Scenario: a flash sale oversells inventory Intermediate

Three incidents follow one another during a flash sale. At each step you choose a response, then see right away what happens.

• If you pick wrong, read the explanation and Try again.
• Scoring counts only your first choice at each step, so think before you click.

Scenario

Flash sale: 10 items, 50 people click Buy

You are the backend dev who owns checkout. Each choice immediately shows its consequences on the metrics below.

Oversold orders40

5xx errors0 %

Checkout p99180 ms

Duplicate orders / double charges0

Step 1/3 · Overselling

The flash sale ends: stock had 10 items but the system created 50 orders. Not a single error in the logs. The code even runs inside a proper BEGIN ... COMMIT.

final stock

9

orders created

50

isolation

READ COMMITTED

BEGIN;
SELECT stock FROM product WHERE id = $1;      -- app sees stock = 10
-- app: if (stock >= 1) ...
UPDATE product SET stock = $2 WHERE id = $1;  -- $2 = stock - 1 computed in the app
INSERT INTO orders ...;
COMMIT;

What do you do?

ARe-read stock once more right before the UPDATE, just to be safeBAdd FOR UPDATE to the SELECT, running each statement with pool.queryawait pool.query('SELECT stock ... FOR UPDATE'); await pool.query('UPDATE ...')CSwitch the transaction to SERIALIZABLE, keep the code as isDOne statement: UPDATE ... SET stock = stock - 1 WHERE id = $1 AND stock >= 1 RETURNING stock, then check rowCount

After playing, read Writing transactions in code for a demo that measures all 5 approaches with 50 concurrent requests.