Cloud Datastore: ways to avoid race conditions

问题:

I have lots of views manipulating entities of same kind:

def view1(request, key):
    user = ndb.Key(urlsafe=key).get()
    user.x = 1
    user.put()
    ...

def view2(request, key):
    user = ndb.Key(urlsafe=key).get()
    user.y = 2
    user.put()
    ...

Obviously, this is error-prone due to possible race conditions (last wins):

1. view1 reads whole user entity data (x=None, y=None)
2. view2 reads whole user entity data (x=None, y=None)
3. view1 user.x = 1 (x=1, y=None)
4. view2 user.y = 2 (x=None, y=2)
5. view1 user.put() (x=1, y=None)
6. view2 user.put() (x=None, y=2)

What are best ways to fix this and what behaviour is considered most decent? Transactions (one of the requests is gonna fail, is this ok)?

回答1:

Wrap your get and put into a transaction. This will ensure you cannot stomp over a different update.

You can read more about transactions with the NDB Client Library documentation.

In your code, you could for example just use the NDB transaction decorator:

@ndb.transactional(retries=1)
def view1(request, key):
    user = ndb.Key(urlsafe=key).get()
    user.x = 1
    user.put()
    ...

@ndb.transactional(retries=1)    
def view2(request, key):
    user = ndb.Key(urlsafe=key).get()
    user.y = 2
    user.put()