python 高速缓存和事务如何保持一致

Posted 2021-05-09

C1 C2 keysToEvict #C1 keysToEvict --- threadLocal; C2 --- medis


'''
关闭C1, isolation level: Read Committed
开启C1, isolation level: Repeatable Read, 无幻读，但少数场景依然是Read Committed

eg:
    A               B               C
    start           
                    start
    ...             X.update(...)
                    commit          start
                                    X.get()
                                    commit <---C2 cache is loaded

    X.get() <--可以读到B的update                 
    ...
    commit
    
    
如果cache切面应用在Service层，需要在Service层提供额外的细粒度的方法对DAO层update/delete包装。

'''

def get(k):
    if k in C1:
        return C1[k]
    if k in keysToEvict:
        return None
    if k in C2:
        C1[k] = C2[k]
        return C2[k]
    return None

def evict(k):
    C1.evict(k)
    keysToEvict.add(k)
    register a <TransactionSynchronization> to flush keysToEvict if neccessary

def put(k, v):
    C1[k] = v
    if tx is readonly:
        C2[k] = v