Merge remote-tracking branch 'origin/master'

db/vlog_gc.cpp

@@ -89,6 +89,13 @@ void VlogGC::gc_counter_decrement() {
 
 
 void VlogGC::do_gc(size_t old_vlog_num, size_t new_vlog_num) {
+  //  判断old_vlog_num是否正在进行gc，如果是，直接返回
+  if (vlog_in_gc(old_vlog_num)) {
+    return ;
+  }
+  // 否则将当前old_vlog_num设置为正在gc
+  add_vlog_in_gc(old_vlog_num);
+
   gc_counter_increment();
   size_t _gc_num_ = get_gc_num();
 
@@ -96,12 +103,15 @@ void VlogGC::do_gc(size_t old_vlog_num, size_t new_vlog_num) {
   add_executor_params(_gc_num_, ep);
   add_vlog_gc(_gc_num_, this);
 
+  // FIXME: 线程的信息必须被保存在函数栈之外，否则函数栈销毁之后，线程会报错exc_bad_access, 这里需要有一个gc_hanlder线程一直运行并处理各个gc请求
   std::thread gc_thread([_gc_num_]() mutable {
     auto _vlog_gc_ = get_vlog_gc(_gc_num_);
     assert(_vlog_gc_ != nullptr);
     _vlog_gc_->exec_gc(_gc_num_);
   });
 
+  gc_thread.detach();
+
 }
 
 
@@ -117,8 +127,8 @@ void VlogGC::exec_gc(size_t gc_num_) {
 
   // start gc process
   auto ep = get_executor_params(gc_num_);
-  gc_executor::exec_gc(ep.vg, ep.old_vlog_num, ep.new_vlog_num);
-//  test_func();
+//  gc_executor::exec_gc(ep.vg, ep.old_vlog_num, ep.new_vlog_num);
+  test_func();
 
   curr_thread_nums_latch_.lock();
   if (curr_thread_nums_ >= max_thread_nums_) {
@@ -128,8 +138,13 @@ void VlogGC::exec_gc(size_t gc_num_) {
   curr_thread_nums_latch_.unlock();
 
   gc_counter_decrement();
+  std::cout << "vlog_gc.cpp line 138" << std::endl;
   del_executor_params(gc_num_);
+  std::cout << "vlog_gc.cpp line 140" << std::endl;
   del_vlog_gc(gc_num_);
+  std::cout << "vlog_gc.cpp line 142" << std::endl;
+
+  del_vlog_in_gc(ep.old_vlog_num);
 }
 
 // vlog: || curr_size(size_t) | value_nums(size_t) || value1 | value2 | ... |

db/vlog_gc.h

@@ -44,6 +44,7 @@ friend class gc_executor;
     memcpy(&slot_num, &value[sizeof(uint16_t)], sizeof(size_t));
     return slot_num;
   }
+
   inline size_t get_gc_num() {
     gc_num_latch_.lock();
     size_t _gc_num_ = gc_num ++;
@@ -51,6 +52,26 @@ friend class gc_executor;
     return _gc_num_;
   }
 
+  inline bool vlog_in_gc(size_t vlog_num) {
+    ovn_map_latch_.lock();
+    bool flag = false;
+    if (ovn_map_.find(vlog_num) != ovn_map_.end()) {
+      flag = true;
+    }
+    ovn_map_latch_.unlock();
+    return flag;
+  }
+  inline void add_vlog_in_gc(size_t vlog_num) {
+    ovn_map_latch_.lock();
+    ovn_map_[vlog_num] = true;
+    ovn_map_latch_.unlock();
+  }
+  inline void del_vlog_in_gc(size_t vlog_num) {
+    ovn_map_latch_.lock();
+    ovn_map_.erase(vlog_num);
+    ovn_map_latch_.unlock();
+  }
+
   SlotPage *slot_page_;
   VlogSet *vlog_set;  // 仅声明为指针，具体定义放在 vlog_gc.cpp
 
@@ -62,6 +83,9 @@ friend class gc_executor;
 
   std::mutex gc_num_latch_;
   size_t gc_num;
+
+  std::mutex ovn_map_latch_;
+  std::unordered_map<size_t, bool> ovn_map_;
 };
 
 

db/vlog_set.cpp

@@ -369,7 +369,7 @@ void VlogSet::mark_del_value(const struct slot_content &sc) {
   vinfo->discard ++;
   vinfo->value_nums --;
   vinfo->curr_size -= value_size;
-  // FIXME: gc process
+  // FIXME: gc process, avoid repeated gc
   if (vlog_need_gc(sc.vlog_num) && !vinfo->processing_gc) {
     // create new vlog
     vinfo->processing_gc = true;