Introduce experimental set

Stop leaking dict interface into rdb_load.

src/core/CMakeLists.txt

@@ -1,6 +1,6 @@
 add_library(dfly_core compact_object.cc dragonfly_core.cc extent_tree.cc 
             external_alloc.cc interpreter.cc mi_memory_resource.cc
-            segment_allocator.cc small_string.cc tx_queue.cc)
+            segment_allocator.cc small_string.cc string_set.cc tx_queue.cc)
 cxx_link(dfly_core base absl::flat_hash_map absl::str_format redis_lib TRDP::lua 
          Boost::fiber crypto)
 
@@ -15,3 +15,4 @@ cxx_test(external_alloc_test dfly_core LABELS DFLY)
 cxx_test(dash_test dfly_core LABELS DFLY)
 cxx_test(interpreter_test dfly_core LABELS DFLY)
 cxx_test(json_test dfly_core TRDP::jsoncons LABELS DFLY)
+cxx_test(string_set_test dfly_core LABELS DFLY)

src/core/compact_object.cc

@@ -21,6 +21,7 @@ extern "C" {
 
 #include "base/logging.h"
 #include "base/pod_array.h"
+#include "core/string_set.h"
 
 #if defined(__aarch64__)
 #include "base/sse2neon.h"
@@ -55,6 +56,11 @@ inline void FreeObjSet(unsigned encoding, void* ptr, pmr::memory_resource* mr) {
       dictRelease((dict*)ptr);
       break;
     }
+    case kEncodingStrMap2: {
+      delete (StringSet*)ptr;
+      break;
+    }
+
     case kEncodingIntSet:
       zfree((void*)ptr);
       break;
@@ -67,6 +73,10 @@ size_t MallocUsedSet(unsigned encoding, void* ptr) {
   switch (encoding) {
     case kEncodingStrMap /*OBJ_ENCODING_HT*/:
       return 0;  // TODO
+    case kEncodingStrMap2: {
+      StringSet* ss = (StringSet*)ptr;
+      return ss->obj_malloc_used() + ss->set_malloc_used();
+    }
     case kEncodingIntSet:
       return intsetBlobLen((intset*)ptr);
   }
@@ -258,6 +268,10 @@ size_t RobjWrapper::Size() const {
           dict* d = (dict*)inner_obj_;
           return dictSize(d);
         }
+        case kEncodingStrMap2: {
+          StringSet* ss = (StringSet*)inner_obj_;
+          return ss->size();
+        }
         default:
           LOG(FATAL) << "Unexpected encoding " << encoding_;
       }
@@ -608,6 +622,7 @@ void CompactObj::ImportRObj(robj* o) {
       if (o->encoding == OBJ_ENCODING_INTSET) {
         enc = kEncodingIntSet;
       } else {
+        LOG(DFATAL) << "This can not be done via ImportRObj for sets";
         enc = kEncodingStrMap;
       }
     }

src/core/compact_object.h

@@ -17,6 +17,7 @@ namespace dfly {
 
 constexpr unsigned kEncodingIntSet = 0;
 constexpr unsigned kEncodingStrMap = 1;  // for set/map encodings of strings
+constexpr unsigned kEncodingStrMap2 = 2;
 
 namespace detail {
 

src/core/string_set.cc

@@ -0,0 +1,509 @@
+// Copyright 2022, Roman Gershman.  All rights reserved.
+// See LICENSE for licensing terms.
+//
+
+#include "core/string_set.h"
+
+#include <absl/numeric/bits.h>
+#include <absl/strings/escaping.h>
+
+#include "base/logging.h"
+#include "core/compact_object.h"  // for hashcode
+
+extern "C" {
+#include "redis/zmalloc.h"
+}
+
+namespace dfly {
+using namespace std;
+
+constexpr size_t kMinSizeShift = 2;
+constexpr size_t kMinSize = 1 << kMinSizeShift;
+constexpr size_t kAllowDisplacements = true;
+
+inline bool CanSetFlat(int offs) {
+  if (kAllowDisplacements)
+    return offs < 2;
+  return offs == 0;
+}
+
+StringSet::StringSet(pmr::memory_resource* mr) : entries_(mr) {
+}
+
+StringSet::~StringSet() {
+  for (auto& entry : entries_) {
+    if (entry.IsLink()) {
+      LinkKey* lk = (LinkKey*)entry.get();
+      while (lk) {
+        sdsfree((sds)lk->ptr);
+        SuperPtr next = lk->next;
+        Free(lk);
+        if (next.IsSds()) {
+          sdsfree((sds)next.get());
+          lk = nullptr;
+        } else {
+          DCHECK(next.IsLink());
+          lk = (LinkKey*)next.get();
+        }
+      }
+    } else if (!entry.IsEmpty()) {
+      sdsfree((sds)entry.get());
+    }
+  }
+  DCHECK_EQ(0u, num_chain_entries_);
+}
+
+void StringSet::Reserve(size_t sz) {
+  sz = std::min<size_t>(sz, kMinSize);
+
+  sz = absl::bit_ceil(sz);
+  capacity_log_ = absl::bit_width(sz);
+  entries_.reserve(sz);
+}
+
+size_t StringSet::SuperPtr::SetString(std::string_view str) {
+  sds sdsptr = sdsnewlen(str.data(), str.size());
+  ptr = sdsptr;
+  return zmalloc_usable_size(sdsAllocPtr(sdsptr));
+}
+
+bool StringSet::SuperPtr::Compare(std::string_view str) const {
+  if (IsEmpty())
+    return false;
+
+  sds sp = GetSds();
+  return str == string_view{sp, sdslen(sp)};
+}
+
+bool StringSet::Add(std::string_view str) {
+  DVLOG(1) << "Add " << absl::CHexEscape(str);
+
+  uint64_t hc = CompactObj::HashCode(str);
+
+  if (entries_.empty()) {
+    capacity_log_ = kMinSizeShift;
+    entries_.resize(kMinSize);
+    auto& e = entries_[BucketId(hc)];
+    obj_malloc_used_ += e.SetString(str);
+    ++size_;
+    ++num_used_buckets_;
+
+    return true;
+  }
+
+  uint32_t bucket_id = BucketId(hc);
+  if (FindAround(str, bucket_id) < 2)
+    return false;
+
+  DCHECK_LT(bucket_id, entries_.size());
+  ++size_;
+
+  // Try insert into flat surface first. Also handle the grow case
+  // if utilization is too high.
+  for (unsigned j = 0; j < 2; ++j) {
+    int offs = FindEmptyAround(bucket_id);
+    if (CanSetFlat(offs)) {
+      auto& entry = entries_[bucket_id + offs];
+      obj_malloc_used_ += entry.SetString(str);
+      if (offs != 0) {
+        entry.SetDisplaced();
+      }
+      ++num_used_buckets_;
+      return true;
+    }
+
+    if (size_ < entries_.size())
+      break;
+
+    Grow();
+    bucket_id = BucketId(hc);
+  }
+
+  auto& dest = entries_[bucket_id];
+  DCHECK(!dest.IsEmpty());
+  if (dest.IsDisplaced()) {
+    sds sptr = dest.GetSds();
+    uint32_t nbid = BucketId(sptr);
+    Link(SuperPtr{sptr}, nbid);
+
+    if (dest.IsSds()) {
+      obj_malloc_used_ += dest.SetString(str);
+    } else {
+      LinkKey* lk = (LinkKey*)dest.get();
+      obj_malloc_used_ += lk->SetString(str);
+      dest.ClearDisplaced();
+    }
+  } else {
+    LinkKey* lk = NewLink(str, dest);
+    dest.SetLink(lk);
+  }
+  DCHECK(!dest.IsDisplaced());
+  return true;
+}
+
+unsigned StringSet::BucketDepth(uint32_t bid) const {
+  SuperPtr ptr = entries_[bid];
+  if (ptr.IsEmpty()) {
+    return 0;
+  }
+
+  unsigned res = 1;
+  while (ptr.IsLink()) {
+    LinkKey* lk = (LinkKey*)ptr.get();
+    ++res;
+    ptr = lk->next;
+    DCHECK(!ptr.IsEmpty());
+  }
+
+  return res;
+}
+
+auto StringSet::NewLink(std::string_view str, SuperPtr ptr) -> LinkKey* {
+  LinkAllocator ea(mr());
+  LinkKey* lk = ea.allocate(1);
+  ea.construct(lk);
+  obj_malloc_used_ += lk->SetString(str);
+  lk->next = ptr;
+  ++num_chain_entries_;
+
+  return lk;
+}
+
+#if 0
+void StringSet::IterateOverBucket(uint32_t bid, const ItemCb& cb) {
+  const Entry& e = entries_[bid];
+  if (e.IsEmpty()) {
+    DCHECK(!e.next);
+    return;
+  }
+  cb(e.value);
+
+  const Entry* next = e.next;
+  while (next) {
+    cb(next->value);
+    next = next->next;
+  }
+}
+#endif
+
+inline bool cmpsds(sds sp, string_view str) {
+  if (sdslen(sp) != str.size())
+    return false;
+  return str.empty() || memcmp(sp, str.data(), str.size()) == 0;
+}
+
+int StringSet::FindAround(string_view str, uint32_t bid) const {
+  SuperPtr ptr = entries_[bid];
+
+  while (ptr.IsLink()) {
+    LinkKey* lk = (LinkKey*)ptr.get();
+    sds sp = (sds)lk->get();
+    if (cmpsds(sp, str))
+      return 0;
+    ptr = lk->next;
+    DCHECK(!ptr.IsEmpty());
+  }
+
+  if (!ptr.IsEmpty()) {
+    DCHECK(ptr.IsSds());
+    sds sp = (sds)ptr.get();
+    if (cmpsds(sp, str))
+      return 0;
+  }
+
+  if (bid && entries_[bid - 1].Compare(str)) {
+    return -1;
+  }
+
+  if (bid + 1 < entries_.size() && entries_[bid + 1].Compare(str)) {
+    return 1;
+  }
+
+  return 2;
+}
+
+void StringSet::Grow() {
+  size_t prev_sz = entries_.size();
+  entries_.resize(prev_sz * 2);
+  ++capacity_log_;
+
+  for (int i = prev_sz - 1; i >= 0; --i) {
+    SuperPtr* current = &entries_[i];
+    if (current->IsEmpty()) {
+      continue;
+    }
+
+    SuperPtr* prev = nullptr;
+    while (true) {
+      SuperPtr next;
+      LinkKey* lk = nullptr;
+      sds sp;
+
+      if (current->IsLink()) {
+        lk = (LinkKey*)current->get();
+        sp = (sds)lk->get();
+        next = lk->next;
+      } else {
+        sp = (sds)current->get();
+      }
+
+      uint32_t bid = BucketId(sp);
+      if (bid != uint32_t(i)) {
+        int offs = FindEmptyAround(bid);
+        if (CanSetFlat(offs)) {
+          auto& dest = entries_[bid + offs];
+          DCHECK(!dest.IsLink());
+
+          dest.ptr = sp;
+          if (offs != 0)
+            dest.SetDisplaced();
+          if (lk) {
+            Free(lk);
+          }
+          ++num_used_buckets_;
+        } else {
+          Link(*current, bid);
+        }
+        *current = next;
+      } else {
+        current->ClearDisplaced();
+        if (lk) {
+          prev = current;
+          current = &lk->next;
+        }
+      }
+      if (next.IsEmpty())
+        break;
+    }
+
+    if (prev) {
+      DCHECK(prev->IsLink());
+      LinkKey* lk = (LinkKey*)prev->get();
+      if (lk->next.IsEmpty()) {
+        bool is_displaced = prev->IsDisplaced();
+        prev->ptr = lk->get();
+        if (is_displaced) {
+          prev->SetDisplaced();
+        }
+        Free(lk);
+      }
+    }
+
+    if (entries_[i].IsEmpty()) {
+      --num_used_buckets_;
+    }
+  }
+
+#if 0
+  unsigned cnt = 0;
+  for (auto ptr : entries_) {
+    cnt += (!ptr.IsEmpty());
+  }
+  DCHECK_EQ(num_used_buckets_, cnt);
+#endif
+}
+
+void StringSet::Link(SuperPtr ptr, uint32_t bid) {
+  SuperPtr& root = entries_[bid];
+  DCHECK(!root.IsEmpty());
+
+  bool is_root_displaced = root.IsDisplaced();
+
+  if (is_root_displaced) {
+    DCHECK_NE(bid, BucketId(root.GetSds()));
+  }
+  LinkKey* head;
+  void* val;
+
+  if (ptr.IsSds()) {
+    if (is_root_displaced) {
+      // in that case it's better to put ptr into root and move root data into its correct place.
+      sds val;
+      if (root.IsSds()) {
+        val = (sds)root.get();
+        root.ptr = ptr.get();
+      } else {
+        LinkKey* lk = (LinkKey*)root.get();
+        val = (sds)lk->get();
+        lk->ptr = ptr.get();
+        root.ClearDisplaced();
+      }
+      uint32_t nbid = BucketId(val);
+      DCHECK_NE(nbid, bid);
+
+      Link(SuperPtr{val}, nbid);  // Potentially unbounded wave of updates.
+      return;
+    }
+
+    LinkAllocator ea(mr());
+    head = ea.allocate(1);
+    ea.construct(head);
+    val = ptr.get();
+    ++num_chain_entries_;
+  } else {
+    head = (LinkKey*)ptr.get();
+    val = head->get();
+  }
+
+  if (root.IsSds()) {
+    head->ptr = root.get();
+    head->next = SuperPtr{val};
+    root.SetLink(head);
+    if (is_root_displaced) {
+      DCHECK_NE(bid, BucketId((sds)head->ptr));
+      root.SetDisplaced();
+    }
+  } else {
+    DCHECK(root.IsLink());
+    LinkKey* chain = (LinkKey*)root.get();
+    head->next = chain->next;
+    head->ptr = val;
+    chain->next.SetLink(head);
+  }
+}
+
+#if 0
+void StringSet::MoveEntry(Entry* e, uint32_t bid) {
+  auto& dest = entries_[bid];
+  if (IsEmpty(dest)) {
+    dest.value = std::move(e->value);
+    Free(e);
+    return;
+  }
+  e->next = dest.next;
+  dest.next = e;
+}
+#endif
+
+int StringSet::FindEmptyAround(uint32_t bid) const {
+  if (entries_[bid].IsEmpty())
+    return 0;
+
+  if (bid + 1 < entries_.size() && entries_[bid + 1].IsEmpty())
+    return 1;
+
+  if (bid && entries_[bid - 1].IsEmpty())
+    return -1;
+
+  return 2;
+}
+
+uint32_t StringSet::BucketId(sds ptr) const {
+  string_view sv{ptr, sdslen(ptr)};
+  return BucketId(CompactObj::HashCode(sv));
+}
+
+#if 0
+uint32_t StringSet::Scan(uint32_t cursor, const ItemCb& cb) const {
+  if (capacity_log_ == 0)
+    return 0;
+
+  uint32_t bucket_id = cursor >> (32 - capacity_log_);
+  const_iterator it(this, bucket_id);
+
+  if (it.entry_ == nullptr)
+    return 0;
+
+  bucket_id = it.bucket_id_;  // non-empty bucket
+  do {
+    cb(*it);
+    ++it;
+  } while (it.bucket_id_ == bucket_id);
+
+  if (it.entry_ == nullptr)
+    return 0;
+
+  if (it.bucket_id_ == bucket_id + 1) {  // cover displacement case
+    // TODO: we could avoid checking computing HC if we explicitly mark displacement.
+    // we have plenty-metadata to do so.
+    uint32_t bid = BucketId((*it).HashCode());
+    if (bid == it.bucket_id_) {
+      cb(*it);
+      ++it;
+    }
+  }
+
+  return it.entry_ ? it.bucket_id_ << (32 - capacity_log_) : 0;
+}
+
+bool StringSet::Erase(std::string_view val) {
+  uint64_t hc = CompactObj::HashCode(val);
+  uint32_t bid = BucketId(hc);
+
+  Entry* current = &entries_[bid];
+
+  if (!current->IsEmpty()) {
+    if (current->value == val) {
+      current->Reset();
+      ShiftLeftIfNeeded(current);
+      --size_;
+      return true;
+    }
+
+    Entry* prev = current;
+    current = current->next;
+    while (current) {
+      if (current->value == val) {
+        current->Reset();
+        prev->next = current->next;
+        Free(current);
+        --size_;
+        return true;
+      }
+      prev = current;
+      current = current->next;
+    }
+  }
+
+  auto& prev = entries_[bid - 1];
+  // TODO: to mark displacement.
+  if (bid && !prev.IsEmpty()) {
+    if (prev.value == val) {
+      obj_malloc_used_ -= prev.value.MallocUsed();
+
+      prev.Reset();
+      ShiftLeftIfNeeded(&prev);
+      --size_;
+      return true;
+    }
+  }
+
+  auto& next = entries_[bid + 1];
+  if (bid + 1 < entries_.size()) {
+    if (next.value == val) {
+      obj_malloc_used_ -= next.value.MallocUsed();
+      next.Reset();
+      ShiftLeftIfNeeded(&next);
+      --size_;
+      return true;
+    }
+  }
+
+  return false;
+}
+
+#endif
+
+void StringSet::iterator::SeekNonEmpty() {
+  while (bucket_id_ < owner_->entries_.size()) {
+    if (!owner_->entries_[bucket_id_].IsEmpty()) {
+      entry_ = &owner_->entries_[bucket_id_];
+      return;
+    }
+    ++bucket_id_;
+  }
+  entry_ = nullptr;
+}
+
+void StringSet::const_iterator::SeekNonEmpty() {
+  while (bucket_id_ < owner_->entries_.size()) {
+    if (!owner_->entries_[bucket_id_].IsEmpty()) {
+      entry_ = &owner_->entries_[bucket_id_];
+      return;
+    }
+    ++bucket_id_;
+  }
+  entry_ = nullptr;
+}
+
+}  // namespace dfly

src/core/string_set.h

@@ -0,0 +1,332 @@
+// Copyright 2022, Roman Gershman.  All rights reserved.
+// See LICENSE for licensing terms.
+//
+#pragma once
+
+#include <memory_resource>
+
+extern "C" {
+#include "redis/object.h"
+}
+
+namespace dfly {
+
+// StringSet is a nice but over-optimized data-structure. Probably is not worth it in the first
+// place but sometimes the OCD kicks in and one can not resist.
+// The advantage of it over redis-dict is smaller meta-data waste.
+// dictEntry is 24 bytes, i.e it uses at least 32N bytes where N is the expected length.
+// dict requires to allocate dictEntry per each addition in addition to the supplied key.
+// It also wastes space in case of a set because it stores a value pointer inside dictEntry.
+// To summarize:
+// 100% utilized dict uses N*24 + N*8 = 32N bytes not including the key space.
+// for 75% utilization (1/0.75 buckets): N*1.33*8 + N*24 = 35N
+//
+// This class uses 8 bytes per bucket (similarly to dictEntry*) but it used it for both
+// links and keys. For most cases, we remove the need for another redirection layer
+// and just store the key, so no "dictEntry" allocations occur.
+// For those cells that require chaining, the bucket is
+// changed in run-time to represent a linked chain.
+// Additional feature - in order to to reduce collisions, we insert items into
+// neighbour cells but only if they are empty (not chains). This way we reduce the number of
+// empty (unused) spaces at full utilization from 36% to ~21%.
+// 100% utilized table requires: N*8 + 0.2N*16 = 11.2N bytes or ~20 bytes savings.
+// 75% utilization: N*1.33*8 + 0.12N*16 = 13N or ~22 bytes savings per record.
+// TODO: to separate hash/compare functions from table logic and make it generic
+// with potential replacements of hset/zset data structures.
+// static_assert(sizeof(dictEntry) == 24);
+
+class StringSet {
+  struct LinkKey;
+  // we can assume that high 12 bits of user address space
+  // can be used for tagging. At most 52 bits of address are reserved for
+  // some configurations, and usually it's 48 bits.
+  // https://www.kernel.org/doc/html/latest/arm64/memory.html
+  static constexpr size_t kLinkBit = 1ULL << 52;
+  static constexpr size_t kDisplaceBit = 1ULL << 53;
+  static constexpr size_t kTagMask = 4095ULL << 51; // we reserve 12 high bits.
+
+  struct SuperPtr {
+    void* ptr = nullptr;  //
+
+    explicit SuperPtr(void* p = nullptr) : ptr(p) {
+    }
+
+    bool IsSds() const {
+      return (uintptr_t(ptr) & kLinkBit) == 0;
+    }
+
+    bool IsLink() const {
+      return (uintptr_t(ptr) & kLinkBit) == kLinkBit;
+    }
+
+    bool IsEmpty() const {
+      return ptr == nullptr;
+    }
+
+    void* get() const {
+      return (void*)(uintptr_t(ptr) & ~kTagMask);
+    }
+
+    bool IsDisplaced() const {
+      return (uintptr_t(ptr) & kDisplaceBit) == kDisplaceBit;
+    }
+
+    // returns usable size.
+    size_t SetString(std::string_view str);
+
+    void SetLink(LinkKey* lk) {
+      ptr = (void*)(uintptr_t(lk) | kLinkBit);
+    }
+
+    bool Compare(std::string_view str) const;
+
+    void SetDisplaced() {
+      ptr = (void*)(uintptr_t(ptr) | kDisplaceBit);
+    }
+
+    void ClearDisplaced() {
+      ptr = (void*)(uintptr_t(ptr) & ~kDisplaceBit);
+    }
+
+    void Reset() {
+      ptr = nullptr;
+    }
+
+    sds GetSds() const {
+      if (IsSds())
+        return (sds)get();
+      LinkKey* lk = (LinkKey*)get();
+      return (sds)lk->get();
+    }
+  };
+
+  struct LinkKey : public SuperPtr {
+    SuperPtr next;  // could be LinkKey* or sds.
+  };
+
+  static_assert(sizeof(SuperPtr) == 8);
+
+ public:
+  class iterator;
+  class const_iterator;
+  // using ItemCb = std::function<void(const CompactObj& co)>;
+
+  StringSet(const StringSet&) = delete;
+
+  explicit StringSet(std::pmr::memory_resource* mr = std::pmr::get_default_resource());
+  ~StringSet();
+
+  StringSet& operator=(StringSet&) = delete;
+
+  void Reserve(size_t sz);
+
+  bool Add(std::string_view str);
+
+  bool Remove(std::string_view str);
+
+  void Erase(iterator it);
+
+  size_t size() const {
+    return size_;
+  }
+
+  bool empty() const {
+    return size_ == 0;
+  }
+
+  size_t bucket_count() const {
+    return entries_.size();
+  }
+
+  // those that are chained to the entries stored inline in the bucket array.
+  size_t num_chain_entries() const {
+    return num_chain_entries_;
+  }
+
+  size_t num_used_buckets() const {
+    return num_used_buckets_;
+  }
+
+  bool Contains(std::string_view val) const;
+
+  bool Erase(std::string_view val);
+
+  iterator begin() {
+    return iterator{this, 0};
+  }
+
+  iterator end() {
+    return iterator{};
+  }
+
+  size_t obj_malloc_used() const {
+    return obj_malloc_used_;
+  }
+
+  size_t set_malloc_used() const {
+    return (num_chain_entries_ + entries_.capacity()) * sizeof(SuperPtr);
+  }
+
+  /// stable scanning api. has the same guarantees as redis scan command.
+  /// we avoid doing bit-reverse by using a different function to derive a bucket id
+  /// from hash values. By using msb part of hash we make it "stable" with respect to
+  /// rehashes. For example, with table log size 4 (size 16), entries in bucket id
+  /// 1110 come from hashes 1110XXXXX.... When a table grows to log size 5,
+  /// these entries can move either to 11100 or 11101. So if we traversed with our cursor
+  /// range [0000-1110], it's guaranteed that in grown table we do not need to cover again
+  /// [00000-11100]. Similarly with shrinkage, if a table is shrinked to log size 3,
+  /// keys from 1110 and 1111 will move to bucket 111. Again, it's guaranteed that we
+  /// covered the range [000-111] (all keys in that case).
+  /// Returns: next cursor or 0 if reached the end of scan.
+  /// cursor = 0 - initiates a new scan.
+  // uint32_t Scan(uint32_t cursor, const ItemCb& cb) const;
+
+  unsigned BucketDepth(uint32_t bid) const;
+
+  // void IterateOverBucket(uint32_t bid, const ItemCb& cb);
+
+  class iterator {
+    friend class StringSet;
+
+   public:
+    iterator() : owner_(nullptr), entry_(nullptr), bucket_id_(0) {
+    }
+
+    iterator& operator++();
+
+    bool operator==(const iterator& o) const {
+      return entry_ == o.entry_;
+    }
+
+    bool operator!=(const iterator& o) const {
+      return !(*this == o);
+    }
+
+   private:
+    iterator(StringSet* owner, uint32_t bid) : owner_(owner), bucket_id_(bid) {
+      SeekNonEmpty();
+    }
+
+    void SeekNonEmpty();
+
+    StringSet* owner_ = nullptr;
+    SuperPtr* entry_ = nullptr;
+    uint32_t bucket_id_ = 0;
+  };
+
+  class const_iterator {
+    friend class StringSet;
+
+   public:
+    const_iterator() : owner_(nullptr), entry_(nullptr), bucket_id_(0) {
+    }
+
+    const_iterator& operator++();
+
+    const_iterator& operator=(iterator& it) {
+      owner_ = it.owner_;
+      entry_ = it.entry_;
+      bucket_id_ = it.bucket_id_;
+
+      return *this;
+    }
+
+    bool operator==(const const_iterator& o) const {
+      return entry_ == o.entry_;
+    }
+
+    bool operator!=(const const_iterator& o) const {
+      return !(*this == o);
+    }
+
+   private:
+    const_iterator(const StringSet* owner, uint32_t bid) : owner_(owner), bucket_id_(bid) {
+      SeekNonEmpty();
+    }
+
+    void SeekNonEmpty();
+
+    const StringSet* owner_ = nullptr;
+    const SuperPtr* entry_ = nullptr;
+    uint32_t bucket_id_ = 0;
+  };
+
+ private:
+  friend class iterator;
+
+  using LinkAllocator = std::pmr::polymorphic_allocator<LinkKey>;
+
+  std::pmr::memory_resource* mr() {
+    return entries_.get_allocator().resource();
+  }
+
+  uint32_t BucketId(uint64_t hash) const {
+    return hash >> (64 - capacity_log_);
+  }
+
+  uint32_t BucketId(sds ptr) const;
+
+  // Returns: 2 if no empty spaces found around the bucket. 0, -1, 1 - offset towards
+  // an empty bucket.
+  int FindEmptyAround(uint32_t bid) const;
+
+  // returns 2 if no object was found in the vicinity.
+  // Returns relative offset to bid: 0, -1, 1 if found.
+  int FindAround(std::string_view str, uint32_t bid) const;
+
+  void Grow();
+
+  void Link(SuperPtr ptr, uint32_t bid);
+  /*void MoveEntry(Entry* e, uint32_t bid);
+
+  void ShiftLeftIfNeeded(Entry* root) {
+    if (root->next) {
+      root->value = std::move(root->next->value);
+      Entry* tmp = root->next;
+      root->next = root->next->next;
+      Free(tmp);
+    }
+  }
+  */
+  void Free(LinkKey* lk) {
+    mr()->deallocate(lk, sizeof(LinkKey), alignof(LinkKey));
+    --num_chain_entries_;
+  }
+
+  LinkKey* NewLink(std::string_view str, SuperPtr ptr);
+
+  // The rule is - entries can be moved to vicinity as long as they are stored
+  // "flat", i.e. not into the linked list. The linked list
+  std::pmr::vector<SuperPtr> entries_;
+  size_t obj_malloc_used_ = 0;
+  uint32_t size_ = 0;
+  uint32_t num_chain_entries_ = 0;
+  uint32_t num_used_buckets_ = 0;
+  unsigned capacity_log_ = 0;
+};
+
+#if 0
+inline StringSet::iterator& StringSet::iterator::operator++() {
+  if (entry_->next) {
+    entry_ = entry_->next;
+  } else {
+    ++bucket_id_;
+    SeekNonEmpty();
+  }
+
+  return *this;
+}
+
+inline StringSet::const_iterator& StringSet::const_iterator::operator++() {
+  if (entry_->next) {
+    entry_ = entry_->next;
+  } else {
+    ++bucket_id_;
+    SeekNonEmpty();
+  }
+
+  return *this;
+}
+#endif
+
+}  // namespace dfly

src/core/string_set_test.cc

@@ -0,0 +1,131 @@
+// Copyright 2022, Roman Gershman.  All rights reserved.
+// See LICENSE for licensing terms.
+//
+
+#include "core/string_set.h"
+
+#include <absl/strings/str_cat.h>
+#include <gmock/gmock.h>
+#include <mimalloc.h>
+
+#include <unordered_set>
+
+#include "base/gtest.h"
+#include "base/logging.h"
+#include "core/mi_memory_resource.h"
+
+extern "C" {
+#include "redis/zmalloc.h"
+}
+
+namespace dfly {
+
+using namespace std;
+
+class StringSetTest : public ::testing::Test {
+ protected:
+  static void SetUpTestSuite() {
+    auto* tlh = mi_heap_get_backing();
+    init_zmalloc_threadlocal(tlh);
+    // SmallString::InitThreadLocal(tlh);
+
+    // static MiMemoryResource mi_resource(tlh);
+    // needed for MallocUsed
+    // CompactObj::InitThreadLocal(&mi_resource);
+  }
+
+  static void TearDownTestSuite() {
+  }
+
+  StringSet ss_;
+};
+
+TEST_F(StringSetTest, Basic) {
+  EXPECT_TRUE(ss_.Add("foo"));
+  EXPECT_TRUE(ss_.Add("bar"));
+  EXPECT_FALSE(ss_.Add("foo"));
+  EXPECT_FALSE(ss_.Add("bar"));
+  EXPECT_EQ(2, ss_.size());
+}
+
+TEST_F(StringSetTest, Ex1) {
+  EXPECT_TRUE(ss_.Add("AA@@@@@@@@@@@@@@"));
+  EXPECT_TRUE(ss_.Add("AAA@@@@@@@@@@@@@"));
+  EXPECT_TRUE(ss_.Add("AAAAAAAAA@@@@@@@"));
+  EXPECT_TRUE(ss_.Add("AAAAAAAAAA@@@@@@"));
+  EXPECT_TRUE(ss_.Add("AAAAAAAAAAAAAAA@"));
+  EXPECT_TRUE(ss_.Add("BBBBBAAAAAAAAAAA"));
+  EXPECT_TRUE(ss_.Add("BBBBBBBBAAAAAAAA"));
+  EXPECT_TRUE(ss_.Add("CCCCCBBBBBBBBBBB"));
+}
+
+TEST_F(StringSetTest, Many) {
+  double max_chain_factor = 0;
+  for (unsigned i = 0; i < 8192; ++i) {
+    EXPECT_TRUE(ss_.Add(absl::StrCat("xxxxxxxxxxxxxxxxx", i)));
+    size_t sz = ss_.size();
+    bool should_print = (sz == ss_.bucket_count()) || (sz == ss_.bucket_count() * 0.75);
+    if (should_print) {
+      double chain_usage = double(ss_.num_chain_entries()) / ss_.size();
+      unsigned num_empty = ss_.bucket_count() - ss_.num_used_buckets();
+      double empty_factor = double(num_empty) / ss_.bucket_count();
+
+      LOG(INFO) << "chains: " << 100 * chain_usage << ", empty: " << 100 * empty_factor << "% at "
+                << ss_.size();
+#if 0
+      if (ss_.size() == 15) {
+        for (unsigned i = 0; i < ss_.bucket_count(); ++i) {
+          LOG(INFO) << "[" << i << "]: " << ss_.BucketDepth(i);
+        }
+        /*ss_.IterateOverBucket(93, [this](const CompactObj& co) {
+          LOG(INFO) << "93->" << (co.HashCode() % ss_.bucket_count());
+        });*/
+      }
+#endif
+    }
+  }
+  EXPECT_EQ(8192, ss_.size());
+
+  LOG(INFO) << "max chain factor: " << 100 * max_chain_factor << "%";
+  /*size_t iter_len = 0;
+  for (auto it = ss_.begin(); it != ss_.end(); ++it) {
+    ++iter_len;
+  }
+  EXPECT_EQ(iter_len, 512);*/
+}
+
+#if 0
+TEST_F(StringSetTest, IterScan) {
+  unordered_set<string> actual, expected;
+  auto insert_actual = [&](const CompactObj& val) {
+    string tmp;
+    val.GetString(&tmp);
+    actual.insert(tmp);
+  };
+
+  EXPECT_EQ(0, ss_.Scan(0, insert_actual));
+  EXPECT_TRUE(actual.empty());
+
+  for (unsigned i = 0; i < 512; ++i) {
+    string s = absl::StrCat("x", i);
+    expected.insert(s);
+    EXPECT_TRUE(ss_.Add(s));
+  }
+
+
+  for (CompactObj& val : ss_) {
+    insert_actual(val);
+  }
+
+  EXPECT_EQ(actual, expected);
+  actual.clear();
+  uint32_t cursor = 0;
+  do {
+    cursor = ss_.Scan(cursor, insert_actual);
+  } while (cursor);
+  EXPECT_EQ(actual, expected);
+}
+
+#endif
+
+}  // namespace dfly

src/server/set_family.cc

@@ -11,17 +11,22 @@ extern "C" {
 #include "redis/util.h"
 }
 
+#include "base/flags.h"
 #include "base/logging.h"
 #include "base/stl_util.h"
+#include "core/string_set.h"
 #include "server/command_registry.h"
 #include "server/conn_context.h"
 #include "server/engine_shard_set.h"
 #include "server/error.h"
 #include "server/transaction.h"
 
+ABSL_FLAG(bool, use_set2, false, "If true use better implementation for sets");
+
 namespace dfly {
 
 using namespace std;
+using absl::GetFlag;
 
 using ResultStringVec = vector<OpResult<vector<string>>>;
 using ResultSetView = OpResult<absl::flat_hash_set<std::string_view>>;
@@ -52,24 +57,103 @@ intset* IntsetAddSafe(string_view val, intset* is, bool* success, bool* added) {
   return is;
 }
 
+bool dictContains(const dict* d, string_view key) {
+  uint64_t h = dictGenHashFunction(key.data(), key.size());
+
+  for (unsigned table = 0; table <= 1; table++) {
+    uint64_t idx = h & DICTHT_SIZE_MASK(d->ht_size_exp[table]);
+    dictEntry* he = d->ht_table[table][idx];
+    while (he) {
+      sds dkey = (sds)he->key;
+      if (sdslen(dkey) == key.size() && (key.empty() || memcmp(dkey, key.data(), key.size()) == 0))
+        return true;
+      he = he->next;
+    }
+    if (!dictIsRehashing(d))
+      break;
+  }
+  return false;
+}
+
+pair<unsigned, bool> RemoveStrSet(ArgSlice vals, CompactObj* set) {
+  unsigned removed = 0;
+  bool isempty = false;
+  if (GetFlag(FLAGS_use_set2)) {
+    StringSet* ss = (StringSet*)set->RObjPtr();
+    for (auto member : vals) {
+      // removed += ss->Erase(member);
+    }
+    isempty = ss->empty();
+  } else {
+    dict* d = (dict*)set->RObjPtr();
+    auto* shard = EngineShard::tlocal();
+
+    for (auto member : vals) {
+      shard->tmp_str1 = sdscpylen(shard->tmp_str1, member.data(), member.size());
+      int result = dictDelete(d, shard->tmp_str1);
+      removed += (result == DICT_OK);
+    }
+    isempty = (dictSize(d) == 0);
+  }
+  return make_pair(removed, isempty);
+}
+
 unsigned AddStrSet(ArgSlice vals, CompactObj* dest) {
   unsigned res = 0;
 
-  dict* ds = (dict*)dest->RObjPtr();
-  auto* es = EngineShard::tlocal();
+  if (GetFlag(FLAGS_use_set2)) {
+    StringSet* ss = (StringSet*)dest->RObjPtr();
+    for (auto member : vals) {
+      res += ss->Add(member);
+    }
+  } else {
+    dict* ds = (dict*)dest->RObjPtr();
+    auto* es = EngineShard::tlocal();
 
-  for (auto member : vals) {
-    es->tmp_str1 = sdscpylen(es->tmp_str1, member.data(), member.size());
-    dictEntry* de = dictAddRaw(ds, es->tmp_str1, NULL);
-    if (de) {
-      de->key = sdsdup(es->tmp_str1);
-      ++res;
+    for (auto member : vals) {
+      es->tmp_str1 = sdscpylen(es->tmp_str1, member.data(), member.size());
+      dictEntry* de = dictAddRaw(ds, es->tmp_str1, NULL);
+      if (de) {
+        de->key = sdsdup(es->tmp_str1);
+        ++res;
+      }
     }
   }
 
   return res;
 }
 
+void InitStrSet(CompactObj* set) {
+  if (GetFlag(FLAGS_use_set2)) {
+    StringSet* ss = new StringSet{CompactObj::memory_resource()};
+    set->InitRobj(OBJ_SET, kEncodingStrMap2, ss);
+  } else {
+    dict* ds = dictCreate(&setDictType);
+    set->InitRobj(OBJ_SET, kEncodingStrMap, ds);
+  }
+}
+
+// f receives a str object.
+template <typename F> void FillFromStrSet(F&& f, void* ptr) {
+  string str;
+  if (GetFlag(FLAGS_use_set2)) {
+    /*for (const CompactObj& co : *(StringSet*)ptr) {
+      co.GetString(&str);
+      f(move(str));
+    }*/
+  } else {
+    dict* ds = (dict*)ptr;
+
+    dictIterator* di = dictGetIterator(ds);
+    dictEntry* de = nullptr;
+    while ((de = dictNext(di))) {
+      str.assign((sds)de->key, sdslen((sds)de->key));
+      f(move(str));
+    }
+    dictReleaseIterator(di);
+  }
+}
+
 // returns (removed, isempty)
 pair<unsigned, bool> RemoveSet(ArgSlice vals, CompactObj* set) {
   bool isempty = false;
@@ -91,14 +175,7 @@ pair<unsigned, bool> RemoveSet(ArgSlice vals, CompactObj* set) {
     isempty = (intsetLen(is) == 0);
     set->SetRObjPtr(is);
   } else {
-    dict* d = (dict*)set->RObjPtr();
-    auto* shard = EngineShard::tlocal();
-    for (auto member : vals) {
-      shard->tmp_str1 = sdscpylen(shard->tmp_str1, member.data(), member.size());
-      int result = dictDelete(d, shard->tmp_str1);
-      removed += (result == DICT_OK);
-    }
-    isempty = (dictSize(d) == 0);
+    return RemoveStrSet(vals, set);
   }
   return make_pair(removed, isempty);
 }
@@ -118,8 +195,7 @@ void InitSet(ArgSlice vals, CompactObj* set) {
     intset* is = intsetNew();
     set->InitRobj(OBJ_SET, kEncodingIntSet, is);
   } else {
-    dict* ds = dictCreate(&setDictType);
-    set->InitRobj(OBJ_SET, kEncodingStrMap, ds);
+    InitStrSet(set);
   }
 }
 
@@ -132,43 +208,126 @@ void ScanCallback(void* privdata, const dictEntry* de) {
 uint64_t ScanStrSet(const CompactObj& co, uint64_t curs, unsigned count, StringVec* res) {
   long maxiterations = count * 10;
 
-  DCHECK_EQ(kEncodingStrMap, co.Encoding());
-  dict* ds = (dict*)co.RObjPtr();
-  do {
-    curs = dictScan(ds, curs, ScanCallback, NULL, res);
-  } while (curs && maxiterations-- && res->size() < count);
+  if (GetFlag(FLAGS_use_set2)) {
+    DCHECK_EQ(kEncodingStrMap2, co.Encoding());
+    LOG(DFATAL) << "TBD: Not implemented";
+  } else {
+    DCHECK_EQ(kEncodingStrMap, co.Encoding());
+    dict* ds = (dict*)co.RObjPtr();
+    do {
+      curs = dictScan(ds, curs, ScanCallback, NULL, res);
+    } while (curs && maxiterations-- && res->size() < count);
+  }
 
   return curs;
 }
 
 using SetType = pair<void*, unsigned>;
 
+uint32_t SetTypeLen(const SetType& set) {
+  if (set.second == kEncodingIntSet) {
+    return intsetLen((const intset*)set.first);
+  }
+
+  if (GetFlag(FLAGS_use_set2)) {
+    DCHECK_EQ(set.second, kEncodingStrMap2);
+    return ((StringSet*)set.first)->size();
+  } else {
+    DCHECK_EQ(set.second, kEncodingStrMap);
+    return dictSize((const dict*)set.first);
+  }
+};
+
+bool IsInSet(const SetType& st, int64_t val) {
+  if (st.second == kEncodingIntSet)
+    return intsetFind((intset*)st.first, val);
+
+  char buf[32];
+  char* next = absl::numbers_internal::FastIntToBuffer(val, buf);
+  string_view str{buf, size_t(next - buf)};
+  if (GetFlag(FLAGS_use_set2)) {
+    LOG(DFATAL) << "TBD: Not implemented";
+    return false;
+  } else {
+    DCHECK_EQ(st.second, kEncodingStrMap);
+    return dictContains((dict*)st.first, str);
+  }
+}
+
+bool IsInSet(const SetType& st, string_view member) {
+  if (st.second == kEncodingIntSet) {
+    long long llval;
+    if (!string2ll(member.data(), member.size(), &llval))
+      return false;
+
+    return intsetFind((intset*)st.first, llval);
+  }
+
+  if (GetFlag(FLAGS_use_set2)) {
+    LOG(DFATAL) << "TBD: Not implemented";
+    return false;
+  } else {
+    DCHECK_EQ(st.second, kEncodingStrMap);
+    return dictContains((dict*)st.first, member);
+  }
+}
+
 // Removes arg from result.
 void DiffStrSet(const SetType& st, absl::flat_hash_set<string>* result) {
-  DCHECK_EQ(kEncodingStrMap, st.second);
-  dict* ds = (dict*)st.first;
-  dictIterator* di = dictGetIterator(ds);
-  dictEntry* de = nullptr;
-  while ((de = dictNext(di))) {
-    sds key = (sds)de->key;
-    result->erase(string_view{key, sdslen(key)});
+  if (GetFlag(FLAGS_use_set2)) {
+  } else {
+    DCHECK_EQ(kEncodingStrMap, st.second);
+    dict* ds = (dict*)st.first;
+    dictIterator* di = dictGetIterator(ds);
+    dictEntry* de = nullptr;
+    while ((de = dictNext(di))) {
+      sds key = (sds)de->key;
+      result->erase(string_view{key, sdslen(key)});
+    }
+    dictReleaseIterator(di);
+  }
+}
+
+void InterStrSet(const vector<SetType>& vec, StringVec* result) {
+  if (GetFlag(FLAGS_use_set2)) {
+  } else {
+    dict* ds = (dict*)vec.front().first;
+    dictIterator* di = dictGetIterator(ds);
+    dictEntry* de = nullptr;
+    while ((de = dictNext(di))) {
+      size_t j = 1;
+      sds key = (sds)de->key;
+      string_view member{key, sdslen(key)};
+
+      for (j = 1; j < vec.size(); j++) {
+        if (vec[j].first != ds && !IsInSet(vec[j], member))
+          break;
+      }
+
+      /* Only take action when all vec contain the member */
+      if (j == vec.size()) {
+        result->push_back(string(member));
+      }
+    }
+    dictReleaseIterator(di);
   }
-  dictReleaseIterator(di);
 }
 
 StringVec PopStrSet(unsigned count, const SetType& st) {
   StringVec result;
-  dict* ds = (dict*)st.first;
-  string str;
-  dictIterator* di = dictGetSafeIterator(ds);
-  for (uint32_t i = 0; i < count; ++i) {
-    dictEntry* de = dictNext(di);
-    DCHECK(de);
-    result.emplace_back((sds)de->key, sdslen((sds)de->key));
-    dictDelete(ds, de->key);
+  if (GetFlag(FLAGS_use_set2)) {
+  } else {
+    dict* ds = (dict*)st.first;
+    string str;
+    dictIterator* di = dictGetSafeIterator(ds);
+    for (uint32_t i = 0; i < count; ++i) {
+      dictEntry* de = dictNext(di);
+      DCHECK(de);
+      result.emplace_back((sds)de->key, sdslen((sds)de->key));
+      dictDelete(ds, de->key);
+    }
+    dictReleaseIterator(di);
   }
-  dictReleaseIterator(di);
-
   return result;
 }
 
@@ -290,58 +449,6 @@ OpStatus NoOpCb(Transaction* t, EngineShard* shard) {
   return OpStatus::OK;
 };
 
-using SetType = pair<void*, unsigned>;
-
-uint32_t SetTypeLen(const SetType& set) {
-  if (set.second == kEncodingStrMap) {
-    return dictSize((const dict*)set.first);
-  }
-  DCHECK_EQ(set.second, kEncodingIntSet);
-  return intsetLen((const intset*)set.first);
-};
-
-bool dictContains(const dict* d, string_view key) {
-  uint64_t h = dictGenHashFunction(key.data(), key.size());
-
-  for (unsigned table = 0; table <= 1; table++) {
-    uint64_t idx = h & DICTHT_SIZE_MASK(d->ht_size_exp[table]);
-    dictEntry* he = d->ht_table[table][idx];
-    while (he) {
-      sds dkey = (sds)he->key;
-      if (sdslen(dkey) == key.size() && (key.empty() || memcmp(dkey, key.data(), key.size()) == 0))
-        return true;
-      he = he->next;
-    }
-    if (!dictIsRehashing(d))
-      break;
-  }
-  return false;
-}
-
-bool IsInSet(const SetType& st, int64_t val) {
-  if (st.second == kEncodingIntSet)
-    return intsetFind((intset*)st.first, val);
-
-  DCHECK_EQ(st.second, kEncodingStrMap);
-  char buf[32];
-  char* next = absl::numbers_internal::FastIntToBuffer(val, buf);
-
-  return dictContains((dict*)st.first, string_view{buf, size_t(next - buf)});
-}
-
-bool IsInSet(const SetType& st, string_view member) {
-  if (st.second == kEncodingIntSet) {
-    long long llval;
-    if (!string2ll(member.data(), member.size(), &llval))
-      return false;
-
-    return intsetFind((intset*)st.first, llval);
-  }
-  DCHECK_EQ(st.second, kEncodingStrMap);
-
-  return dictContains((dict*)st.first, member);
-}
-
 template <typename F> void FillSet(const SetType& set, F&& f) {
   if (set.second == kEncodingIntSet) {
     intset* is = (intset*)set.first;
@@ -354,15 +461,7 @@ template <typename F> void FillSet(const SetType& set, F&& f) {
       f(string{buf, size_t(next - buf)});
     }
   } else {
-    dict* ds = (dict*)set.first;
-    string str;
-    dictIterator* di = dictGetIterator(ds);
-    dictEntry* de = nullptr;
-    while ((de = dictNext(di))) {
-      str.assign((sds)de->key, sdslen((sds)de->key));
-      f(move(str));
-    }
-    dictReleaseIterator(di);
+    FillFromStrSet(move(f), set.first);
   }
 }
 
@@ -428,7 +527,11 @@ OpResult<uint32_t> OpAdd(const OpArgs& op_args, std::string_view key, ArgSlice v
           return OpStatus::OUT_OF_MEMORY;
         }
         // frees 'is' on a way.
-        co.InitRobj(OBJ_SET, kEncodingStrMap, tmp.ptr);
+        if (GetFlag(FLAGS_use_set2)) {
+          co.InitRobj(OBJ_SET, kEncodingStrMap2, tmp.ptr);
+        } else {
+          co.InitRobj(OBJ_SET, kEncodingStrMap, tmp.ptr);
+        }
         inner_obj = co.RObjPtr();
         break;
       }
@@ -691,25 +794,7 @@ OpResult<StringVec> OpInter(const Transaction* t, EngineShard* es, bool remove_f
       }
     }
   } else {
-    dict* ds = (dict*)sets.front().first;
-    dictIterator* di = dictGetIterator(ds);
-    dictEntry* de = nullptr;
-    while ((de = dictNext(di))) {
-      size_t j = 1;
-      sds key = (sds)de->key;
-      string_view member{key, sdslen(key)};
-
-      for (j = 1; j < sets.size(); j++) {
-        if (sets[j].first != ds && !IsInSet(sets[j], member))
-          break;
-      }
-
-      /* Only take action when all sets contain the member */
-      if (j == sets.size()) {
-        result.push_back(string(member));
-      }
-    }
-    dictReleaseIterator(di);
+    InterStrSet(sets, &result);
   }
 
   return result;
@@ -1193,25 +1278,39 @@ bool SetFamily::ConvertToStrSet(const intset* is, size_t expected_len, robj* des
   char buf[32];
   int ii = 0;
 
-  dict* ds = dictCreate(&setDictType);
+  if (GetFlag(FLAGS_use_set2)) {
+    StringSet* ss = new StringSet{CompactObj::memory_resource()};
+    if (expected_len)
+      ss->Reserve(expected_len);
 
-  if (expected_len) {
-    if (dictTryExpand(ds, expected_len) != DICT_OK) {
-      dictRelease(ds);
-      return false;
+    while (intsetGet(const_cast<intset*>(is), ii++, &intele)) {
+      char* next = absl::numbers_internal::FastIntToBuffer(intele, buf);
+      string_view str{buf, size_t(next - buf)};
+      CHECK(ss->Add(str));
     }
+
+    dest->ptr = ss;
+    dest->encoding = OBJ_ENCODING_HT;
+  } else {
+    dict* ds = dictCreate(&setDictType);
+
+    if (expected_len) {
+      if (dictTryExpand(ds, expected_len) != DICT_OK) {
+        dictRelease(ds);
+        return false;
+      }
+    }
+
+    /* To add the elements we extract integers and create redis objects */
+    while (intsetGet(const_cast<intset*>(is), ii++, &intele)) {
+      char* next = absl::numbers_internal::FastIntToBuffer(intele, buf);
+      sds s = sdsnewlen(buf, next - buf);
+      CHECK(dictAddRaw(ds, s, NULL));
+    }
+
+    dest->ptr = ds;
+    dest->encoding = OBJ_ENCODING_HT;
   }
-
-  /* To add the elements we extract integers and create redis objects */
-  while (intsetGet(const_cast<intset*>(is), ii++, &intele)) {
-    char* next = absl::numbers_internal::FastIntToBuffer(intele, buf);
-    sds s = sdsnewlen(buf, next - buf);
-    CHECK(dictAddRaw(ds, s, NULL));
-  }
-
-  dest->ptr = ds;
-  dest->encoding = OBJ_ENCODING_HT;
-
   return true;
 }