400行代碼實現本地Key-Value緩存

Key-Value緩存有很多，用的較多的是memcache、redis，他們都是以獨立服務的形式運行，在工作中有時需要嵌入一個本地的key-value緩存，當然已經有LevelDb等，但感覺還是太重量級了。

本文實現了一種超級輕量的緩存，

1、實現代碼僅僅需要400行；

2、性能高效，value長度在1K時測試速度在每秒200萬左右

3、緩存是映射到文件中的，所以沒有malloc、free的開銷，以及帶來的內存洩露、內存碎片等；

4、如果服務掛掉了，重啟後緩存內容繼續存在；

5、如果把緩存映射到磁盤文件就算機器掛了，緩存中內容還是會存在，當然有可能會出現數據損壞的情況；

6、一定程度上實現了LRU淘汰算法，實現的LRU不是全局的只是一條鏈上的，所以只能說在一定程序上實現了；

7、穩定，已經在多個項目中運用，線上部署的機器有幾十台，運行了大半年了沒出過問題；

8、普通的緩存key、value都是字符串的形式，此緩存的key、value都可以是class、struct對象結構使用更方便；

 老規矩直接上代碼：

template<typename K, typename V>
class HashTable
{
public:
    HashTable(const char *tablename, uint32_t tableLen, uint32_t nodeTotal);
    virtual ~HashTable();

    bool Add(K &key, V &value)
    {
        AutoLock autoLock(m_MutexLock);

        //check is exist
        uint32_t nodeId = GetIdByKey(key);
        if(nodeId != m_InvalidId) return false;

        nodeId = GetFreeNode();
        if(nodeId == m_InvalidId) return false;

        uint32_t hashCode = key.HashCode();
        Entry *tmpNode = m_EntryAddr + nodeId;
        tmpNode->m_Key = key;
        tmpNode->m_Code = hashCode;
        tmpNode->m_Value = value;

        uint32_t index = hashCode % m_HeadAddr->m_TableLen;
        AddNodeToHead(index, nodeId);

        return true;
    }
    
    bool Del(K &key)
    {
        AutoLock autoLock(m_MutexLock);

        uint32_t nodeId = GetIdByKey(key);
        if(nodeId == m_InvalidId) return false;
        
        uint32_t index = key.HashCode() % m_HeadAddr->m_TableLen;
        
        return RecycleNode(index, nodeId);
    }

    bool Set(K &key, V &value)
    {
        AutoLock autoLock(m_MutexLock);
        
        uint32_t nodeId = GetIdByKey(key);
        if(nodeId == m_InvalidId) return false;

        (m_EntryAddr + nodeId)->m_Value = value;

        return true;
    }
        
    bool Get(K &key, V &value)
    {
        AutoLock autoLock(m_MutexLock);
        
        uint32_t nodeId = GetIdByKey(key);
        if(nodeId == m_InvalidId) return false;

        value = (m_EntryAddr + nodeId)->m_Value;

        return true;
    }
    
    bool Exist(K &key)
    {
        AutoLock autoLock(m_MutexLock);
        
        uint32_t nodeId = GetIdByKey(key);
        if(nodeId == m_InvalidId) return false;

        return true;
    }

    uint32_t Count()
    {
        AutoLock autoLock(m_MutexLock);
        return m_HeadAddr->m_UsedCount;
    }

    //if exist set else add
    bool Replace(K &key, V &value)
    {
        AutoLock autoLock(m_MutexLock);
        
        if(Exist(key)) return Set(key, value);
        else return Add(key, value);
    }

    /***********************************************
    ****LRU: when visit a node, move it to head ****
    ************************************************/
    //if no empty place,recycle tail
    bool LruAdd(K &key, V &value, K &recyKey, V &recyValue, bool &recycled)
    {
        AutoLock autoLock(m_MutexLock);
        
        if(Exist(key)) return false;

        if(Add(key, value)) return true;

        uint32_t index = key.HashCode() % m_HeadAddr->m_TableLen;
        uint32_t tailId = GetTailNodeId(index);
        
        if(tailId == m_InvalidId) return false;
        
        Entry *tmpNode = m_EntryAddr + tailId;
        recyKey   = tmpNode->m_Key;
        recyValue = tmpNode->m_Value;
        recycled  = true;

        RecycleNode(index, tailId);
        
        return Add(key, value);
    }
    
    bool LruSet(K &key, V &value)
    {
        AutoLock autoLock(m_MutexLock);

        if(Set(key, value)) return MoveToHead(key);
        else return false;
    }
    
    bool LruGet(K &key, V &value)
    {
        AutoLock autoLock(m_MutexLock);
    
        if(Get(key, value)) return MoveToHead(key);
        else return false;
    }

    //if exist set else add; if add failed recycle tail than add
    bool LruReplace(K &key, V &value, K &recyKey, V &recyValue, bool &recycled)
    {
        AutoLock autoLock(m_MutexLock);

        recycled = false;
        
        if(Exist(key)) return LruSet(key, value);
        else return LruAdd(key, value, recyKey, recyValue, recycled);
    }

    void Clear()
    {
        AutoLock autoLock(m_MutexLock);
        
        m_HeadAddr->m_FreeBase = 0;
        m_HeadAddr->m_RecycleHead = 0;
        m_HeadAddr->m_UsedCount = 0;
        for(uint32_t i = 0; i < m_HeadAddr->m_TableLen; ++i)
        {
            (m_ArrayAddr+i)->m_Head = m_InvalidId;
            (m_ArrayAddr+i)->m_Tail = m_InvalidId;
        }
    }

    int GetRowKeys(vector<K> &keys, uint32_t index)
    {
        AutoLock autoLock(m_MutexLock);
        
        if(index >= m_HeadAddr->m_TableLen) return -1;

        keys.clear();
        keys.reserve(16);
        
        int count = 0;
        Array *tmpArray = m_ArrayAddr + index;
        uint32_t nodeId = tmpArray->m_Head;
        while(nodeId != m_InvalidId)
        {
            Entry *tmpNode = m_EntryAddr + nodeId;
            keys.push_back(tmpNode->m_Key);
            nodeId = tmpNode->m_Next;
            ++count;
        }

        return count;
    }
    
    void *Padding(uint32_t size)
    {
        AutoLock autoLock(m_MutexLock);

        if(size > m_HeadSize - sizeof(TableHead)) return NULL;
        else return m_HeadAddr->m_Padding;
    }

private:
    static const uint32_t m_InvalidId = 0xffffffff;
    static const uint32_t m_HeadSize = 1024;
    struct TableHead
    {
        uint32_t m_TableLen;
        uint32_t m_NodeTotal;
        uint32_t m_FreeBase;
        uint32_t m_RecycleHead;
        uint32_t m_UsedCount;
        char     m_TableName[256];
        uint32_t m_Padding[0];
    };

    struct Array
    {
        uint32_t m_Head;
        uint32_t m_Tail;
    };

    struct Entry
    {
        V m_Value;
        K m_Key;
        uint32_t m_Code;
        uint32_t m_Next;
        uint32_t m_Prev;
    };
    
    size_t     m_MemSize;
    uint8_t   *m_MemAddr;
    TableHead *m_HeadAddr;
    Array     *m_ArrayAddr;
    Entry     *m_EntryAddr;

    ThreadMutex m_MutexLock;

    bool MoveToHead(K &key);
    uint32_t GetIdByKey(K &key);
    void AddNodeToHead(uint32_t index, uint32_t nodeId);
    bool MoveNodeToHead(uint32_t index, uint32_t nodeId);
    bool RecycleNode(uint32_t index, uint32_t nodeId);
    uint32_t GetTailNodeId(uint32_t index);
    uint32_t GetFreeNode();

    DISABLE_COPY_AND_ASSIGN(HashTable);
};

template<typename K, typename V>
HashTable<K, V>::HashTable(const char *tablename, uint32_t tableLen, uint32_t nodeTotal)
{
    AbortAssert(tablename != NULL);

    m_MemSize = m_HeadSize + tableLen*sizeof(Array) + nodeTotal*sizeof(Entry);
    m_MemAddr = (uint8_t*)MemFile::Realloc(tablename, m_MemSize);
    AbortAssert(m_MemAddr != NULL);

    m_HeadAddr = (TableHead*)(m_MemAddr);
    m_ArrayAddr = (Array*)(m_MemAddr + m_HeadSize);
    m_EntryAddr = (Entry*)(m_MemAddr + m_HeadSize + tableLen*sizeof(Array));

    m_HeadAddr->m_TableLen = tableLen;
    m_HeadAddr->m_NodeTotal = nodeTotal;
    strncpy(m_HeadAddr->m_TableName, tablename, sizeof(m_HeadAddr->m_TableName));
    
    if(m_HeadAddr->m_UsedCount == 0)//if first use init array to invalid id 
    {
        for(uint32_t i = 0; i < tableLen; ++i)
        {
            (m_ArrayAddr+i)->m_Head = m_InvalidId;
            (m_ArrayAddr+i)->m_Tail = m_InvalidId;
        }

        m_HeadAddr->m_FreeBase = 0;
        m_HeadAddr->m_RecycleHead = 0;
    }
}

template<typename K, typename V>
HashTable<K, V>::~HashTable()
{
    MemFile::Release(m_MemAddr, m_MemSize);
}

template<typename K, typename V>
bool HashTable<K, V>::MoveToHead(K &key)
{
    uint32_t nodeId = GetIdByKey(key);
    uint32_t index = key.HashCode() % m_HeadAddr->m_TableLen;

    return MoveNodeToHead(index, nodeId);
}

template<typename K, typename V>
uint32_t HashTable<K, V>::GetIdByKey(K &key)
{
    uint32_t hashCode = key.HashCode();
    uint32_t index = hashCode % m_HeadAddr->m_TableLen;
    Array *tmpArray = m_ArrayAddr + index;
    
    uint32_t nodeId = tmpArray->m_Head;
    while(nodeId != m_InvalidId)
    {
        Entry *tmpNode = m_EntryAddr + nodeId;
        if(tmpNode->m_Code == hashCode && key.Equals(tmpNode->m_Key)) break;
        
        nodeId = tmpNode->m_Next;
    }

    return nodeId;
}

template<typename K, typename V>
void HashTable<K, V>::AddNodeToHead(uint32_t index, uint32_t nodeId)
{
    if(index >= m_HeadAddr->m_TableLen || nodeId >= m_HeadAddr->m_NodeTotal) return;

    Array *tmpArray = m_ArrayAddr + index;
    Entry *tmpNode = m_EntryAddr + nodeId;
    if(m_InvalidId == tmpArray->m_Head)
    {
        tmpArray->m_Head = nodeId;
        tmpArray->m_Tail = nodeId;
    }
    else
    {
        tmpNode->m_Next = tmpArray->m_Head;
        (m_EntryAddr + tmpArray->m_Head)->m_Prev = nodeId;
        tmpArray->m_Head = nodeId;
    }
}

template<typename K, typename V>
bool HashTable<K, V>::MoveNodeToHead(uint32_t index, uint32_t nodeId)
{
    if(index >= m_HeadAddr->m_TableLen || nodeId >= m_HeadAddr->m_NodeTotal) return false;
    
    Array *tmpArray = m_ArrayAddr + index;
    Entry *tmpNode = m_EntryAddr + nodeId;
    
    //already head
    if(tmpArray->m_Head == nodeId)
    {
        return true;
    }

    uint32_t nodePrev = tmpNode->m_Prev;
    uint32_t nodeNext = tmpNode->m_Next;
    (m_EntryAddr+nodePrev)->m_Next = nodeNext;
    if(nodeNext != m_InvalidId)
    {
        (m_EntryAddr+nodeNext)->m_Prev = nodePrev;
    }
    else
    {
        tmpArray->m_Tail = nodePrev;
    }

    tmpNode->m_Prev = m_InvalidId;
    tmpNode->m_Next = tmpArray->m_Head;
    (m_EntryAddr + tmpArray->m_Head)->m_Prev = nodeId;
    tmpArray->m_Head = nodeId;

    return true;
}

template<typename K, typename V>
bool HashTable<K, V>::RecycleNode(uint32_t index, uint32_t nodeId)
{
    if(index >= m_HeadAddr->m_TableLen || nodeId >= m_HeadAddr->m_NodeTotal) return false;
    
    Array *tmpArray = m_ArrayAddr + index;
    Entry *tmpNode = m_EntryAddr + nodeId;

    uint32_t nodePrev = tmpNode->m_Prev;
    uint32_t nodeNext = tmpNode->m_Next;

    if(nodePrev != m_InvalidId)
    {
        (m_EntryAddr + nodePrev)->m_Next = nodeNext;
    }
    else
    {
        tmpArray->m_Head = nodeNext;
    }

    if(nodeNext != m_InvalidId)
    {
        (m_EntryAddr + nodeNext)->m_Prev = nodePrev;
    }
    else
    {
        tmpArray->m_Tail = nodePrev;
    }

    (m_EntryAddr+nodeId)->m_Next = m_HeadAddr->m_RecycleHead;
    m_HeadAddr->m_RecycleHead = nodeId;
    --(m_HeadAddr->m_UsedCount);

    return true;
}

template<typename K, typename V>
uint32_t HashTable<K, V>::GetTailNodeId(uint32_t index)
{
    if(index >= m_HeadAddr->m_TableLen) return m_InvalidId;
    
    Array *tmpArray = m_ArrayAddr + index;

    return tmpArray->m_Tail;
}

template<typename K, typename V>
uint32_t HashTable<K, V>::GetFreeNode()
{
    uint32_t nodeId = m_InvalidId;
    if(m_HeadAddr->m_UsedCount < m_HeadAddr->m_FreeBase)//get from recycle list
    {
        nodeId = m_HeadAddr->m_RecycleHead;
        m_HeadAddr->m_RecycleHead = (m_EntryAddr+nodeId)->m_Next;
        ++(m_HeadAddr->m_UsedCount);
    }
    else if(m_HeadAddr->m_UsedCount < m_HeadAddr->m_NodeTotal)//get from free mem
    {
        nodeId = m_HeadAddr->m_FreeBase;
        ++(m_HeadAddr->m_FreeBase);
        ++(m_HeadAddr->m_UsedCount);
    }
    else
    {
        nodeId = m_InvalidId;
    }

    //init node
    if(nodeId < m_HeadAddr->m_NodeTotal)
    {
        Entry *tmpNode = m_EntryAddr + nodeId;
        memset(tmpNode, 0, sizeof(Entry));

        tmpNode->m_Next = m_InvalidId;
        tmpNode->m_Prev = m_InvalidId;
    }
    
    return nodeId;
}