CMU 15-445/645 PROJECT #1 - BUFFER POOL上（实现线程安全的LRU）

LRU的实现，实际上就是在算法题的基础上，加上线程完全。

用C++来写的话，可以使用哈希表和list，不用自己去实现双向链表

线程安全直接用C++11锁来做

 

缓冲buffer未完待续。

 

 

During the semester, you will be building a new disk-oriented storage manager for the BusTub DBMS. Such a storage manager assumes that the primary storage location of the database is on disk.

The first programming project is to implement a buffer pool in your storage manager. The buffer pool is responsible for moving physical pages back and forth from main memory to disk. It allows a DBMS to support databases that are larger than the amount of memory that is available to the system. The buffer pool's operations are transparent to other parts in the system. For example, the system asks the buffer pool for a page using its unique identifier (page_id_t) and it does not know whether that page is already in memory or whether the system has to go retrieve it from disk.

Your implementation will need to be thread-safe. Multiple threads will be accessing the internal data structures at the same and thus you need to make sure that their critical sections are protected with latches (these are called "locks" in operating systems).

You will need to implement the following two components in your storage manager:

• LRU Replacement Policy
• Buffer Pool Manager

This is a single-person project that will be completed individually (i.e. no groups).

• Release Date: Sep 14, 2020
• Due Date: Sep 27, 2020 @ 11:59pm

 

For each of the following components, we are providing you with stub classes that contain the API that you need to implement. You should not modify the signatures for the pre-defined functions in these classes. If you modify the signatures, the test code that we use for grading will break and you will get no credit for the project. You also should not add additional classes in the source code for these components. These components should be entirely self-contained.

 

PROJECT SPECIFICATION

For each of the following components, we are providing you with stub classes that contain the API that you need to implement. You should not modify the signatures for the pre-defined functions in these classes. If you modify the signatures, the test code that we use for grading will break and you will get no credit for the project. You also should not add additional classes in the source code for these components. These components should be entirely self-contained.

If a class already contains data members, you should not remove them. For example, the BufferPoolManager contains DiskManager and Replacer objects. These are required to implement the functionality that is needed by the rest of the system. On the other hand, you may need to add data members to these classes in order to correctly implement the required functionality. You can also add additional helper functions to these classes. The choice is yours.

You are allowed to use any built-in C++17 containers in your project unless specified otherwise. It is up to you to decide which ones you want to use. Note that these containers are not thread-safe and that you will need to include latches in your implementation to protect them. You may not bring in additional third-party dependencies (e.g. boost).

 

TASK #1 - LRU REPLACEMENT POLICY

This component is responsible for tracking page usage in the buffer pool. You will implement a new sub-class called LRUReplacer in src/include/buffer/lru_replacer.h and its corresponding implementation file in src/buffer/lru_replacer.cpp. LRUReplacer extends the abstract Replacer class (src/include/buffer/replacer.h), which contains the function specifications.

The size of the LRUReplacer is the same as buffer pool since it contains placeholders for all of the frames in the BufferPoolManager. However, not all the frames are considered as in the LRUReplacer. The LRUReplacer is initialized to have no frame in it. Then, only the newly unpinned ones will be considered in the LRUReplacer.

You will need to implement the LRU policy discussed in the class. You will need to implement the following methods:

• Victim(T*) : Remove the object that was accessed the least recently compared to all the elements being tracked by the Replacer, store its contents in the output parameter and return True. If the Replacer is empty return False.
• Pin(T) : This method should be called after a page is pinned to a frame in the BufferPoolManager. It should remove the frame containing the pinned page from the LRUReplacer.
• Unpin(T) : This method should be called when the pin_count of a page becomes 0. This method should add the frame containing the unpinned page to the LRUReplacer.
• Size() : This method returns the number of frames that are currently in the LRUReplacer.

The implementation details are up to you. You are allowed to use built-in STL containers. You can assume that you will not run out of memory, but you must make sure that the operations are thread-safe.

 

LRU的实现，建议使用C++ STL unordered_map和list，在外面加一把大锁即可。

注意这些方法的定义和Leetcode上题目描述不同，结合测试用例和英文原文理解。

namespace bustub {

/**
 * LRUReplacer implements the lru replacement policy, which approximates the Least Recently Used policy.
 */
class LRUReplacer : public Replacer {
 public:
  /**
   * Create a new LRUReplacer.
   * @param num_pages the maximum number of pages the LRUReplacer will be required to store
   */
  explicit LRUReplacer(size_t num_pages);

  /**
   * Destroys the LRUReplacer.
   */
  ~LRUReplacer() override;

  bool Victim(frame_id_t *frame_id) override;

  void Pin(frame_id_t frame_id) override;

  void Unpin(frame_id_t frame_id) override;

  size_t Size() override;

 private:
  using ListIterator = typename std::list<frame_id_t>::const_iterator;
  using CacheMap = std::unordered_map<frame_id_t,ListIterator>;

  size_t num_pages;
  std::list<frame_id_t> m_lruList;
  CacheMap m_lruMap;
  std::mutex m_lock;
};

}  // namespace bustub

 

namespace bustub {

LRUReplacer::LRUReplacer(size_t num_pages):num_pages(num_pages) {
  m_lruMap.reserve(this->num_pages);
}

LRUReplacer::~LRUReplacer() = default;

bool LRUReplacer::Victim(frame_id_t *frame_id) {
  std::lock_guard<std::mutex> guard(m_lock);
  if(Size()==0) { return false;}
  *frame_id = m_lruList.back();
  m_lruMap.erase(m_lruList.back());
  m_lruList.pop_back();
  return true;
}

void LRUReplacer::Pin(frame_id_t frame_id) {
  std::lock_guard<std::mutex> guard(m_lock);
  if(m_lruMap.find(frame_id)==m_lruMap.end()) { return;}
  m_lruList.erase(m_lruMap[frame_id]);
  m_lruMap.erase(frame_id);
}

void LRUReplacer::Unpin(frame_id_t frame_id) {
  std::lock_guard<std::mutex> guard(m_lock);
  // key already exists, just update the queue
  if(auto mapIter = m_lruMap.find(frame_id);mapIter!=m_lruMap.end()){
    //m_lruList.splice(m_lruList.begin(),m_lruList,mapIter->second);
  }else{
    m_lruList.push_front(frame_id);
    m_lruMap[frame_id] = m_lruList.begin();
  }
}

size_t LRUReplacer::Size() { return m_lruMap.size(); }

}  // namespace bustub

然后是测试代码

namespace bustub {

TEST(LRUReplacerTest, SampleTest) {
  LRUReplacer lru_replacer(7);

  // Scenario: unpin six elements, i.e. add them to the replacer.
  lru_replacer.Unpin(1);
  lru_replacer.Unpin(2);
  lru_replacer.Unpin(3);
  lru_replacer.Unpin(4);
  lru_replacer.Unpin(5);
  lru_replacer.Unpin(6);
  lru_replacer.Unpin(1);
  EXPECT_EQ(6, lru_replacer.Size());

  // Scenario: get three victims from the lru.
  int value;
  lru_replacer.Victim(&value);
  EXPECT_EQ(1, value);
  lru_replacer.Victim(&value);
  EXPECT_EQ(2, value);
  lru_replacer.Victim(&value);
  EXPECT_EQ(3, value);

  // Scenario: pin elements in the replacer.
  // Note that 3 has already been victimized, so pinning 3 should have no effect.
  lru_replacer.Pin(3);
  lru_replacer.Pin(4);
  EXPECT_EQ(2, lru_replacer.Size());

  // Scenario: unpin 4. We expect that the reference bit of 4 will be set to 1.
  lru_replacer.Unpin(4);

  // Scenario: continue looking for victims. We expect these victims.
  lru_replacer.Victim(&value);
  EXPECT_EQ(5, value);
  lru_replacer.Victim(&value);
  EXPECT_EQ(6, value);
  lru_replacer.Victim(&value);
  EXPECT_EQ(4, value);
}

}  // namespace bustub

TASK #2 - BUFFER POOL MANAGER

Next, you need to implement the buffer pool manager in your system (BufferPoolManager). The BufferPoolManager is responsible for fetching database pages from the DiskManager and storing them in memory. The BufferPoolManager can also write dirty pages out to disk when it is either explicitly instructed to do so or when it needs to evict a page to make space for a new page.

To make sure that your implementation works correctly with the rest of the system, we will provide you with some of the functions already filled in. You will also not need to implement the code that actually reads and writes data to disk (this is called the DiskManager in our implementation). We will provide that functionality for you.

All in-memory pages in the system are represented by Page objects. The BufferPoolManager does not need to understand the contents of these pages. But it is important for you as the system developer to understand that Page objects are just containers for memory in the buffer pool and thus are not specific to a unique page. That is, each Page object contains a block of memory that the DiskManager will use as a location to copy the contents of a physical page that it reads from disk. The BufferPoolManager will reuse the same Page object to store data as it moves back and forth to disk. This means that the same Page object may contain a different physical page throughout the life of the system. The Page object's identifer (page_id) keeps track of what physical page it contains; if a Page object does not contain a physical page, then its page_id must be set to INVALID_PAGE_ID.

Each Page object also maintains a counter for the number of threads that have "pinned" that page. Your BufferPoolManager is not allowed to free a Page that is pinned. Each Page object also keeps track of whether it is dirty or not. It is your job to record whether a page was modified before it is unpinned. Your BufferPoolManager must write the contents of a dirty Page back to disk before that object can be reused.

Your BufferPoolManager implementation will use the LRUReplacer class that you created in the previous steps of this assignment. It will use the LRUReplacer to keep track of when Page objects are accessed so that it can decide which one to evict when it must free a frame to make room for copying a new physical page from disk.

You will need to implement the following functions defined in the header file (src/include/buffer/buffer_pool_manager.h) in the source file (src/buffer/buffer_pool_manager.cpp):

• FetchPageImpl(page_id)
• NewPageImpl(page_id)
• UnpinPageImpl(page_id, is_dirty)
• FlushPageImpl(page_id)
• DeletePageImpl(page_id)
• FlushAllPagesImpl()

For FetchPageImpl,you should return NULL if no page is available in the free list and all other pages are currently pinned. FlushPageImpl should flush a page regardless of its pin status.

Refer to the function documentation for details on how to implement these functions. Don't touch the non-impl versions, we need those to grade your code.

 

未完待续