如何对数据库原子进行多次更新和查询操作？

Question

I am building a pet project (delivery service - customers and couriers) using plain Servlets and JDBC. When the customer sends new order form via JSP, I need to analyze data, create new order entity, and store it in the database. I also generate a so called track number per each new or pending order (just a sequence of symbols and numbers), so the customer may search his order and check status. So the create order logic is

1. Fill in order entity
2. Download list of all track numbers in use from the DB
3. Keep generating track number until it doesn't match any one in the list
4. Assign track number to the order entity
5. Store the order entity in the database
6. Send confirmation email with track number to customer

The thing is that operations 2-5 must be atomic (a request thread was sent to sleep by the scheduler after it has downloaded list of track numbers in use and generated a track number, but before it has stored new order. In the worst case scenario another thread could have generated the same track number and store the order in the database which will lead to errors), and I can't figure out how to do it optimally.

My current solution is to place methods getActiveTracks(), generateTrackNumber() and storeOrder() in the synchronized block using ReentrantLock with timeout. But I wonder if it will have any negative impact on concurrent requests from multiple users. My second thought was to make it as a transaction at the DAO layer, but I don't know how to manage several operations simultaneously (query the list of track numbers from the orders table, check if a newly generated track number corresponds to any one from the list, and if not, update the orders table with new order)

Edit I should have indicated that I'm not using track numbers as primary keys and I do have unique ids to each order record (just a BIGSERIAL type). The track number is more of a convenience for 'customer' (its 4-symbol long, eg X5F4, which customer can easily memorize and enter into the search bar on the website). These track numbers may repeat many times since the system can distinguish each record by order id and status (COMPLETED, NEW, PENDING). So in my situation I generate track number until it doesn't match the one with NEW or PENDING (it doesn't matter if it matches COMPLETED).

So each of my order rows are like

id	track_number	status	date
3847	X45G	NEW	2021-07-19
4657	X45G	COMPLETED	2021-05-03

I need to perform following operations when creating new order

1. Query DB for the list of all new or pending orders SELECT track_number FROM orders WHERE status IN('NEW', 'PENDING')
2. Generate track number until it doesn't match any one in the list
3. Update DB INSERT INTO orders VALUES(DEFAULT, ?, ?, ?)

And I need these three operations to be atomic

Answer 1

Two main things:

1. Realize that even with synchronized blocks if your application ever moves into a clustered environment you will very likely create collisions in a busy environment when generating the track number.
2. Obtaining the track number this way will do nothing but slow down over time as the list grows.

Instead, I'd start with using a GUID for each "track number". A simple:

import java.util.UUID;


String uuid = UUID.randomUUID().toString();

yourOrderObject.setId(uuid);

No transactions needed.

Second, you should think of storing in the database and sending an email as separate tasks. I would not want to roll back the transaction if the email couldn't be sent for some reason.

Doing things this way means that you eliminate steps 2 and 3 and likely speed up the system considerably.