database operations successfully commit. The algorithm now supports partial failures and still provides transactional guarantees without resorting to 2PC.
There is a simpler technique for assuring idempotent updates if the only concern is ordering. Let’s change our sample schema just a bit to illustrate the challenge and the solution (see figure 8). Suppose you also want to track the last date of sale and purchase for the user. You can rely on a similar scheme of updating the date with a message, but there is one problem.
Suppose two purchases occur within a short time window, and our message system doesn’t ensure ordered operations. You now have a situation where, depending
id name amt_sold amt_bought last_sale last_purchase
transaction
xid
seller_id
buyer_id
amount
upon which order the messages are processed in, you will have an incorrect value for last_purchase. Fortunately, this kind of update can be handled with a minor modification to the SQL, as illustrated in figure 9.
By simply not allowing the last_purchase time to go backward in time, you have made the update operations order independent. You can also use this approach to protect any update from out-of-order updates. As an alternative to using time, you can also try a monotonically increasing transaction ID.
A short side note on ordered message delivery is relevant. Message systems offer the ability to ensure that messages are delivered in the order they are received. This can be expensive to support and is often unnecessary, and, in fact, at times gives a false sense of security.
The examples provided here illustrate how message ordering can be relaxed and still provide a consistent view of the database, eventually. The overhead required to relax the ordering is nominal and in most cases is significantly less than enforcing ordering in the message system.
Further, a Web application is semantically an event-driven system regardless of the style of interaction. The client requests arrive to the system in arbitrary order. Processing time required per request varies. Request scheduling throughout the components of the systems is nondeterministic, resulting in nondeterministic queuing of messages. Requiring the order to be preserved gives a false sense of security. The simple reality is that nondeterministic inputs will lead to nondeterministic outputs.
Up to this point, the focus has been on trading consistency for availability. The other side of the coin is understanding the influence that soft state and eventual consistency has on application design.
As software engineers we tend to look at our systems
For each message in queue:
Peek message
Begin transaction
Update user set last_purchase=message.trans_date where id=message.buyer_id and last_purchase< message.trans_date; End transaction
If transaction successful
Remove message from queue
End for
References:
Archives