Originally posted at 4/13/2011
As much as I rile against unbounded result set, I recently run into the first real case where “I need all of the data now” was a valid use case.
The issue is charging micro payments from customers in a particular market. In that market, accounts are usually handled using prepaid codes. So you open an account and load some amount of money into it. Once that money is over, there is no way to charge the account any longer. Users will occasionally re-charge their account. Now, we are talking about recurring billing scenario, where we need to charge users every week some amount of money.
So far, seems pretty reasonable, I hope. The catch is that it is pretty routine for billing to fail for insufficient funds. The strategy to resolve this is to analyze the user’s behavior and retry at likely times when we hope to catch the account with some money in it.
Since failures are common, and since most people behave in a similar fashion, what happens is that billing cycles trends to focus on the likely times when we hope to find some money there. In other words, let us say that we noticed that on Sunday evening, most people charge their account…
What this means is that we will have to try to charge those accounts in that time period. To make things interesting, there are other parties as well, which probably also managed to figure out the times when most accounts have money in them. At that point, it actually becomes a race, who will be able to charge the accounts first, and get the money.
Starting at a given time, charge as many accounts as fast as possible. Charging each account is a separate action, we don’t have the ability to do batching.
How would you solve this problem?
My solution, in the next post…