ -30,7 +30,7 @@ class VectorRep : public MemTableRep {
   // single buffer and pass that in as the parameter to Insert)
   // REQUIRES: nothing that compares equal to key is currently in the
   // collection.
-  virtual void Insert(const char* key) override;
+  virtual void Insert(KeyHandle handle) override;
 
   // Returns true iff an entry that compares equal to key is in the collection.
   virtual bool Contains(const char* key) const override;
@@ -106,7 +106,8 @@ class VectorRep : public MemTableRep {
   const KeyComparator& compare_;
 };
 
-void VectorRep::Insert(const char* key) {
+void VectorRep::Insert(KeyHandle handle) {
+  auto* key = static_cast<char*>(handle);
   assert(!Contains(key));
   WriteLock l(&rwlock_);
   assert(!immutable_);
@@ -134,7 +135,8 @@ size_t VectorRep::ApproximateMemoryUsage() {
 }
 
 VectorRep::VectorRep(const KeyComparator& compare, Arena* arena, size_t count)
-  : bucket_(new Bucket()),
+  : MemTableRep(arena),
+    bucket_(new Bucket()),
     immutable_(false),
     sorted_(false),
     compare_(compare) { bucket_.get()->reserve(count); }