What Starbucks Can Teach Us About Asynchronous Messaging and Two‑Phase Commit

1. A Cup of Hot Chocolate (Hotto Cocoa o Kudasai)

After a two‑week trip to Japan, the author observed the massive number of Starbucks stores in Shinjuku and Roppongi and wondered how Starbucks handles orders.

Starbucks maximizes orders by using an asynchronous workflow: the cashier marks the cup, places it in a queue (the line of cups on the coffee machine), decoupling cashier and barista so the cashier can keep taking orders.

2. Correlation

Because the order of coffee completion may differ from order placement, Starbucks must correlate each finished cup with the correct customer. In the US they write the customer's name on the cup; elsewhere they may use the drink type as an identifier.

3. Exception Handling

Starbucks deals with failures similarly to asynchronous messaging systems:

If payment fails, they discard the coffee or remove the cup from the queue.

If the coffee is wrong or unsatisfactory, they remake it.

If the machine breaks, they refund.

3.1 Discard

The simplest strategy is to do nothing or discard the work, which is acceptable when the loss is small.

3.2 Retry

When an operation fails, you can either roll back completed steps or retry the failed ones. Retry is viable if the failure is transient (e.g., an external system is down).

3.3 Compensation

Compensation rolls back all completed actions to restore consistency, such as refunding a charge in a financial transaction.

4. Two‑Phase Commit

Applying two‑phase commit to Starbucks would require a prepare phase (order placed, receipt printed, cash held) and an execute phase (cash, receipt, and coffee exchanged simultaneously). This would drastically reduce throughput.

5. Conversation Pattern

The coffee ordering process exemplifies a simple conversation pattern: a short synchronous interaction for ordering and payment, followed by a longer asynchronous interaction for preparation and delivery. Similar patterns appear in e‑commerce, where order numbers are assigned synchronously and fulfillment happens asynchronously, with compensation and retry mechanisms.

Observing everyday asynchronous processes can inspire robust message‑oriented system designs.