Row-Level Locking Strategies

When multiple users or processes try to change the same data at the same time, databases need a way to prevent conflicts and ensure data integrity. This is where row-level locking comes in.

A row-level lock allows a transaction to claim exclusive or shared access to specific rows, preventing other transactions from making conflicting changes until the lock is released. It's crucial for high-concurrency applications.

PostgreSQL automatically applies row-level locks during Data Manipulation Language (DML) operations like INSERT, UPDATE, and DELETE.

• INSERT: Places an exclusive lock on the newly inserted row.
• UPDATE: Places an exclusive lock on the row being modified.
• DELETE: Places an exclusive lock on the row being deleted.

These implicit locks ensure that only one transaction can modify a specific row at a time.

Sometimes you need to lock rows before modifying them, especially when your application logic involves reading data, making decisions, and then updating. This is where SELECT ... FOR UPDATE is invaluable.

It acquires an exclusive lock on the selected rows, preventing other transactions from updating or deleting them until your transaction commits or rolls back.

Try this example. If you run SELECT ... FOR UPDATE in one database session, then try to UPDATE the same row from another session, the second session will wait.

Session 1:

The previous code snippet shows how FOR UPDATE works. If you ran the SELECT in Session 1, then immediately tried to run this UPDATE in a different Session 2, Session 2 would wait until Session 1 either COMMITs or ROLLBACKs.

Session 2 (will wait):

What if you want to prevent updates, but allow other transactions to read the data or even acquire their own shared lock?

SELECT ... FOR SHARE acquires a shared lock. This means:

• Other transactions can read the rows.
• Other transactions can acquire their own FOR SHARE locks.
• Other transactions cannot acquire FOR UPDATE locks or modify the rows.

If Session 1 holds a FOR SHARE lock, Session 2 can also acquire a FOR SHARE lock, but a FOR UPDATE or DML operation on the same row will wait.

Session 1:

SELECT ... FOR NO KEY UPDATE is similar to FOR UPDATE but is weaker. It acquires an exclusive lock that doesn't block FOR KEY SHARE locks.

It's useful when you're updating non-key columns and don't need to prevent concurrent foreign key operations, which are typically very short-lived.

SELECT ... FOR KEY SHARE is the weakest explicit row-level lock. It allows other transactions to acquire FOR SHARE, FOR NO KEY UPDATE, and even other FOR KEY SHARE locks.

It primarily prevents other transactions from deleting the locked rows or acquiring an exclusive lock that would modify key columns. It's often used by foreign key constraints.

A critical strategy to prevent deadlocks (where two transactions wait for each other indefinitely) is to always acquire locks on multiple rows in a consistent order.

For example, if you need to lock rows with product_id = 5 and product_id = 10, always lock 5 first, then 10 across all transactions. This prevents a scenario where one transaction locks 5 then tries for 10, while another locks 10 then tries for 5.

Consider two concurrent transactions. Transaction A runs SELECT * FROM users WHERE user_id = 1 FOR UPDATE;. What happens if Transaction B immediately tries to run UPDATE users SET email = 'new@example.com' WHERE user_id = 1;?

We've explored how PostgreSQL manages concurrency with row-level locks:

• Implicit locks protect DML operations.
• FOR UPDATE provides exclusive row locks for modifications.
• FOR SHARE provides shared locks, allowing reads but blocking updates.
• FOR NO KEY UPDATE and FOR KEY SHARE offer more granular control.
• Consistently ordering lock acquisition is a key strategy to prevent deadlocks.

Mastering these strategies ensures your application handles concurrent writes efficiently and reliably!