The last 10 items

Scenario

A campaign product has 10 units left. At 09:00:00 a hundred dealers press "Order" within the same second. Finance later finds more invoices than units in the warehouse. Today you reproduce that second on purpose and make it impossible.

1. Predict first

Read how place() lowers stock (in your code after mission 05, it is the reserve step). It reads the product, checks stock >= quantity, then writes stock - quantity.

• 100 dealers, 1 unit each, stock 10. How many orders will be accepted? Write a number.
• What will the final stock be?

2. Reproduce

./lab race sets product 20000 to stock 10, fires 100 orders at once, and prints what happened:

./lab race

3. Watch two requests interleave (on paper)

Draw two columns, request A and request B. Both read stock = 1. Write each step in time order until both have written. This drawing is the whole lesson: if you can draw it, you understand it.

4. The 4 steps (on paper)

1. What is broken: one sentence with the words "read", "check", "write".
2. Numbers: orders accepted vs stock, final stock.
3. Path: the three statements Hibernate sends for one line of an order.
4. What can break: where exactly can another request slip in?

5. Fix it

Make the check and the decrement one atomic step. Options, pick one and be able to defend it:

• a single conditional UPDATE ... SET stock = stock - ? WHERE id = ? AND stock >= ? and check the number of updated rows (Spring Data: @Modifying @Query),
• optimistic locking with a @Version column (needs a Flyway migration) and a retry on conflict,
• SELECT ... FOR UPDATE (pessimistic lock) before the check.

If you return stock when a payment fails (mission 05), make that atomic too.

./lab race
./lab load orders 20

The second test proves you did not break normal orders, and shows what your choice costs on the hot campaign rows.

6. Evidence

	Before	After
orders accepted (stock 10)
final stock
orders 20/s p95
orders 20/s failed %

7. In your own words

One sentence: why does a correct-looking if not protect stock?

Check question

You choose optimistic locking. 100 dealers race for 10 units. How many requests will see a version conflict at least once, roughly, and what should your code do with them?

Java corner

• What does @Modifying change about how Spring Data runs a @Query?
• After a bulk UPDATE query, the Product entity already loaded in the same persistence context still shows the old stock. Why, and when does that matter?