diff --git a/conanfile.py b/conanfile.py
index 1cf54553..5a75270a 100644
--- a/conanfile.py
+++ b/conanfile.py
@@ -40,7 +40,7 @@ def build_requirements(self):
         self.build_requires("gtest/1.14.0")
 
     def requirements(self):
-        self.requires("homestore/[~=4,      include_prerelease=True]@oss/master")
+        self.requires("homestore/[~=4,      include_prerelease=True]@user/testing")
         self.requires("sisl/[~=10,          include_prerelease=True]@oss/master")
         # Remove when HomeStore Replication Service is mature
         self.requires("nuraft_mesg/[~=1,    include_prerelease=True]@oss/main")
diff --git a/src/lib/homestore/heap_chunk_selector.cpp b/src/lib/homestore/heap_chunk_selector.cpp
index 59e9c710..d3853d6b 100644
--- a/src/lib/homestore/heap_chunk_selector.cpp
+++ b/src/lib/homestore/heap_chunk_selector.cpp
@@ -1,38 +1,69 @@
 #include "heap_chunk_selector.h"
 
+#include <execution>
+#include <algorithm>
+#include <sisl/logging/logging.h>
+
 namespace homeobject {
 
+//add_chunk may be called in Homeobject(when sealing a shard) or in Homestore(when adding chunks to vdev) concurently
 void HeapChunkSelector::add_chunk(csharedChunk& chunk) {
-    uint32_t pdevId = VChunk(chunk).get_pdev_id();
-    auto it = _pdev_heap_map.find(pdevId);
-    if (it == _pdev_heap_map.end()) it = _pdev_heap_map.try_emplace(pdevId, std::make_shared<std::pair<std::mutex, VChunkHeap>>()).first;
-    std::unique_lock<std::mutex> l(it->second->first);
+    VChunk vchunk(chunk);
+    auto&& pdevId = vchunk.get_pdev_id();
+    auto&& [it, ok] = m_pdev_heap_map.try_emplace(pdevId, std::make_shared<std::pair<std::mutex, VChunkHeap>>());
+    
+    {
+        std::lock_guard<std::mutex> l(lock); 
+        if(ok) m_pdev_avalable_blk_map.emplace(pdevId, vchunk.available_blks());
+        else m_pdev_avalable_blk_map[pdevId] += vchunk.available_blks();
+        m_chunks.emplace(chunk);
+    }
+
+    std::lock_guard<std::mutex> l(it->second->first); 
     it->second->second.emplace(chunk);
 }
 
+//select_chunk will only be called in homestore when allocating a block
 csharedChunk HeapChunkSelector::select_chunk(homestore::blk_count_t count, const homestore::blk_alloc_hints& hint) {
-    auto it = _pdev_heap_map.find(hint.dev_id_hint);
-    if (it == _pdev_heap_map.end()) return nullptr;
-    std::unique_lock<std::mutex> l(it->second->first);
-    if(it->second->second.empty()) return nullptr;
+    auto pdevID = hint.dev_id_hint;
+    if(pdevID == homestore::INVALID_DEV_ID) {
+    // this is the first shard of this pg, select a pdev with the most available blocks for it
+        std::lock_guard<std::mutex> l(lock);
+        auto it = std::max_element(m_pdev_avalable_blk_map.begin(), m_pdev_avalable_blk_map.end(),
+        [](const std::pair<uint32_t, uint32_t>& lhs, const std::pair<uint32_t, uint32_t>& rhs) { 
+            return lhs.second < rhs.second; }
+        );
+        pdevID = it->first;
+    }
+
+    auto it = m_pdev_heap_map.find(pdevID);
+    if (it == m_pdev_heap_map.end()) {
+        LOGINFO("No pdev found for pdev {}", pdevID);
+        return nullptr;
+    }
+
+    
+    it->second->first.lock();
+    if(it->second->second.empty()) {
+        LOGINFO("No more available chunks found for pdev {}", pdevID);
+        return nullptr;
+    }
     auto& vchunk = it->second->second.top();
     it->second->second.pop();
+    it->second->first.unlock();
+    
+    {
+        std::lock_guard<std::mutex> l(lock);
+        m_pdev_avalable_blk_map[pdevID] -= vchunk.available_blks();
+    }
+
     return vchunk.get_internal_chunk();
 }
 
 void HeapChunkSelector::foreach_chunks(std::function< void(csharedChunk&) >&& cb) {
-    VChunkHeap tempHeap;
-    for(auto& p : _pdev_heap_map) {
-        std::unique_lock<std::mutex> l(p.second->first);
-        auto& heap = p.second->second;
-        while(!heap.empty()) {
-            auto& vchunk = heap.top();
-            heap.pop();
-            cb(vchunk.get_internal_chunk());
-            tempHeap.emplace(std::move(vchunk));
-        }        
-        heap.swap(tempHeap);
-    }
+    //we should call `cb` on all the chunks, selected or not
+    std::lock_guard<std::mutex> l(lock); 
+    std::for_each(std::execution::par_unseq, m_chunks.begin(), m_chunks.end(), 
+    [cb = std::move(cb)](const VChunk& p){ cb(p.get_internal_chunk()); });
 }
-
 }   // namespace homeobject
\ No newline at end of file
diff --git a/src/lib/homestore/heap_chunk_selector.h b/src/lib/homestore/heap_chunk_selector.h
index e0f0d862..ebe55627 100644
--- a/src/lib/homestore/heap_chunk_selector.h
+++ b/src/lib/homestore/heap_chunk_selector.h
@@ -9,9 +9,18 @@
 #include <mutex>
 #include <utility>
 #include <memory>
+#include <map>
+#include <functional>
 
 #include <folly/concurrency/ConcurrentHashMap.h>
 
+using VChunk = homestore::VChunk;
+
+template <>
+struct std::hash< VChunk > {
+    std::size_t operator()(VChunk const& v) const noexcept { return std::hash< uint16_t >()(v.get_chunk_id()); }
+};
+
 namespace homeobject {
 
 using csharedChunk = homestore::cshared<homestore::Chunk>;  
@@ -19,14 +28,8 @@ using csharedChunk = homestore::cshared<homestore::Chunk>;
 class HeapChunkSelector : public homestore::ChunkSelector {
 public:
     HeapChunkSelector() = default;
-    HeapChunkSelector(const HeapChunkSelector&) = delete;
-    HeapChunkSelector(HeapChunkSelector&&) noexcept = delete;
-    HeapChunkSelector& operator=(const HeapChunkSelector&) = delete;
-    HeapChunkSelector& operator=(HeapChunkSelector&&) noexcept = delete;
     ~HeapChunkSelector() = default;
 
-    using VChunk = homestore::VChunk;
-
     class VChunkComparator {
     public:
         bool operator()(VChunk& lhs, VChunk& rhs) {
@@ -41,6 +44,14 @@ class HeapChunkSelector : public homestore::ChunkSelector {
     csharedChunk select_chunk([[maybe_unused]]homestore::blk_count_t nblks, const homestore::blk_alloc_hints& hints);
 
 private:
-    folly::ConcurrentHashMap< uint32_t, std::shared_ptr<std::pair< std::mutex, VChunkHeap > > >_pdev_heap_map;
+    folly::ConcurrentHashMap< uint32_t, std::shared_ptr< std::pair< std::mutex, VChunkHeap > > > m_pdev_heap_map;
+    //for now, uint32_t is enough for the sum of all the available blocks of a pdev.
+    //if necessary , we can change this to uint64_t to hold a larger sum.
+    std::unordered_map< uint32_t, uint32_t> m_pdev_avalable_blk_map;
+
+    //hold all the chunks , selected or not
+    std::unordered_set< VChunk > m_chunks;
+    std::mutex lock;
+    
 };
 } // namespace homeobject
\ No newline at end of file